A multi-source heterogeneous industrial equipment data management method

Abstract

The application discloses a multi-source heterogeneous industrial equipment data management method, and particularly relates to the technical field of industrial data management, and is used for solving the problem that the existing knowledge graph generates false association due to data quality problems and continuously strengthens in iterative updating, resulting in system cognitive distortion; is achieved by collecting and synchronizing data units from multiple sources, constructing a physical topology vector space and a data association vector space and calculating the space divergence degree to identify potential false association, then extracting key operation events in the operation log, analyzing the cause-effect direction and intensity change mode of potential false association and events in the event isolation window, and finally updating the knowledge graph according to the analysis result, and marking the association with event-driven continuous characteristics as an event-dependent association relationship.

Description

Technical Field

[0001] This invention relates to the field of industrial data management technology, and in particular to a method for managing data from multi-source heterogeneous industrial equipment. Background Technology

[0002] In the industrial manufacturing sector, to achieve equipment monitoring, production optimization, and predictive maintenance, it is necessary to integrate data from multiple heterogeneous devices such as programmable logic controllers, sensors, manufacturing execution systems, and maintenance records. Existing technologies typically employ methods such as data integration platforms or knowledge graphs to extract, transform, and load this data, and attempt to establish a unified semantic model to support upper-layer applications, involving the association, alignment, and continuous updating of multimodal industrial data.

[0003] However, existing technologies face a contradiction when achieving cross-source data semantic fusion: in order to build and maintain a seemingly complete knowledge model, the system will make inferences and fill in gaps based on local patterns, given the normal occurrence of quality fluctuations and missing information in the underlying data. This results in the generation of associations containing false components lacking factual basis. More seriously, such false associations will be repeatedly referenced and reinforced in subsequent incremental data updates, causing the entire knowledge model to gradually deviate from the actual equipment status and operating logic, resulting in systemic cognitive distortion. This exposes applications such as production decision-making and fault diagnosis based on such models to underlying credibility risks. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a method for managing data from multi-source heterogeneous industrial equipment.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: A method for managing data from multi-source heterogeneous industrial equipment includes: S1. Collect data units from multi-source heterogeneous industrial equipment, and associate each data unit with the equipment identifier and collection timestamp of its source equipment; S2. Obtain the physical topology network of industrial equipment and map the device nodes in the physical topology network to the topology vector space; S3. Calculate the data association strength between device nodes based on data units, and map the data association strength to the association vector space; S4. Calculate the spatial divergence of each pair of device nodes in the topology vector space and the association vector space, and identify potential false associations in the knowledge graph based on the spatial divergence. S5. Extract key operation events from the operation log, analyze the causal relationship between potential false associations and key operation events within the event isolation window corresponding to the key operation events, and compare the intensity change pattern of potential false associations in the pre-event stage and the post-event stage to generate event dependency analysis results. S6. Update the relationships in the knowledge graph based on the results of event dependency analysis. For potential false relationships with event-driven continuous association characteristics, mark them as event-dependent relationships in the knowledge graph.

[0006] Furthermore, S1 includes: Based on the communication protocols and data interfaces of industrial equipment, multimodal time-series data from programmable logic controllers, sensors, and manufacturing execution systems are analyzed from different sources to generate data units; Map the logical identifiers of the same industrial physical equipment across different data sources to a unified equipment identifier; The timestamps of data units collected from different data sources are synchronized and aligned based on a unified network time protocol.

[0007] Furthermore, S2 includes: Based on the factory piping and instrumentation diagram or equipment connection relationship database, construct a physical topology graph with industrial equipment as equipment nodes and physical connection relationships between equipment as edges; Based on the device identifier, associate the device nodes and data units in the physical topology diagram; The graph embedding algorithm maps the connection structure of device nodes in the physical topology graph to a low-dimensional dense topology vector space, so that device nodes with similar connections have similar vector representations in the topology vector space.

[0008] Furthermore, S3 includes: Based on the collected data units with collection timestamps, the temporal data association strength between each pair of device nodes is quantified using a time series similarity measurement method. Based on the calculated data association strength between all pairs of device nodes, a data association strength matrix is ​​constructed. A dimensionality reduction algorithm is used to map the association strength relationship between each pair of device nodes in the data association strength matrix to a low-dimensional dense association vector space, so that the vector representations of device node pairs with similar association strengths in the association vector space are also similar.

[0009] Furthermore, S4 includes: For each pair of device nodes, obtain their vector representation in the topology vector space and their vector representation in the associated vector space. The spatial distance metric method is used to calculate the distance between the vector representations of corresponding device nodes in the topological vector space, and the distance between the vector representations in the associated vector space. The ratio of the distance in the associated vector space to the distance in the topological vector space is defined as the spatial divergence of the corresponding device node. Device node pairs whose spatial deviation exceeds a preset deviation threshold are identified as potential false associations in the knowledge graph.

[0010] Furthermore, distances are calculated using spatial distance metrics, including Euclidean distance.

[0011] Furthermore, device node pairs whose spatial deviation exceeds a preset deviation threshold are identified as potentially false associations. This includes dynamically setting different preset deviation thresholds for different types of device node pairs based on the degree of determinism of the physical connections of industrial equipment.

[0012] Furthermore, S5 includes: From the operation logs of industrial equipment, extract key operational events that indicate equipment status switching, process formula changes, or control command triggering, along with their precise occurrence times; Centered on the precise occurrence time of critical operational events, time ranges are defined both before and after the event to form an event isolation window that includes the pre-event phase and the post-event phase. For each potential false association identified, based on the collection timestamp, the corresponding data units in the pre-event stage and the post-event stage are extracted respectively, and the statistical characteristics of the association strength of the corresponding association in the pre-event and post-event stages are calculated. By comparing the statistical characteristics of association strength before and after an event using time series analysis, and combining the abrupt change points of association strength with the chronological order of events, the causal relationship between potential spurious associations and key operational events is determined, generating event dependency analysis results that include causal relationships and intensity change patterns.

[0013] Furthermore, the analysis of key operational events includes classifying events into planned operational events or sudden abnormal events based on event description text or event codes; and forming event isolation windows, which includes setting different pre-event and post-event phase durations for planned operational events and sudden abnormal events based on the event classification results.

[0014] Furthermore, S6 includes: Based on the results of event dependency analysis, potential false associations that are determined to be caused by key operational events and whose association strength statistical characteristics are consistently higher in the later event stage than in the earlier event stage are identified as associations with event-driven persistent association characteristics. Based on the device node pairs corresponding to the determined associations, locate the corresponding association edges in the knowledge graph; Add event-dependent relationship tags to the edges of the relationships, and store the event identifiers and event isolation window information of the key operation events that trigger the relationships as the association attributes of the event-dependent relationship tags to complete the update of the knowledge graph.

[0015] The beneficial effects of this invention are: 1. By constructing a dual vector space of physical topology and data association and calculating its spatial deviation, potential false associations in the knowledge graph that contradict the physical connection basis can be directly identified. This effectively removes unreliable association components introduced by data quality fluctuations or local pattern inference, fundamentally curbing the self-reinforcing trend of false associations in the incremental update process of the knowledge graph. This enables the knowledge model to maintain an accurate representation of the actual state and operating logic of the equipment, laying a reliable data foundation for production decision-making and fault diagnosis applications based on this model.

[0016] 2. By analyzing the temporal causal relationships between key operational events and potential false associations, it is possible to finely distinguish between truly unfounded false associations and temporary associations triggered by specific events and possessing causal logic. This makes updating the knowledge graph no longer a simple addition or subtraction of associations, but rather an ability to assign dynamic semantic labels such as event dependencies to associations. This forms a knowledge model that better reflects the actual operating mechanism of industrial processes and possesses causal explanatory capabilities, ultimately improving the accuracy and interpretability of semantic fusion of multi-source heterogeneous data in industrial data management. Attached Figure Description

[0017] Figure 1 This is a flowchart of a multi-source heterogeneous industrial equipment data management method according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0019] Example: Figure 1 This invention provides a method for managing data from multi-source heterogeneous industrial equipment, comprising: S1. Collect data units from multi-source heterogeneous industrial equipment, and associate each data unit with the equipment identifier and collection timestamp of its source equipment; S2. Obtain the physical topology network of industrial equipment and map the device nodes in the physical topology network to the topology vector space; S3. Calculate the data association strength between device nodes based on data units, and map the data association strength to the association vector space; S4. Calculate the spatial divergence of each pair of device nodes in the topology vector space and the association vector space, and identify potential false associations in the knowledge graph based on the spatial divergence. S5. Extract key operation events from the operation log, analyze the causal relationship between potential false associations and key operation events within the event isolation window corresponding to the key operation events, and compare the intensity change pattern of potential false associations in the pre-event stage and the post-event stage to generate event dependency analysis results. S6. Update the relationships in the knowledge graph based on the results of event dependency analysis. For potential false relationships with event-driven continuous association characteristics, mark them as event-dependent relationships in the knowledge graph.

[0020] S1. Collect data units from multi-source heterogeneous industrial equipment, and associate each data unit with the equipment identifier and collection timestamp of its source equipment. Specifically, this is implemented as follows: First, data connections need to be established and parsed according to the communication protocols and data interface characteristics of various devices in the industrial field. For programmable logic controllers (PLCs), connections are typically made via industrial Ethernet protocols, such as PROFINET or EtherNet / IP. Real-time input / output status, internal variable values, and program running status information recorded in their cyclic output data areas or data blocks are read and processed according to their original scan cycles or triggered by changing events to form raw time-series data streams. For sensors, data acquisition is based on their interface type. For example, for analog input modules supporting the Modbus RTU protocol, measured values ​​are read from their registers via serial communication; for smart sensors equipped with IO-Link interfaces, periodic process data and event-driven service data containing measured values, parameters, and status information are obtained through the master station. For manufacturing execution systems (MES), time-series records related to business operations, such as production orders, process parameters, material batches, and equipment utilization rates, are queried through their standard application programming interfaces (APIs) or database access interfaces. The raw data obtained from different sources differ in format and frequency. After parsing, it needs to be uniformly converted into structured data units containing numerical values, time points, and data source identities. For example, a data unit from a temperature sensor might contain fields such as "Device Identifier: TIC-101", "Data Value: 25.3", "Unit: Degrees Celsius", and "Acquisition Timestamp: 2025-05-10 14:02:00.123". This process ensures that device data from different manufacturers and using different communication standards is transformed into a set of data units with clear semantics that can be consistently processed in subsequent steps. The generation frequency of data units can be set according to the data source update cycle or application requirements; for example, high-speed sensors can be acquired at the millisecond level, while production events can be recorded at the second or minute level.

[0021] After generating data units, it is necessary to address the issue of the same physical equipment having different logical identifiers in different data sources to achieve unified data association. In practice, a device identifier mapping table needs to be pre-established or obtained. This table records the correspondence between the registered name or address code of each physical device in the factory in different systems and the unified device identifier used for that device in subsequent analyses. For example, the same centrifugal pump might be named PUMP001 in the programmable logic controller's variable table, have an asset number of P-2025-1001 in the manufacturing execution system's asset database, and be labeled P-101 on the piping and instrumentation diagram. The content of the device identifier mapping table can be obtained from manually maintained electronic documents of the factory equipment ledger, or it can be automatically generated by accessing the equipment master data records in the enterprise asset management system database. During implementation, by querying this mapping table, the original logical identifier carried in the data unit is replaced with the unified device identifier agreed upon in the mapping table. For example, the tag number P-101 on the piping and instrumentation diagram is uniformly used as the unique identifier for that device in all subsequent analyses. If no corresponding relationship is found in the mapping table, the data unit can be marked as pending processing or ignored. Through this step, regardless of the data source, as long as it describes the same physical device, the generated data units will be associated with the same unified device identifier, laying the foundation for subsequent device node-based analysis.

[0022] After unifying equipment identification, the synchronization of data units from different data sources needs to be addressed. Since programmable logic controllers, sensors, and manufacturing execution systems may operate on their own independent clocks, their recorded timestamps may be inaccurate, leading to errors in direct time-series correlation analysis. In implementation, a network time protocol (NTP) is used as the unified time reference source. Specifically, a NTP client is configured on the data acquisition server deployed on the factory network to synchronize with an internal or higher-precision time server. The synchronization period between the NTP client and server can be set, for example, every minute. When collecting raw data from various data sources and generating data units, the acquisition program not only records the time information provided by the data source itself, but more importantly, it records the server system time synchronized with the NTP at the actual time the acquisition action occurs, and uses this time as the official acquisition timestamp for that data unit. For data sources that can provide high-precision time stamps, such as devices supporting the IEEE 1588 precision time protocol, their built-in high-precision timestamps are preferred, but their time reference must be compared and calibrated with the NTP server time. For business data with coarser time granularity, such as production events recorded in a manufacturing execution system on a minute-by-minute basis, a more precise acquisition timestamp, down to the second or millisecond level, is assigned based on the recorded event time and the moment the data record arrived at the acquisition interface. Through this series of operations, all data units, regardless of their source, have their acquisition timestamps unified to the same high-precision, consistent time benchmark, thereby ensuring the accuracy and reliability of subsequent operations such as time-series-based correlation analysis and event window segmentation. The output of the entire step S1 is a series of standardized data units with unified device identifiers and synchronized acquisition timestamps, forming the data foundation for all subsequent analyses.

[0023] S2. Obtain the physical topology network of the industrial equipment, and map the device nodes in the physical topology network to the topology vector space. Specifically, this is implemented as follows: First, physical connection information needs to be obtained from the factory's engineering design documents or asset database. Common sources include electronic factory piping and instrumentation diagrams, which use standard graphic symbols to label all equipment, instruments, valves, and the pipes and signal lines connecting them. During implementation, by parsing the vector graphic file of this diagram or its associated component attribute database, each graphic element representing industrial equipment and its unique identifier, such as the equipment tag number, is automatically identified. Simultaneously, the lines connecting these equipment graphic elements are automatically identified, and each such line is defined as an edge representing a physical connection relationship. Each edge can be assigned attributes, such as indicating the direction of material flow or signal transmission. Another implementation is based on an equipment connection relationship database, which records the physical connection relationships between equipment in a structured table format. For example, each record includes a starting equipment identifier field, an ending equipment identifier field, and a connection type field. By reading and parsing the table records of such a database, the set of equipment nodes and the set of connection edges can also be extracted. Based on the information extracted from any of the above sources, a formalized graph structure is constructed. In this graph structure, each industrial equipment instance is represented as a device node, and the node attributes at least include the unified device identifier obtained from step S1. Each physical connection instance is represented as an edge connecting two device nodes. For example, through parsing, a local physical topology graph can be constructed, containing device node P-101, device node V-201, and an edge connecting these two nodes. This physical topology graph fully depicts the static physical connection layout of the equipment within the factory and serves as the benchmark for subsequent spatial deviation calculations.

[0024] After constructing the physical topology diagram, the device nodes within it need to be precisely associated with the standardized data units generated and continuously accumulated in step S1. The sole basis for this association process is the unified device identifier. In practice, each device node in the physical topology diagram is traversed, and the device identifier recorded in its node attributes is read. Subsequently, in the data unit set generated in step S1, a query operation is performed to find all data units whose device identifier field values ​​are completely consistent with the current node's device identifier. These data units are then associated with the device node, typically by storing the data unit's index or reference in the device node's attributes. For example, a device node in the physical topology diagram with the device identifier TIC-101 will be associated with all temperature measurement data units with the device identifier TIC-101. This operation allows each device node in the static topology diagram to point to its corresponding dynamic runtime sequence data. To ensure the integrity of the association, a verification step can be designed during implementation, such as checking whether a device node in the physical topology diagram exists whose device identifier cannot be matched with any data unit in the data unit set; in such cases, a warning log can be generated. This step accurately integrates the physical topology with real-time historical operational data, providing direct data support for calculating the correlation strength between device nodes from a data perspective.

[0025] After associating device nodes with data units, the structured information of the physical topology graph needs to be transformed into a low-dimensional dense vector representation that facilitates mathematical calculations and comparisons, i.e., mapped to the topological vector space. This process is accomplished by implementing a graph embedding algorithm. Specifically, the constructed physical topology graph is used as the input graph for the algorithm. The core objective of the graph embedding algorithm is to learn a mapping function that generates a low-dimensional dense real-valued vector for each device node in the graph as its topological vector representation, ensuring that device nodes with similar connections in the original graph also have similar vector representations in the topological vector space in terms of distance metric. One optional implementation uses a graph embedding method based on random walks. This method first performs numerous random walks on the input physical topology graph to generate a set of device node sequences. The random walk strategy can be configured; for example, in directional networks, random walks can be set to prioritize walks along edges. The length of the random walk, i.e., the number of device nodes in each sequence, is a configurable parameter, and its setting can be based on a combination of graph diameter and computational resources, such as 20, 50, or 100. Next, the generated set of device node sequences is treated as training corpus, with each device node considered a word. Word vector learning techniques are used to train the node vector model, such as the Skip-gram model architecture. The training process involves adjusting the node vectors to maximize the probability of predicting neighboring device nodes appearing in the context window for each central device node in the sequence. Key parameters involved in this training process include: the dimension of the topology vector, i.e., the length of the output vector corresponding to each device node. This parameter can be set based on the total number of device nodes in the physical topology graph. For example, it can be set as the logarithm of the total number of nodes multiplied by a coefficient between 4 and 16, typically set to values ​​like 64, 128, or 256; the learning rate, used to control the step size of parameter updates during model training, with an initial value set to, for example, 0.025, using a linear decay strategy; and the number of iterations, referring to the number of rounds used to fully train the model using all node sequence data, which can be set to 5 or 10. After model training is complete, each device node obtains a topology vector of fixed dimensions. For example, device nodes P-101 and V-201, which are directly connected by pipes in the physical topology graph, will have a high cosine similarity in their topological vectors after model learning because they frequently appear as neighbors in the sequence generated by random walks. Conversely, two device nodes that are far apart in the physical topology graph and have long connection paths will have lower similarity in their topological vectors. Through this process, the complex physical connection network is encoded into a low-dimensional, dense topological vector space containing structural similarity information. The position vector of each device node in this space quantifies its structural role in the factory's physical topology, laying a solid foundation for the subsequent quantitative comparison with the associated vector space in step S4.

[0026] S3. Calculate the data association strength between device nodes based on data units, and map the data association strength to the association vector space. The specific implementation is as follows: First, calculations are performed based on the standardized data units with synchronized acquisition timestamps generated in step S1. For any pair of device nodes in the physical topology diagram, all associated data units need to be obtained from the association relationships established in step S2. These data units constitute two time series, each containing a numerical data value and its precise acquisition timestamp. Before calculation, data alignment preprocessing is usually required. This involves sorting the data units corresponding to the two device nodes in ascending order according to their acquisition timestamps for a selected analysis time range, such as the most recent 24 hours, and resampling or linear interpolation on a unified and continuous time axis to ensure that the two time series have corresponding data values ​​at the same number of uniformly distributed time points. After preprocessing, a time series similarity metric is used to calculate the data association strength between the pair of device nodes. A typical implementation is to calculate the Pearson correlation coefficient. This calculation requires two preprocessed time series numerical arrays of equal length as input and outputs a real number between -1 and +1. The specific calculation process is as follows: First, calculate the arithmetic mean of all data values ​​in both time series arrays. Then, subtract the arithmetic mean of the data value at each time point to obtain the deviation value at that point. Next, calculate the sum of the deviation values ​​of the two series at all time points. Finally, divide this sum by the square root of the sum of the squares of the deviation values ​​of the two series. This calculation yields a correlation coefficient. The closer the absolute value of this coefficient is to 1, the stronger the linear correlation between the data of the two device nodes within the selected time range, i.e., the higher the data association strength. Another optional implementation is to use a dynamic time warping algorithm to calculate the minimum warped path distance between the two time series. This distance reflects the similarity of the sequence shapes. Then, this distance value is mapped to a similarity score between 0 and 1 through a transformation function, such as a negative exponential function, as the data association strength. The characteristics of actual industrial data must be considered during the calculation. For example, data gaps that still exist after preprocessing can be imputed using the average of data from before and after the preprocessing. Potential transient impulse noise can be smoothed using a sliding median filter. This step needs to be repeated for all pairs of device nodes that need to be evaluated in the physical topology diagram, ultimately obtaining a quantified data association strength value between each pair of device nodes. The larger this value, the stronger the synchronization or interdependence of the two devices in terms of operational data.

[0027] After calculating the data association strength between all device node pairs, these scattered strength values ​​need to be organized into a global, structured view, i.e., a data association strength matrix needs to be constructed. In practice, firstly, all N device nodes in the physical topology diagram that need to participate in the calculation are determined, and each is assigned a fixed index number from 1 to N. Then, an N x N two-dimensional array is initialized as the data association strength matrix, where the row and column indices correspond to the indices of these N device nodes. The element in the i-th row and j-th column of the matrix stores the data association strength value between the device node with index i and the device node with index j. Since data association strength is usually defined as symmetric, i.e., the association strength between device node i and device node j is equal to the association strength between device node j and device node i, in practice, it is only necessary to calculate the device node pair strength when i is less than or equal to j, and fill the results into the i-th row and j-th column and j-th row and i-th column positions of the matrix, thus constructing a symmetric matrix. The elements on the diagonal of the matrix, representing the correlation strength between each device node and itself, are typically set to a generally accepted maximum correlation value, such as 1.0. The construction process is completed through a programmed loop: it iterates through all device node pairs whose index combination (i, j) satisfies the condition i <= j, retrieves the corresponding data correlation strength value from the cache or database of the calculation results from the first step, and then assigns this value to both matrix elements (i, j) and (j, i). For example, if the index of device node P-101 is 5 and the index of device node V-201 is 8, and their Pearson correlation coefficient is 0.85, then 0.85 is filled into the 5th row, 8th column and the 8th row, 5th column of the matrix. The resulting data correlation strength matrix is ​​an N x N real symmetric matrix that comprehensively depicts the pairwise correlation between all device nodes at the runtime data level within the current analysis time window in a compact two-dimensional table format, providing direct input for the next step of overall dimensionality reduction mapping.

[0028] After obtaining the data association strength matrix, it is necessary to map the high-dimensional associations it contains to a lower-dimensional, denser vector space, namely the association vector space. This step is achieved by implementing a dimensionality reduction algorithm. Specifically, the data association strength matrix is ​​used as the input to the dimensionality reduction algorithm. The core objective of the dimensionality reduction algorithm is to learn a mapping from device nodes to low-dimensional real vectors, so that device node pairs with similar data association strengths in the original high-dimensional relational space have closer spatial distances between their corresponding vector representations in the reduced low-dimensional association vector space. A typical implementation method is to use the classic multidimensional scaling algorithm. This algorithm takes the data association strength matrix as input and first needs to convert the association strength values ​​in the matrix into a distance metric. This conversion can be done by calculating the distance for each association strength value in the matrix as 1 minus the absolute value of the association strength value. This generates a distance matrix. Next, the algorithm solves for this distance matrix using eigenvalue decomposition. Specifically, it calculates the bicentered matrix of the distance matrix, then performs eigenvalue decomposition on the bicentered matrix, selecting the k largest positive eigenvalues ​​and their corresponding eigenvectors. Ultimately, the k-dimensional coordinates of each device node in the association vector space are formed by multiplying its corresponding eigenvector component by the square root of that eigenvalue. Another optional implementation is to use a spectral embedding algorithm. This algorithm treats the data association strength matrix as the affinity matrix of a graph. It calculates the normalized Laplacian matrix of this matrix and performs eigenvalue decomposition on the Laplacian matrix, selecting the eigenvectors corresponding to the k smallest eigenvalues ​​(excluding the smallest eigenvalue). These eigenvectors are then arranged column-wise, and their row vectors represent the k-dimensional association vector of each device node. Regardless of the specific algorithm used, one key parameter is the target dimension k, i.e., the dimension of the association vector space. The setting of this parameter can be based on a comprehensive consideration of the original number of device nodes and subsequent computational needs. For example, it can be determined by analyzing the inflection point of the eigenvalue spectrum, i.e., selecting the dimension corresponding to the point before a significant decrease in the eigenvalue magnitude; or it can be set to an empirical value, such as 8, 16, or 32, but k must be less than the original number of device nodes. After the algorithm is executed, each device node obtains a k-dimensional real vector, i.e., its representation in the association vector space. For example, after dimensionality reduction using a multidimensional scaling algorithm, the Euclidean distance between two vector points of device nodes P-101 and V-201 with high data association strength will be very small in the association vector space. Conversely, the distance between the vector points of two device nodes with weak data association strength will be large. Through this step, the complex association network, expressed in matrix form and mined from the original data, is encoded into a compact low-dimensional association vector space containing global association patterns. The position of each device node in this space comprehensively reflects its runtime data association pattern with all other device nodes in the network. This provides a crucial benchmark for comparison with the topological vector space to discover discrepancies between physical constraints and data performance.

[0029] S4. Calculate the spatial divergence degree of each pair of device nodes in the topology vector space and the association vector space, and identify potential false associations in the knowledge graph based on the spatial divergence degree. The specific implementation is as follows: First, for each pair of device nodes to be evaluated in the physical topology graph, we need to obtain their vector representations in the topology vector space generated in step S2 and the association vector space generated in step S3. The vector representation in the topology vector space is a low-dimensional real-valued vector learned by each device node through a graph embedding algorithm, encoding the node's structural location information in the factory's physical network. The vector representation in the association vector space is a low-dimensional real-valued vector obtained by the same device node through a dimensionality reduction algorithm, encoding the global association pattern information between the node and all other nodes in the network based on runtime sequence data. During the acquisition operation, based on the unique device identifier of each device node, queries are performed in the database or index file storing the topology vector representations and the database or index file storing the association vector representations, respectively. These vectors are real-valued arrays of the same dimension, for example, 64-dimensional real-valued arrays, ensuring that subsequent distance calculations are mathematically feasible. This step is fundamental to all subsequent calculations and must guarantee the accurate correspondence between the vector data and the device node identifiers.

[0030] After successfully obtaining the vector representations of a pair of device nodes in two vector spaces, the next step is to calculate the distance between the device node pair in the two different spaces using a spatial distance metric. Specifically, Euclidean distance is used as the spatial distance metric. For the topological vector space, the calculation process is as follows: The topological vector of device node A is vector Atopo, which is a real number array containing k elements; the topological vector of device node B is vector Btopo, which is also a real number array containing k elements. The Euclidean distance is calculated by first calculating the difference between the two vectors in each corresponding dimension, i.e., subtracting the first element of vector Atopo from the first element of vector Btopo, subtracting the second element of vector Atopo from the second element of vector Btopo, and so on until the kth element. Then, the difference calculated in each dimension is squared. Next, the squared differences of all k dimensions are summed to obtain a total. Finally, the square root of this sum is taken, and the result is the Euclidean distance between device node A and device node B in the topological vector space, denoted as distance Dtopo. For the associated vector space, the exact same calculation process is used: the associated vector of device node A is vector Aassoc, and the associated vector of device node B is vector Bassoc; their Euclidean distance Dassoc is then calculated. The entire calculation process is executed automatically by the program, ensuring the accuracy of the numerical calculation. For example, if the values ​​of two vectors are close in multiple dimensions, the calculated distance value is smaller, indicating that the two nodes are closer in the vector space; if the values ​​differ significantly, the distance value is larger, indicating that the nodes are farther apart. By calculating separately, two quantified distance values ​​are obtained for the same pair of device nodes in the topological vector space characterizing the physical structure and the associated vector space characterizing data behavior. These two distance values ​​are comparable because they are based on the same metric, namely Euclidean distance, and the same vector dimensions.

[0031] After obtaining the distance values ​​Dtopo and Dassoc of the device node pair in the two spaces, they need to be merged into a single evaluation metric, namely, spatial divergence. Specifically, the ratio of the distance Dassoc in the association vector space to the distance Dtopo in the topology vector space is defined as the spatial divergence of the device node pair. This ratio has a clear physical meaning: the denominator Dtopo reflects the structural proximity of the two device nodes in the physical connection network, and its value is usually negatively correlated with the tightness of the physical connection; that is, the more direct the physical connection or the shorter the path, the smaller the Dtopo value should theoretically be. The numerator Dassoc reflects the similarity of the association patterns exhibited by the two device nodes in their runtime sequence data; the smaller the value, the more synchronized the data behavior. Therefore, when the distance Dassoc in the data association space is much greater than the distance Dtopo in the physical topology space, the spatial divergence will be a value much greater than 1. This indicates that the two devices should be physically closely connected, but their runtime data shows significant differences, thus constituting divergence. Conversely, if the spatial deviation is close to or less than 1, it indicates that the data association pattern is roughly consistent with or more closely matches the expected physical topology. This calculation needs to be performed on each of the evaluated device node pairs, generating a quantified spatial deviation value for each pair. This value is the core criterion for determining whether potential spurious associations exist.

[0032] After calculating the spatial deviation of all device node pairs, potential spurious associations in the knowledge graph are identified based on this value. The key to this step is setting a reasonable preset deviation threshold and comparing the spatial deviation of each device node pair with this threshold. Specifically, device node pairs with spatial deviations exceeding the preset threshold are identified as potential spurious associations in the knowledge graph that require further review. The preset deviation threshold is not a fixed value but is dynamically adjusted based on the determinism of the physical connections of industrial equipment. The determinism of physical connections can be pre-classified according to connection type and assigned different base threshold values. In implementation, a connection type classification table needs to be established first, defining different types of physical connections and their determinism levels. For example, device node pairs directly and uniquely connected via rigid pipes or physical cables have high determinism and can be classified as strongly deterministic connections. Device node pairs indirectly connected via wireless signals, logical links, or multiple intermediate devices have lower determinism and can be classified as weakly deterministic connections. For each connection type, a preset deviation threshold is set by analyzing historical normal operating data. Specifically, a sufficiently long historical period, confirmed as normal system operation, is selected, and the spatial deviation of all device node pairs belonging to that connection type within that period is calculated, forming a set of spatial deviation values ​​for that type. Then, statistical analysis is performed on this set, such as calculating its percentiles. The preset deviation threshold is set to the value corresponding to a higher percentile of this set, such as the 95th or 98th percentile. This means that under historical normal conditions, only no more than 5% or 2% of the spatial deviations of this type of connection will exceed this threshold. Therefore, in real-time or offline analysis, when the spatial deviation of a new device node pair exceeds the preset deviation threshold dynamically determined according to its connection type, there is reason to believe that its deviation degree has exceeded the historical normal range, thus identifying it as a potential spurious association. For example, a pump and valve pair classified as a deterministic strong connection has a historical spatial deviation 95th percentile of 1.5; therefore, this value of 1.5 is set as the preset deviation threshold for this type of connection. In the analysis, if the calculated spatial deviation of the device node is 2.1, since 2.1 is greater than 1.5, this association will be identified as a potentially spurious association. Through this dynamic threshold setting and comparison method based on historical data statistical analysis and connection type classification, it is possible to systematically and adaptively filter out suspicious links with significant contradictions between physical constraints and data performance from the massive associations in the knowledge graph, providing a clear target for subsequent root cause analysis.

[0033] S5. Extract key operation events from the operation log, analyze the causal relationship between potential spurious associations and key operation events within the event isolation window corresponding to the key operation events, and compare the intensity change pattern of potential spurious associations in the pre-event and post-event stages to generate event dependency analysis results. The specific implementation is as follows: First, key operational events need to be extracted from the industrial equipment's operation logs. Operation logs are typically stored in text files or database tables, with each record containing a timestamp field and an event description field. During parsing, log records are read line by line, and predefined keywords or patterns are identified from the event description field using string matching rules or regular expressions to determine if the record represents a key operational event, such as equipment status switching, process recipe changes, or control command triggering. Simultaneously, the date and time of the event are precisely parsed from the timestamp field of the same record, with a time accuracy of at least the second, ensuring that its time base is synchronized with the network time protocol used in step S1. For example, a log record containing "2025-10-27 14:30:05|INFO|Pump P-101 switched from automatic mode to manual mode," by matching the keyword "switched," can be parsed as a key operational event of equipment status switching, with its precise occurrence time being 2025-10-27 14:30:05. Furthermore, based on features in the event description text or event coding, the parsed events are classified into planned operational events or sudden abnormal events. Classification can be achieved based on a predefined classification rule table, which lists specific keywords or coding prefixes used to identify planned operational events, such as planned initiation, work order number, and recipe number, as well as specific keywords or coding prefixes used to identify sudden abnormal events, such as alarm, fault, anomaly, and emergency stop. Classification is completed by matching the event descriptions against this rule table. This step lays the foundation for subsequent differential analysis.

[0034] After extracting and classifying key operational events, an analysis time range needs to be defined for each event, forming an event isolation window. In practice, using the precise occurrence time of the analyzed event as the central reference point, continuous time intervals are defined both forward and backward. The forward-defined time interval constitutes the pre-event phase, used to analyze the steady-state or gradual change process of the system before the event occurs; the backward-defined time interval constitutes the post-event phase, used to analyze the system's response and the establishment of a new steady state after the event occurs. The total length of the event isolation window and the time lengths of the preceding and following phases are not fixed values ​​but are differentiated based on the event classification results. For events classified as planned operational events, since their occurrence time, execution process, and expected impact are usually predictable, the length of their pre-event phase can be set relatively short, mainly to capture the system state immediately before the event triggers, for example, 2 to 5 minutes before the event occurs; the length of the post-event phase needs to be set relatively long to fully cover the entire process of system state adjustment and stabilization after the operation is executed, for example, 20 to 60 minutes after the event occurs. For events classified as sudden abnormal events, their occurrence is both accidental and sudden. The focus of the analysis is on tracing the signs before the anomaly occurs. Therefore, the duration of the pre-event phase should be set sufficiently long, such as 10 to 30 minutes before the event, to allow for retrospective analysis of gradual or abrupt changes in data indicators. The duration of the post-event phase can be set shorter, primarily to capture the direct impact of the fault, such as 5 to 15 minutes after the event. The specific duration is chosen based on the process characteristics of the industrial process, equipment response time, and historical experience. For example, for a large, slow-responding heating process, the post-event phase of planned operations may require a longer duration.

[0035] Subsequently, for each group of potential false associations identified in step S4, the changes in association strength are analyzed within the event isolation window. Specifically, for two device nodes involved in a group of potential false associations, based on the data units with acquisition timestamps collected in step S1, all data units within the time range of the pre-event and post-event phases of the current event are extracted. The extraction method is as follows: based on the device node identifier, all associated data units are found, and then those data units whose acquisition timestamp values ​​are greater than or equal to the start time and less than or equal to the end time of the pre-event phase are selected to form the pre-event phase data subset; similarly, data units whose acquisition timestamp values ​​are greater than or equal to the start time and less than or equal to the end time of the post-event phase are selected to form the post-event phase data subset. Then, based on these data subsets, the statistical characteristics of the association strength of this group of device nodes in the two phases are calculated. During calculation, for the data subset of the preceding event phase, the data values ​​corresponding to the two device nodes are arranged in chronological order to form two time series. Using the same time series similarity measurement method as in step S3, a correlation strength value is calculated as the statistical feature of the correlation strength for this phase, such as the mean correlation strength of the preceding event phase. The same operation is performed on the data subset of the following event phase to obtain the mean correlation strength of the following event phase. The statistical features of correlation strength can also include variance, maximum value, minimum value, etc., to more comprehensively describe the distribution and stability of the correlation strength.

[0036] After calculating the statistical characteristics of the correlation strength between the preceding and following event stages, time series analysis techniques are used to compare their changes and determine the causal orientation. In practice, the statistical characteristics of the correlation strength between the preceding and following event stages are directly compared. For example, the difference between the mean correlation strength of the following event stage and the mean correlation strength of the preceding event stage is calculated. If the difference is positive and greater than a minimum significant change threshold, the correlation strength is considered to have increased; if the difference is negative and its absolute value is greater than the threshold, the correlation strength is considered to have decreased; otherwise, it is considered to have remained essentially unchanged. This minimum significant change threshold can be determined based on the natural fluctuation range of the correlation strength of the group of device node pairs in long-term historical data, for example, set to 1 or 2 times the standard deviation of its historical correlation strength. This is the intensity change pattern. Determining the causal orientation requires more refined time series analysis. One implementation method is to slide a short time window at fixed time intervals within the entire event isolation window coverage period, calculating the instantaneous correlation strength of the potential spurious correlation within each short window, thereby obtaining a high-temporal-resolution sequence of correlation strength changes over time. A mutation point detection algorithm is applied to this intensity time series. For example, by calculating the first difference of the sequence and setting a mutation detection threshold, when the absolute value of the first difference continuously exceeds the threshold, that moment is determined to be a mutation point in the correlation intensity. The mutation detection threshold can be set based on the standard deviation of the fluctuation of the intensity sequence outside the neighborhood of the mutation point. Then, the time sequence of the identified correlation intensity mutation point is strictly compared with the precise occurrence time of the critical operational event: if the time of the correlation intensity mutation point occurs after the occurrence time of the critical operational event, it is determined that the occurrence of the critical operational event caused the subsequent change in correlation intensity, and the causal point is that the event caused the correlation change; if the time of the correlation intensity mutation point occurs before the occurrence time of the critical operational event, it is determined that the abnormal change in correlation intensity may have been recorded as a precursor or trigger before the occurrence of the critical operational event, and the causal point is that the correlation change predicts the event. Finally, the generated event dependency analysis result is a structured record, the content of which includes at least the identifier of the analyzed critical operational event, the identifier of the device node pairs involved in the potential spurious correlation, the determined causal point category, and a description of the specific correlation intensity change pattern.

[0037] S6. Update the relationships in the knowledge graph based on the event dependency analysis results. For potential false relationships with event-driven persistent association characteristics, mark them as event-dependent relationships in the knowledge graph. The specific implementation is as follows: First, the event dependency analysis results generated in step S5 for each potential spurious association are refined and screened. The goal of the screening is to identify associations with event-driven, persistent association characteristics. The screening process requires a logical AND operation based on two explicit conditions. The first condition is the causal orientation condition, which involves reading the causal orientation field from the event dependency analysis results and filtering out association records that are clearly determined to be caused by an event, indicating that the change in the association is attributed to the occurrence of a key operational event. The second condition is the continuously strengthening association strength condition, which requires further analysis of the strength change patterns recorded in the event dependency analysis results. The strength change patterns record the comparison between the statistical characteristics of the association strength in the later event stage and the statistical characteristics of the association strength in the earlier event stage. In implementation, a quantitative criterion for judging a stable increase needs to be set. This criterion contains two sub-criterions that must be met simultaneously. The first sub-criterion is the significant increase criterion, which means that the increase in the mean association strength in the later event stage relative to the mean association strength in the earlier event stage must exceed a preset minimum significant increase threshold. The magnitude of the increase is calculated by subtracting the mean association strength of the preceding event phase from the mean association strength of the subsequent event phase, and then dividing by the mean association strength of the preceding event phase. The minimum significant increase threshold can be obtained by analyzing a series of historically confirmed normal event-driven association cases. Specifically, the distribution of increase values ​​in these cases is calculated, and a higher percentile of this distribution is taken as the threshold, such as the 90th or 95th percentile, corresponding to values ​​of 0.1 (10%) or 0.15 (15%). The second sub-criteria is the stability criterion, which requires that the variance of the multiple instantaneous association strength values ​​calculated within the subsequent event phase must be lower than a preset volatility stability threshold. These instantaneous association strength values ​​are calculated in step S5 using a sliding window method throughout the entire subsequent event phase. The volatility stability threshold is also obtained based on the analysis of historical normal cases, calculating the distribution of the variance of these cases' subsequent event phase variances, and taking a lower percentile of this distribution as the threshold, such as the 10th or 20th percentile. Only potential spurious associations that simultaneously meet the above-mentioned causal orientation condition, significant enhancement criterion, and stability criterion will be ultimately identified as associations with event-driven persistent association characteristics, and thus enter the subsequent update process.

[0038] After identifying the set of relationships with event-driven, persistent association characteristics, it is necessary to find and manipulate the specific representations corresponding to these abstract relationships within the actual stored knowledge graph structure. In practice, based on the unique device node pair identifiers corresponding to each identified relationship, queries and location are performed in the knowledge graph's graph database or graph structure model. The knowledge graph uses device nodes as entities and the data associations between device nodes as relation edges. The specific process of the location operation is to use the two device node identifiers in the device node pair identifiers as query keys to search the knowledge graph for the existence of a relation edge that starts and ends with these two device nodes. This relation edge may be generated based on general rules during initial construction or preliminarily established in step S3 based on the strength of the data association. For example, for the relationship identified between device nodes P-101 and V-201, the relation edge connecting these two nodes, representing a data association, is located in the knowledge graph. This step ensures that subsequent label updates can accurately apply to the correct semantic relationship instances in the knowledge graph.

[0039] After successfully locating the corresponding relationship edge in the knowledge graph, its attributes are semantically labeled to reflect its event dependency characteristics. Specifically, a new event-dependent relationship label is added to the relationship edge. This involves modifying the attribute list of the relationship edge, adding a new attribute field named "Relationship Type," and setting its value to "Event Dependency." Simultaneously, the core information of the key operational event that triggered the formation or enhancement of this relationship must be stored as the association attribute of the event-dependent relationship label. These association attributes include the event identifier of the key operational event, which is the event number or description string uniquely determined from the log in step S5. The association attributes also include the event isolation window information corresponding to the key operational event, including the start and end times of the event isolation window. This information is directly obtained from the event dependency analysis result record generated in step S5 and written into the attributes of the relationship edge. After storage, the knowledge graph update is complete. For example, for the relationship edge connecting device nodes P-101 and V-201, add "Relationship Type: Event Dependency" to its attributes, and add the attributes "Trigger Event Identifier: Pump P-101 Automatic Mode Switching_20251027143005", "Event Window Start Time: 2025-10-27 14:28:05", and "Event Window End Time: 2025-10-27 14:50:05". After this step, the knowledge graph not only records the relationships between devices, but also further reveals the dependency and response patterns of specific relationships on specific operational events. This elevates the semantic level of the knowledge graph from static structural relationships to dynamic causal relationships, providing richer queryable and inferable knowledge for subsequent root cause tracing, impact surface analysis, or production strategy optimization.

[0040] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0041] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0042] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. Computer-readable storage media can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0043] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0044] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0045] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0046] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0047] If a function is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0048] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0049] In conclusion, the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for managing data from multi-source heterogeneous industrial equipment, characterized in that, include: S1. Collect data units from multi-source heterogeneous industrial equipment, and associate each data unit with the equipment identifier and collection timestamp of its source equipment; S2. Obtain the physical topology network of industrial equipment and map the device nodes in the physical topology network to the topology vector space; S3. Calculate the data association strength between device nodes based on data units, and map the data association strength to the association vector space; S4. Calculate the spatial divergence of each pair of device nodes in the topology vector space and the association vector space, and identify potential false associations in the knowledge graph based on the spatial divergence. S5. Extract key operation events from the operation log, analyze the causal relationship between potential false associations and key operation events within the event isolation window corresponding to the key operation events, and compare the intensity change pattern of potential false associations in the pre-event stage and the post-event stage to generate event dependency analysis results. S6. Update the relationships in the knowledge graph based on the results of event dependency analysis. For potential false relationships with event-driven continuous association characteristics, mark them as event-dependent relationships in the knowledge graph.

2. The method for managing multi-source heterogeneous industrial equipment data according to claim 1, characterized in that, S1 includes: Based on the communication protocols and data interfaces of industrial equipment, multimodal time-series data from programmable logic controllers, sensors, and manufacturing execution systems are analyzed from different sources to generate data units; Map the logical identifiers of the same industrial physical equipment across different data sources to a unified equipment identifier; The timestamps of data units collected from different data sources are synchronized and aligned based on a unified network time protocol.

3. The data management method for multi-source heterogeneous industrial equipment according to claim 1, characterized in that, S2 include: Based on the factory piping and instrumentation diagram or equipment connection relationship database, construct a physical topology graph with industrial equipment as equipment nodes and physical connection relationships between equipment as edges; Based on the device identifier, associate the device nodes and data units in the physical topology diagram; The graph embedding algorithm maps the connection structure of device nodes in the physical topology graph to a low-dimensional dense topology vector space, so that device nodes with similar connections have similar vector representations in the topology vector space.

4. The data management method for multi-source heterogeneous industrial equipment according to claim 1, characterized in that, S3 includes: Based on the collected data units with collection timestamps, the temporal data association strength between each pair of device nodes is quantified using a time series similarity measurement method. Based on the calculated data association strength between all pairs of device nodes, a data association strength matrix is ​​constructed. A dimensionality reduction algorithm is used to map the association strength relationship between each pair of device nodes in the data association strength matrix to a low-dimensional dense association vector space, so that the vector representations of device node pairs with similar association strengths in the association vector space are also similar.

5. A method for managing data of multi-source heterogeneous industrial equipment according to claim 1, characterized in that, S4 include: For each pair of device nodes, obtain their vector representation in the topology vector space and their vector representation in the associated vector space. The spatial distance metric method is used to calculate the distance between the vector representations of corresponding device nodes in the topological vector space, and the distance between the vector representations in the associated vector space. The ratio of the distance in the associated vector space to the distance in the topological vector space is defined as the spatial divergence of the corresponding device node. Device node pairs whose spatial deviation exceeds a preset deviation threshold are identified as potential false associations in the knowledge graph.

6. The data management method for multi-source heterogeneous industrial equipment according to claim 5, characterized in that, Distances are calculated using spatial distance metrics, including Euclidean distance.

7. The data management method for multi-source heterogeneous industrial equipment according to claim 5, characterized in that, Device node pairs whose spatial deviation exceeds a preset deviation threshold are identified as potentially false associations. This includes dynamically setting different preset deviation thresholds for different types of device node pairs based on the degree of determinism of the physical connection of industrial equipment.

8. The data management method for multi-source heterogeneous industrial equipment according to claim 1, characterized in that, S5 includes: From the operation logs of industrial equipment, extract key operational events that indicate equipment status switching, process formula changes, or control command triggering, along with their precise occurrence times; Centered on the precise occurrence time of critical operational events, time ranges are defined both before and after the event to form an event isolation window that includes the pre-event phase and the post-event phase. For each potential false association identified, based on the collection timestamp, the corresponding data units in the pre-event stage and the post-event stage are extracted respectively, and the statistical characteristics of the association strength of the corresponding association in the pre-event and post-event stages are calculated. By comparing the statistical characteristics of association strength before and after an event using time series analysis, and combining the abrupt change points of association strength with the chronological order of events, the causal relationship between potential spurious associations and key operational events is determined, generating event dependency analysis results that include causal relationships and intensity change patterns.

9. A method for managing data of multi-source heterogeneous industrial equipment according to claim 8, characterized in that, The process of parsing key operational events includes classifying events into planned operational events or sudden abnormal events based on event description text or event codes; and forming event isolation windows, which involves setting different pre-event and post-event phase durations for planned operational events and sudden abnormal events based on the event classification results.

10. A method for managing data of multi-source heterogeneous industrial equipment according to claim 1, characterized in that, S6 include: Based on the results of event dependency analysis, potential false associations that are determined to be caused by key operational events and whose association strength statistical characteristics are consistently higher in the later event stage than in the earlier event stage are identified as associations with event-driven persistent association characteristics. Based on the device node pairs corresponding to the determined associations, locate the corresponding association edges in the knowledge graph; Add event-dependent relationship tags to the edges of the relationships, and store the event identifiers and event isolation window information of the key operation events that trigger the relationships as the association attributes of the event-dependent relationship tags to complete the update of the knowledge graph.

Images