CN-121997230-A - Intelligent factory new energy safety early warning method based on knowledge graph

CN121997230ACN 121997230 ACN121997230 ACN 121997230ACN-121997230-A

Abstract

The invention discloses a knowledge graph-based intelligent factory new energy safety early warning method which comprises the steps of constructing a unified time sequence data set by collecting multi-source data in operation of a plurality of devices, modeling complex dependency relations among various operation parameters by utilizing a Vine Copula method to obtain condition dependency strength among variables, converting an identified abnormal state into an event sequence, establishing a trigger relation among the events by adopting a Hawkes process, reversely adjusting the dependency structure among the variables by the event trigger strength, realizing dynamic update of the structure, constructing a risk graph structure combining the variable dependency and an event trigger mechanism, developing path reasoning, identifying potential risks of target variables and outputting early warning information. The method can comprehensively analyze the relationship between the data dependence and the event influence among the devices, and realize dynamic identification and early warning of potential risks in a complex industrial system.

Inventors

WANG CUI
RONG LETIAN
GAO YUANYUAN
SHAO YUANYUAN
ZHENG RONG

Assignees

青岛城市学院

Dates

Publication Date: 20260508
Application Date: 20260127

Claims (8)

1. A knowledge graph-based intelligent factory new energy safety pre-warning method is characterized by comprising the following steps of: Collecting multi-source heterogeneous operation data of a plurality of new energy devices in an intelligent factory, preprocessing the multi-source heterogeneous operation data, and constructing a standardized time sequence feature set; based on working condition variables in the standardized time sequence feature set, a Vine Copula method is adopted to construct a high-dimensional dependency structure diagram comprising a plurality of nodes and edges; Extracting condition dependency intensity among working condition variables from the high-dimensional dependency structure diagram to form a dependency coefficient matrix; constructing an abnormal event sequence based on the abnormal state identified in the standardized time sequence feature set, modeling the abnormal event sequence by adopting a Hawkes process, introducing the condition dependent intensity among the working condition variables in the dependency coefficient matrix into a trigger intensity adjusting parameter in a Hawkes process, and generating an abnormal event trigger intensity sequence; Reversely updating the high-dimensional dependency structure diagram by using an abnormal event trigger intensity sequence generated in Hawkes process to generate a dynamically updated high-dimensional dependency structure diagram; and constructing a risk knowledge graph comprising a dynamically updated high-dimensional dependency structure diagram and an abnormal event triggering intensity sequence, and carrying out path reasoning based on the risk knowledge graph to output risk early warning information of a target working condition variable of the new energy equipment.
2. The intelligent factory new energy safety pre-warning method based on the knowledge graph of claim 1, wherein the pre-processing comprises data cleaning, time alignment and normalization.
3. The knowledge-graph-based intelligent factory new energy safety pre-warning method according to claim 1, wherein constructing a high-dimensional dependency structure diagram comprising a plurality of nodes and edges comprises: Collecting all new energy equipment working condition variables in the standardized time sequence feature set, and establishing a new energy equipment working condition variable set; Respectively carrying out edge distribution fitting treatment on each working condition variable in the working condition variable set of the new energy equipment to generate an edge distribution function set, and converting the working condition variable set of the new energy equipment into a pseudo-observation data set for dependency modeling; Based on the pseudo-observation data set, selecting Vine Copula structural types of the model, and determining the arrangement sequence of the working condition variables of the new energy equipment in the dependent structural modeling; Constructing a first-layer tree structure, wherein each node in the first-layer tree structure corresponds to a working condition variable of new energy equipment, and each side is connected with two working condition variables and used for representing the marginal dependency relationship; Determining a condition variable set according to two working condition variables connected with edges in the first-layer tree structure, recursively constructing a second-layer tree structure and a tree structure of a subsequent level thereof on the basis of the condition variable set, wherein each edge of each layer represents a condition dependency relationship of two working condition variables under a given condition variable set; In the construction process of each layer of tree structure, parameter estimation of a dependency function is carried out, and two working condition variables, the level, a condition variable set and fitting parameters thereof of each side are uniformly recorded as a structural dependency relation unit; after the construction of all tree structure levels is completed, extracting all working condition variable nodes as node sets of the high-dimensional dependency structure diagram, and extracting all structural dependency relation units as edge sets; The high-dimensional dependency structure diagram is composed of node sets and edge sets, wherein each node represents a working condition variable of the new energy equipment, and each edge comprises two working condition variables, a tree structure level, a dependent condition variable set and a dependency strength parameter and is used for representing a multi-level statistical dependency structure among the working condition variables of the new energy equipment.
4. The knowledge-graph-based intelligent factory new energy safety pre-warning method according to claim 1, wherein the forming of the dependency coefficient matrix comprises: extracting all edge sets from the high-dimensional dependency structure diagram, wherein each edge is connected with two new energy equipment working condition variables and comprises dependency strength parameters obtained by a modeling process; Extracting two new energy equipment working condition variables connected with the corresponding edges from each edge in the edge set, determining unique index positions of the two new energy equipment working condition variables in the new energy equipment working condition variable set, and recording corresponding dependent intensity parameter values of the two new energy equipment working condition variables; Based on the working condition variable set of the new energy equipment, a two-dimensional matrix structure is established, and each row and each column of the two-dimensional matrix respectively correspond to one working condition variable of the new energy equipment; and filling all the new energy equipment working condition variables and the dependence intensity parameters thereof extracted from the edge set into corresponding positions in the two-dimensional matrix, constructing a numerical matrix representing the dependence intensity of the variables on the conditions, and generating a dependence coefficient matrix.
5. The knowledge-graph-based intelligent factory new energy safety pre-warning method according to claim 1, wherein the generating of the abnormal event trigger intensity sequence comprises: monitoring historical data of all new energy equipment working condition variables in the standardized time sequence feature set, and identifying an abnormal state through a set rule; Constructing each identified abnormal state as an abnormal event unit, wherein each abnormal event unit comprises abnormal time, abnormal type, the condition variable identifier of the new energy equipment and the characteristic value thereof; arranging all abnormal event units according to time sequence to construct a new energy equipment abnormal event sequence; Extracting dependency strength parameters between the working condition variables corresponding to the abnormal event units and other working condition variables from the dependency coefficient matrix; The new energy equipment abnormal event sequence is used as an input event stream, the dependent intensity parameter is used as a trigger gain input, and the trigger gain input is input into a Hawkes process model for establishing a self-excitation and cross-excitation mechanism between abnormal events; And calculating the triggering intensity of each abnormal event unit in a given time interval through Hawkes process models, and outputting an abnormal event triggering intensity sequence of the new energy equipment according to the time sequence structure and the historical occurrence condition of the abnormal event sequence, wherein the abnormal event triggering intensity sequence corresponds to the triggering intensity value generated by each event unit.
6. The knowledge-graph-based intelligent factory new energy safety pre-warning method according to claim 1, wherein Hawkes comprises the following steps: Extracting each abnormal event unit in the abnormal event sequence of the new energy equipment, recording the abnormal occurrence time, the corresponding working condition variable identification and the abnormal type of the new energy equipment, and constructing an input event stream arranged in time sequence; Setting basic intensity parameters corresponding to each type of abnormal event; Extracting the condition dependency intensity between the working condition variable of the new energy equipment corresponding to each abnormal event unit and other working condition variables from the dependency coefficient matrix to form a dynamic trigger gain factor set; identifying causal transfer paths between the abnormal event unit and all historical events according to the time position of the abnormal event unit in the abnormal event sequence, determining the propagation level between the abnormal event sequences, and generating a corresponding chain attenuation weight set; For each target abnormal event unit, searching all historical abnormal events occurring before the time point of each target abnormal event unit, and calculating trigger contribution values of the historical abnormal events to the target events one by one; when the trigger contribution value of each historical event is calculated, the trigger kernel function output result is called, and the trigger contribution value is multiplied by a dynamic trigger gain factor and a chain attenuation weight corresponding to the historical abnormal event to be used as the current contribution value; Adding the contribution values of all the historical abnormal events with the basic intensity parameters of the current event to obtain a final trigger intensity value of the corresponding target abnormal event unit; repeating the calculation steps for all the abnormal event units in the abnormal event sequence, and outputting the trigger intensity values of all the abnormal event units according to time sequence to form the abnormal event trigger intensity sequence.
7. The knowledge-graph-based intelligent factory new energy safety pre-warning method according to claim 1, wherein the generating of the dynamically updated high-dimensional dependency structure comprises: Extracting all edge sets from the high-dimensional dependency structure diagram, wherein each edge is connected with two new energy equipment working condition variables and is associated with an original condition dependency strength; Searching the abnormal event units containing two corresponding variables in the abnormal event triggering intensity sequence for the working condition variables of the two new energy equipment connected with each side, screening all events occurring in a specified time window, and extracting the triggering intensity values of the abnormal event units; Grouping and calculating the screened trigger intensity values according to the working condition variable pairs to obtain the combined trigger intensity mean value of the corresponding side corresponding variable pairs; for each edge, constructing a dependency intensity correction function according to the numerical relation between the joint trigger intensity mean value and the original condition dependency intensity, and outputting updated condition dependency intensity; assigning the updated condition dependent intensity to the corresponding edge structure, replacing the edge parameters in the original edge set, and generating a new edge set; and keeping the node set in the original high-dimensional dependency structure diagram unchanged, and combining the updated edge set with the original node set to generate the dynamic updated high-dimensional dependency structure diagram.
8. The knowledge-graph-based intelligent factory new energy safety pre-warning method according to claim 1, wherein outputting risk pre-warning information of target working condition variables of new energy equipment comprises: the method comprises the steps of fusing a high-dimensional dependency structure diagram after dynamic updating with an abnormal event triggering intensity sequence, and constructing a risk knowledge graph, wherein the risk knowledge graph comprises a node set and a directed edge set, the nodes represent new energy equipment working condition variables and corresponding abnormal states thereof, and the directed edges represent the dependency relationship between variable pairs and the propagation direction of an abnormal event; For each edge in the high-dimensional dependency structure diagram, matching abnormal event units corresponding to the working condition variable pairs in the trigger intensity sequence, extracting the trigger intensity values of the abnormal event units, and performing weighted combination on the condition dependency intensities and the trigger intensities to generate edge weight values of directed edges; Marking abnormal event nodes with trigger intensity larger than a set trigger threshold in a risk knowledge graph based on the generated edge weight of the directed edge as initial risk source nodes; Starting from each initial risk source node, traversing all reachable paths according to a risk knowledge graph structure, recording all passed nodes, edge weights and propagation sequences on the paths, and calculating the accumulated risk propagation strength of each path; Establishing a path set generated by traversing as a path record set, wherein each path record comprises a path starting point, a path end point, a working condition variable sequence of passing new energy equipment, accumulated risk propagation intensity and path depth; Screening propagation paths which take target new energy equipment working condition variables as path end points in the path record set and have accumulated risk propagation intensity higher than a set risk propagation threshold value, and taking the propagation paths as candidate risk paths; and (3) sorting the path structure, the target working condition variable identification, the accumulated risk propagation intensity and the corresponding trigger time window in the candidate risk path into risk early warning information of the target working condition variable of the new energy equipment.

Description

Intelligent factory new energy safety early warning method based on knowledge graph Technical Field The invention relates to the technical field of industrial intelligent monitoring and safety early warning, in particular to a novel intelligent factory energy safety early warning method based on a knowledge graph. Background With the acceleration of the digitization of intelligent factories, energy Internet and industrial equipment, the safe operation management of new energy equipment has become an important component of the reliability guarantee of industrial systems. Currently, a large amount of heterogeneous data generated in the running process of new energy equipment is generally collected through a sensor and uploaded to an edge computing node or a cloud platform, and state monitoring and abnormal early warning are performed by using rule matching, feature extraction or a machine learning algorithm. The existing new energy safety early warning method still has significant defects in the aspects of dependency modeling and risk propagation mechanism construction for processing multi-source equipment data. On one hand, the traditional method mostly adopts fixed correlation coefficient, distance measurement or artificial characteristic engineering to carry out variable correlation modeling, and nonlinear and multi-level statistical dependency structures among the multidimensional working condition variables of the new energy equipment are difficult to reveal, so that the real coupling relation among the variables cannot be effectively described. On the other hand, the mainstream anomaly detection algorithm is mostly based on static threshold judgment or classification model output, a dynamic trigger mechanism between events cannot be constructed, complex propagation phenomena such as chain faults, secondary risks and the like cannot be reflected, linkage updating capability between variable dependence and event triggering is also lacked, and timeliness and accuracy of early warning are affected. Therefore, how to provide a knowledge graph-based intelligent factory new energy safety pre-warning method is a problem to be solved by those skilled in the art. Disclosure of Invention The invention aims to provide a knowledge graph-based intelligent factory new energy safety early warning method, which utilizes Vine Copula method to construct nonlinear dependency structure among new energy equipment multisource variables, combines Hawkes process to establish triggering mechanism among abnormal events, realizes dynamic coupling update of variable dependency relationship and event propagation intensity, and finally carries out path reasoning and outputs risk early warning information of new energy equipment target variables based on constructed risk knowledge graph. The method has the capabilities of modeling high-dimensional dependence, describing chain risk and realizing dynamic evolution and interpretable reasoning. According to the embodiment of the invention, the intelligent factory new energy safety pre-warning method based on the knowledge graph comprises the following steps: Collecting multi-source heterogeneous operation data of a plurality of new energy devices in an intelligent factory, preprocessing the multi-source heterogeneous operation data, and constructing a standardized time sequence feature set; Based on working condition variables in the standardized time sequence feature set, a Vine Copula method is adopted to construct a high-dimensional dependency structure diagram containing a plurality of nodes and edges, and the working condition variable high-dimensional dependency structure diagram of the new energy equipment comprises a plurality of working condition variable nodes and corresponding condition dependency edge weights; Extracting condition dependency intensity among working condition variables from the high-dimensional dependency structure diagram to form a dependency coefficient matrix; constructing an abnormal event sequence based on the abnormal state identified in the standardized time sequence feature set, modeling the abnormal event sequence by adopting a Hawkes process, introducing the condition dependent intensity among the working condition variables in the dependency coefficient matrix into a trigger intensity adjusting parameter in a Hawkes process, and generating an abnormal event trigger intensity sequence; Reversely updating the high-dimensional dependency structure diagram by using an abnormal event trigger intensity sequence generated in Hawkes process to generate a dynamically updated high-dimensional dependency structure diagram; and constructing a risk knowledge graph comprising a dynamically updated high-dimensional dependency structure diagram and an abnormal event triggering intensity sequence, and carrying out path reasoning based on the risk knowledge graph to output risk early warning information of a target working condition variable of the new energy equipmen