Whole-process intelligent disposal management method and system based on alarm signal

Abstract

The invention discloses a whole-process intelligent disposal management method based on an alarm signal, and relates to the technical field of XX. The method comprises the following steps: monitoring a transmission folder in real time by adopting a Kettle data processing tool, and importing an alarm signal data file transmitted by an OCS system into a database once the alarm signal data file is found; executing a cleaning program on the warehousing data, and extracting core information of the alarm signal; constructing and continuously updating an expert knowledge base, performing matching in the expert knowledge base according to the extracted core information, and correspondingly generating an expert knowledge model; according to the generated expert knowledge model, processing steps of the alarm signals are automatically split according to stages to generate sub-processes; and judging whether an automatic disposal criterion is met or not according to a matching result, if yes, automatically executing, and if not, executing after field personnel confirms. According to the invention, the alarm signal data file transmitted by the power grid monitoring system is automatically sensed and monitored in real time, and the timeliness and accuracy of abnormity monitoring are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of power systems, and in particular to a full-process intelligent disposal management method and system based on alarm signals. Background Art

[0002] As an important basic industry of the national economy, the safe and stable operation of the power system is related to the national economy and people's livelihood. As the scale of the power grid continues to expand, the number of equipment continues to increase, the operation status of the power grid becomes more and more complex, and various abnormal alarm signals are increasing. Traditional manual methods are difficult to handle a large number of abnormal situations in a timely and efficient manner.

[0003] Traditional methods have some shortcomings in the intelligent handling of alarm signals. First, there is a lack of fusion and utilization of multi-source heterogeneous data, making it difficult to obtain comprehensive contextual information to support decision-making; second, there is a lack of knowledge-driven handling model generation mechanism, and manual experience is difficult to systematize and standardize; third, there is a lack of process automation and intelligent management, and the handling process is inefficient; fourth, there is a lack of optimization and iteration mechanism for human-machine collaboration, and the knowledge base and process templates are difficult to continuously improve. Summary of the invention

[0004] In view of the deficiencies in the prior art in intelligent processing of abnormal alarm signals, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to transform the expert disposal knowledge model into an intelligent workflow, introduce a process intelligent management mechanism, and support process adaptive adjustment, parallel execution and exception handling.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, an embodiment of the present invention provides a full-process intelligent disposal management method based on alarm signals, which includes using the Kettle data processing tool to monitor the transmission folder in real time, and once the alarm signal data file transmitted by the OCS system is found, it is imported into the database; a cleaning program is executed on the input data to extract the core information of the alarm signal; an expert knowledge base is constructed and continuously updated, and the expert knowledge base is matched according to the extracted core information to generate an expert knowledge model accordingly; according to the generated expert knowledge model, the disposal steps of the alarm signal are automatically split into stages to generate sub-processes; according to the matching results, it is judged whether the automatic disposal criteria are met, and if so, it is automatically executed, and if not, it is executed after confirmation by on-site personnel; the CART classification tree algorithm is used to predict potential defects based on historical alarm signal data and disposal results, so as to provide a decision-making basis for maintenance.

[0008] As an optimal solution of the full-process intelligent disposal management method based on alarm signals described in the present invention, the cleaning procedure for incoming data includes the following steps: when executing the cleaning procedure for incoming data, an adaptive data cleaning mechanism based on semantic rules is introduced; based on domain ontology and background knowledge, a set of cleaning semantic rule sets R is constructed; for each incoming data D, semantic analysis is performed on it based on the cleaning semantic rule set R to obtain a semantic parse tree T; a scoring function Score(T,R) is defined to measure the score of the semantic parse tree T under the cleaning semantic rule R, and a threshold parameter θ is set; if Score(T,R)>θ, data cleaning is performed according to the semantic structure of T; if Score(T,R)≤θ, the uncertain part of the semantic analysis is fed back, and the semantic rule R is manually reviewed and corrected; based on the manual correction, the scoring function NewScore(T,R') is reconstructed; the incoming data D is re-cleaned based on the latest scoring function NewScore(T,R').

[0009] As a preferred solution of the full-process intelligent disposal management method based on alarm signals described in the present invention, the specific formula of the scoring function Score(T,R) is as follows:

[0010] Score(T,R)=λ1StructureSim(T,R)+λ2ConceptSim(T,R)

[0011] +λ3ContextSim(T,R)+λ4RuleCoverageSim(T,R)

[0012]

[0013] Among them, StructureSim(T,R) represents the similarity score between the semantic parse tree T and the rule R at the structural level, ConceptSim(T,R) represents the similarity score between the semantic parse tree T and the rule R at the conceptual level, ContextSim(T,R) represents the similarity score between the semantic parse tree T and the rule R at the context level, RuleCoverageSim(T,R) represents the coverage of the semantic parse tree T by the rule R, λ1, λ2, λ3, λ4 represent the weight parameters of each score (obtained by tuning the validation set data), N T represents the node set in the semantic parse tree T, n represents the node, MLP represents the multi-layer perceptron network, σ represents the sigmoid function, E T Represents the edge set in the semantic parse tree T, e represents the edge, CosineSim() represents the calculation of the cosine similarity of two vectors, e t1 and e t2represents the entities at both ends of the edge, Embed represents the entity vector representation (mapped to the vector representation in the low-dimensional dense vector space through the embedding technology), P T represents the path set in the semantic parse tree T, p represents the path, BLEU represents the commonly used evaluation index in natural language generation tasks, and p desc represents the path description, GenDesc represents the context generation model, and E R Represents the edge set in rule R.

[0014] The specific formula for reconstructing the scoring function NewScore(T,R') is as follows:

[0015] NewScore(T,R')=αScore(T,R')+(1-α)SimWord(T,D)

[0016] Among them, SimWord represents the bag-of-words similarity, α represents the coupling parameter, Score(T,R') represents the semantic matching degree score of the semantic parse tree T under the manually corrected new rule R', R' represents the manually corrected new rule, T represents the semantic parse tree, and D represents the stored data.

[0017] As a preferred solution of the full-process intelligent disposal management method based on alarm signals described in the present invention, the following steps are included: constructing and continuously updating the expert knowledge base: based on the existing manually annotated historical alarm data, using natural language processing and machine learning technology to automatically extract the core information of the alarm signal; integrating multi-source heterogeneous data from the monitoring system, working condition database, and equipment archives into the expert knowledge base to provide comprehensive context information for abnormal analysis; modeling the expert knowledge base as a knowledge graph for the power grid field, effectively organizing entities and their relationships, and supporting richer queries and reasoning; using graph neural networks to automatically learn historical abnormal patterns on the knowledge graph, accurately match new alarm signals, and generate corresponding expert disposal models; if the new abnormal alarm signal does not exist in the expert knowledge base, manually enter the expert library information and match the expert library again; for the expert disposal model generated by the match, perform expert manual review and feedback correction, and improve the expert knowledge base through continuous human-machine collaborative iterative optimization; the optimized expert disposal model is combined with real-time monitoring data to dynamically generate personalized and explainable abnormal disposal solutions.

[0018] As a preferred solution of the full-process intelligent disposal management method based on alarm signals described in the present invention, the following steps are included in the fusion of multi-source heterogeneous data from the monitoring system, working condition database, and equipment archives into the expert knowledge base: constructing an overall framework G including three knowledge graphs G1, G2, and G3 of monitoring, working condition, and equipment; in each sub-graph G iThe knowledge graph embedding model is trained separately to obtain the embedding representation vector v i ; Define the knowledge graph node alignment scoring function Align(v i ,v j )=δ(W·Concat(v i ,v j )), where δ represents a nonlinear function, W represents a weight matrix, Concat(v i ,v j ) represents the two vectors v i and v j For splicing, v i and v j Represents two different knowledge graphs G i and G j Node embedding vector; maximize the global objective function of alignment score on all node pairs (i, j), and optimize the alignment matrix M; under the optimized alignment matrix M, perform feature exchange of each sub-graph to achieve data fusion; the specific formula of the global objective function of alignment score is as follows:

[0019] max∑Align(v i ,v j )·M ij

[0020] st∑M ij =1,M ij ≥0

[0021] Where M represents the alignment matrix, i and j represent row and column variables respectively, and Align(v i ,v j ) represents the knowledge graph node alignment scoring function, v i and v j Represents two different knowledge graphs G i and G j The node embedding vector of .

[0022] As a preferred solution of the full-process intelligent disposal management method based on alarm signals described in the present invention, the disposal steps of alarm signals are automatically split into stages to generate sub-processes, including the following steps: using the rich semantic information contained in the knowledge graph, dynamically planning the optimal disposal path according to the context association of the alarm signal; modeling the disposal path as an intelligent workflow, introducing process intelligence technology, and supporting adaptive adjustment, parallel execution and exception handling of the process; regarding the disposal process as a sequential decision problem, using reinforcement learning technology to automatically learn and optimize the optimal decomposition path based on a large number of historical cases; while splitting the internal process based on the expert knowledge base, cross-system process orchestration and integration are performed to make the execution of the disposal plan cover a wider range of scenarios; for special and complex abnormal situations, experts are allowed to intervene manually, and the pre-generated process is reviewed, modified and supplemented through visual modeling tools to generate customized disposal process plans; for new knowledge and new practices generated during the execution process, they are automatically summarized and fed back to the expert knowledge base through knowledge extraction technology, and the disposal process templates are continuously optimized and enriched.

[0023] As a preferred solution of the full-process intelligent disposal management method based on alarm signals described in the present invention, generating a customized disposal process plan includes the following steps: designing state s as the currently generated process sequence, action a as the insert / modify node operation, and defining the reward function r(s,a) as the process quality score after inserting the new node; under the reinforcement learning framework, state s, action a, reward r(s,a) construct a Markov decision process to find the optimal generation strategy π * =argmax∑r(s,a); Based on the existing manual processing data, the supervised learning algorithm is used to * Perform pre-training to build a data set of state-action pairs; in the human-computer interaction mode, receive the expert's operation action a and feedback reward r in real time, and continuously adjust and optimize the strategy π through the policy iteration algorithm * ; Build a model interpretability mechanism to explain the reasons for each step of the operation. If the expert does not understand or objects to a certain step, feedback correction will be provided through the model interpretability mechanism; For the final strategy of optimization iteration π * , encoded as a process generation model and integrated into a visual modeling tool to assist experts in process review and customization; the specific formula of the reward function r(s,a) is as follows:

[0024] r(s,a)=w1SimStructure(s',s)+w2SimSemantic(s',s)+w3CostScore(s')

[0025] Among them, s' represents the newly generated process after executing action a, s represents the reference high-quality process, SimStructure and SimSemantic are used to measure the similarity between the newly generated process s' and the reference high-quality process s at the structural and semantic levels, respectively, and w1, w2, and w3 represent the corresponding weights, respectively.

[0026] In the second aspect, an embodiment of the present invention provides a full-process intelligent disposal management system based on alarm signals, which includes a transmission detection module, which is used to use the Kettle data processing tool to monitor the transmission folder in real time, and once the alarm signal data file transmitted by the OCS system is found, it is imported into the database; a cleaning execution module, which is used to execute the cleaning procedure on the incoming data and extract the core information of the alarm signal; a matching generation module, which is used to build and continuously update the expert knowledge base, match the extracted core information in the expert knowledge base, and generate an expert knowledge model accordingly; a process splitting module, which is used to automatically split the alarm signal disposal steps into stages according to the generated expert knowledge model to generate sub-processes; a judgment module, which is used to judge whether the automatic disposal criteria are met according to the matching results, and if so, it is automatically executed, and if not, it is executed after confirmation by the on-site personnel; a defect detection module, which is used to use the CART classification tree algorithm to predict potential defects based on historical alarm signal data and disposal results, and provide a decision-making basis for maintenance.

[0027] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the full-process intelligent disposal management method based on alarm signals as described in the first aspect of the present invention are implemented.

[0028] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored, wherein: when the computer program instructions are executed by a processor, the steps of the full-process intelligent disposal management method based on alarm signals as described in the first aspect of the present invention are implemented.

[0029] The beneficial effects of the present invention are as follows: the present invention realizes real-time automatic perception and monitoring of alarm signal data files transmitted by the power grid monitoring system, thereby improving the timeliness and accuracy of abnormal monitoring; constructs a knowledge graph for the power grid field based on multi-source heterogeneous data fusion, automatically learns historical abnormal patterns, generates accurate expert handling knowledge models, and provides efficient and explainable abnormal handling decision support; transforms the expert handling model into an intelligent workflow, introduces a process intelligent management mechanism, supports adaptive adjustment of the process, parallel execution and abnormal handling, and improves the pertinence and executability of the handling plan; establishes a human-machine collaborative optimization mechanism, continuously optimizes and enriches the expert knowledge base, iteratively generates customized handling processes, and continuously improves the accuracy and high level of abnormal handling decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0031] Figure 1 The framework flow chart of the full-process intelligent disposal management method based on alarm signals.

[0032] Figure 2 Computer equipment diagram of the full-process intelligent disposal management method based on alarm signals. DETAILED DESCRIPTION

[0033] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0034] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0035] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0036] Example 1

[0037] Reference Figure 1-2 , which is the first embodiment of the present invention, and provides a full-process intelligent disposal management method based on alarm signals, including:

[0038] Furthermore, this embodiment also provides a full-process intelligent disposal management system based on alarm signals, including:

[0039] S1: Use Kettle data processing tool to monitor the transmission folder in real time. Once the alarm signal data file transmitted by the OCS system is found, it will be imported into the database.

[0040] Specifically, the Kettle data processing tool is used to detect the file transfer folder once every second. Once a file is found, the alarm signal data file ending with .dt transmitted by the OCS system is entered into the database.

[0041] S2: Execute the cleaning procedure on the incoming data to extract the core information of the alarm signal.

[0042] It should be noted that the data stored in the database is cleaned through a data cleaning program to clean the alarm signal into data of the corresponding format. The format after cleaning is: substation + interval + occurrence time + alarm signal content.

[0043] Specifically, when executing the cleaning procedure for the incoming data, an adaptive data cleaning mechanism based on semantic rules is introduced; based on the domain ontology and background knowledge, a set of cleaning semantic rule sets R is constructed; for each incoming data D, semantic parsing is performed on it based on the cleaning semantic rule set R to obtain a semantic parse tree T; a scoring function Score(T,R) is defined to measure the score of the semantic parse tree T under the cleaning semantic rule R, and a threshold parameter θ is set; if Score(T,R)>θ, data cleaning is performed according to the semantic structure of T; if Score(T,R)≤θ, the uncertain part of the semantic parsing is fed back, and the semantic rule R is manually reviewed and corrected; based on the manual correction, the scoring function NewScore(T,R') is reconstructed; and the data D is re-cleaned based on the latest scoring function NewScore(T,R').

[0044] It should be noted that the design formula of the scoring function Score(T,R) is as follows:

[0045] Score(T,R)=λ1StructureSim(T,R)+λ2ConceptSim(T,R)

[0046] +λ3ContextSim(T,R)+λ4RuleCoverageSim(T,R)

[0047]

[0048] Among them, StructureSim(T,R) represents the similarity score between the semantic parse tree T and the rule R at the structural level, ConceptSim(T,R) represents the similarity score between the semantic parse tree T and the rule R at the conceptual level, ContextSim(T,R) represents the similarity score between the semantic parse tree T and the rule R at the context level, RuleCoverageSim(T,R) represents the coverage of the semantic parse tree T by the rule R, λ1, λ2, λ3, λ4 represent the weight parameters of each score (obtained by tuning the validation set data), NT represents the node set in the semantic parse tree T, n represents the node, MLP represents the multi-layer perceptron network, σ represents the sigmoid function, E T Represents the edge set in the semantic parse tree T, e represents the edge, CosineSim() represents the calculation of the cosine similarity of two vectors, e t1 and e t2 represents the entities at both ends of the edge, Embed represents the entity vector representation (mapped to the vector representation in the low-dimensional dense vector space through the embedding technology), P T represents the path set in the semantic parse tree T, p represents the path, BLEU represents the commonly used evaluation index in natural language generation tasks, and p desc represents the path description, GenDesc represents the context generation model, and E R Represents the edge set in rule R.

[0049] It should be noted that in ConceptSim(T,R) and RuleCoverageSim(T,R) T Although the meaning is the same, the specific calculation method is different; in ConceptSim (T, R), E T As a set of conceptual edges, for each edge e, the embedding similarity of the entities at both ends of the edge is calculated; in RuleCoverageSim(T,R), E T is regarded as the set of all edges that need to be covered by rule R, through the edge set E in rule R R The intersection is taken to measure the coverage.

[0050] Furthermore, the specific formula of reconstructing the scoring function NewScore(T,R') is as follows:

[0051] NewScore(T,R')=αScore(T,R')+(1-α)SimWord(T,D)

[0052] Among them, SimWord represents the bag-of-words similarity, α represents the coupling parameter, Score(T,R') represents the semantic matching degree score of the semantic parse tree T under the manually corrected new rule R', R' represents the manually corrected new rule, T represents the semantic parse tree, and D represents the stored data.

[0053] S3: Build and continuously update the expert knowledge base, match the extracted core information in the expert knowledge base, and generate an expert knowledge model accordingly.

[0054] Specifically, the following steps are included:

[0055] S3.1: Based on the existing manually annotated historical alarm data, use natural language processing and machine learning technology to automatically extract the core information of the alarm signal.

[0056] Preferably, the core information of the alarm signal includes the alarm signal content, alarm level, abnormality classification, cause, possible consequences, and handling steps.

[0057] S3.2: Integrate multi-source heterogeneous data from monitoring systems, working condition databases, and equipment archives into the expert knowledge base to provide comprehensive contextual information for anomaly analysis.

[0058] It should be noted that this expert knowledge base is provided by many grid dispatching experts from the Southern Power Grid. This expert database records in detail the abnormal signal classification, causes, possible consequences, push methods, handling steps, handling requirements for each step, and output information for each step corresponding to different alarm signals.

[0059] Specifically, the fusion of multi-source heterogeneous data from the monitoring system, working condition database, and equipment archives into the expert knowledge base includes the following steps: constructing an overall framework G including three knowledge graphs G1, G2, and G3 of monitoring, working condition, and equipment; i The knowledge graph embedding model is trained separately to obtain the embedding representation vector v i ; Define the knowledge graph node alignment scoring function Align(v i ,v j )=δ(W·Concat(v i ,v j )), where δ represents a nonlinear function, W represents a weight matrix, Concat(v i ,v j ) represents the two vectors v i and v j For splicing, v i and v j Represents two different knowledge graphs G i and G j Node embedding vector; maximize the global objective function of alignment score on all node pairs (i, j) and optimize the alignment matrix M; under the optimized alignment matrix M, perform feature exchange of each sub-graph to achieve data fusion.

[0060] It should be noted that the specific formula of the global objective function of the alignment score is as follows:

[0061] max∑Align(v i ,v j )·M ij

[0062] st∑Mij =1,M ij ≥0

[0063] Where M represents the alignment matrix, i and j represent row and column variables respectively, and Align(v i ,v j ) represents the knowledge graph node alignment scoring function, v i and v j Represents two different knowledge graphs G i and G j The node embedding vector of .

[0064] S3.3: Model the expert knowledge base as a knowledge graph for the power grid domain, effectively organize entities and their relationships, and support richer queries and reasoning.

[0065] S3.4: Use graph neural networks to automatically learn historical abnormal patterns on the knowledge graph, accurately match new alarm signals, and generate corresponding expert handling models.

[0066] S3.5: If the new abnormal alarm signal does not exist in the expert knowledge base, the expert database information is manually entered and matched again.

[0067] S3.6: For the matched expert treatment models, experts shall conduct manual review and feedback correction, and improve the expert knowledge base through continuous human-machine collaborative iterative optimization.

[0068] S3.7: Combine the optimized expert handling model with real-time monitoring data to dynamically generate personalized and explainable exception handling solutions.

[0069] S4: Based on the generated expert knowledge model, the handling steps of the alarm signal are automatically split into stages to generate sub-processes.

[0070] Specifically, the following steps are included:

[0071] S4.1: Utilize the rich semantic information contained in the knowledge graph and dynamically plan the optimal disposal path based on the contextual association of the alarm signal.

[0072] S4.2: Model the disposal path as an intelligent workflow and introduce process intelligence technology to support adaptive adjustment, parallel execution and exception handling of the process.

[0073] S4.3: Treat the disposal process as a sequential decision-making problem and use reinforcement learning technology to automatically learn and optimize the optimal decomposition path based on a large number of historical cases.

[0074] S4.4: While splitting internal processes based on the expert knowledge base, orchestrate and integrate cross-system processes to enable the execution of disposal plans to cover a wider range of scenarios.

[0075] S4.5: For special and complex abnormal situations, experts are allowed to intervene manually and review, modify and supplement the pre-generated processes through visual modeling tools to generate customized disposal process plans.

[0076] Preferably, when experts are allowed to intervene manually to modify the pre-generated process, the process generation strategy can be optimized through reinforcement learning. The specific approach is as follows: design the state s as the currently generated process sequence, the action a as the node insertion / modification operation, and define the reward function r(s,a) as the process quality score after the new node is inserted; under the reinforcement learning framework, the state s, action a, and reward r(s,a) construct a Markov decision process to find the optimal generation strategy π * =argmax∑r(s,a); Based on the existing manual processing data, the supervised learning algorithm is used to * Perform pre-training to build a data set of state-action pairs; in the human-computer interaction mode, receive the expert's operation action a and feedback reward r in real time, and continuously adjust and optimize the strategy π through the policy iteration algorithm * ; Build a model interpretability mechanism to explain the reasons for each step of the operation. If the expert does not understand or objects to a certain step, feedback correction will be provided through the model interpretability mechanism; For the final strategy of optimization iteration π * , coded into a process generation model and integrated into a visual modeling tool to assist experts in process review and customization.

[0077] Preferably, the specific formula of the reward function r(s,a) is as follows:

[0078] r(s,a)=w1SimStructure(s',s)+w2SimSemantic(s',s)+w3CostScore(s')

[0079] Among them, s' represents the newly generated process after executing action a, s represents the reference high-quality process, SimStructure and SimSemantic are used to measure the similarity between the newly generated process s' and the reference high-quality process s at the structural and semantic levels, respectively, and w1, w2, and w3 represent the corresponding weights, respectively.

[0080] Specifically, SimStructure(s',s) represents structural similarity measurement: the process is regarded as a directed acyclic graph, with nodes representing tasks and edges representing control flow transfers; the topological structural features of s' and s are compared, such as the number of nodes, the number of edges, the average path length, etc.; the kernel function or structured model is used to obtain the kernel representation of the two flow charts; the value of the structural similarity SimStructure(s',s) is calculated through the kernel representation, and the higher the similarity, the higher the score. SimSemantic(s',s) represents semantic similarity measurement: the task of each node is represented by a natural language description text, and the semantic vector representation of the node text is calculated using a pre-trained language model such as BERT; semantically similar node pairs are found in s' and s, and the semantic similarity of the node pairs is calculated; the similarity scores of all node pairs are summarized to obtain the overall semantic similarity SimSemantic(s',s). The reward function r(s,a) is obtained by integrating the two dimensions of structural similarity and semantic similarity in a linear weighted manner. The reward function r(s,a) constructed in this way comprehensively considers the topological rationality, semantic rationality and execution cost of the process, and can guide the reinforcement learning model to better optimize the generation strategy π.

[0081] S4.6: For new knowledge and new practices generated during the execution process, they are automatically summarized and fed back to the expert knowledge base through knowledge extraction technology to continuously optimize and enrich the disposal process templates.

[0082] S5: Determine whether the automatic disposal criteria are met based on the matching results. If so, it is automatically executed. If not, it is executed after confirmation by on-site personnel.

[0083] Specifically, the input data of various dimensions required for executing automated criteria are collected, including alarm signal content, level, impact scope, risk assessment, and historical cases; a multi-level threat assessment model is constructed based on machine learning algorithms and historical data to optimize the weights and thresholds of each criterion condition; the optimized multi-dimensional criterion decision model TS(ω * ), organized into an efficient and executable criterion rule set through decision tree or rule engine technology; for a new alarm signal sample x, the criterion rule set is executed on the rule engine to calculate TS(x; ω * ), and obtain the decision result of whether it meets the requirements of automated disposal; for special and complex situations that cannot be fully covered by the rule engine, start the human-computer interaction interface for experts to manually judge and intervene in the decision-making process; according to the decision result, approve whether to execute the automated process disposal or require on-site personnel to intervene in manual disposal; for the cases of judgment errors in the actual execution process, summarize and feedback to the experts for review, and update and optimize the multi-dimensional judgment decision model TS(ω) online *) Establish a statistical analysis mechanism for criterion quality, continuously track key indicators such as accuracy and recall rate of the criterion, and regularly generate analysis reports to provide a basis for continuous improvement.

[0084] It should be noted that the construction of a multi-level threat assessment model includes the following steps: establishing a threat assessment model based on the alarm signal content NR, the alarm level LE, the impact range YX, and the risk assessment FX NR 、f LE 、f YX 、f FX ; For each model f NR 、f LE 、f YX 、f FX Based on historical data, four threat assessment models f are trained using the mean square error loss function NR 、f LE 、f YX 、f FX The model parameter μ NR , μ LE , μ YX , μ FX ; Four threat assessment models f NR 、f LE 、f YX 、f FX The output of is weighted linearly combined to construct a comprehensive threat score function TS; on the training data, the optimal value of each weight is estimated through cross-validation and grid search. * ; For a new test sample x, calculate the comprehensive threat level TS(x;ω * ); if TS(x;ω * )<τ, τ is a pre-set threshold, then it is determined that x meets the automatic processing conditions, otherwise it does not meet the conditions and the human-computer interaction judgment machine needs to be started.

[0085] Specifically, the specific formula of the mean square error loss function is as follows:

[0086] L(μ)=∑(f(x;μ)-y) 2

[0087] Among them, x represents the model input, y represents the true label, and μ represents the parameter to be optimized.

[0088] Specifically, the specific formula of the comprehensive threat degree scoring function TS is as follows:

[0089] TS(UR,LE,YX,FX)=ω NR f NR (NR; μ NR )+ω LE f LE (LE; μLE )

[0090] +ω YX f YX (YX; μ YX )+ω FX f FX (FX; μ FX )

[0091] Among them, UR, LE, YX, and FX represent the alarm signal content, alarm level, impact range, and risk assessment, respectively. NR 、f LE 、f YX 、f FX They represent the threat assessment model based on the alarm signal content UR, alarm level LE, impact range YX, and risk assessment FX, respectively. UR ,ω LE ,ω YX ,ω FX Represents the weight of each model, μ NR , μ LE , μ YX , μ FX They represent the threat assessment model f NR 、f LE 、f YX 、f FX The model parameters.

[0092] S6: If there are defects in the prediction results, the CART classification tree algorithm is used to predict potential defects based on historical alarm signal data and disposal results.

[0093] It should be noted that the definition of a defective prediction result is that unexpected situations or unsatisfactory results occur during the actual execution of the automated disposal.

[0094] Furthermore, based on historical alarm signal data and actual disposal result data, the CART classification tree algorithm is used to train the defect prediction model to identify different types of potential disposal defects. For alarm signals that require manual disposal in S5, they are input into the trained defect prediction model to automatically analyze and diagnose potential disposal defect types and risks. The predicted potential defect information is presented in a visual form on the human-computer interaction interface to provide reference and warning for the actual disposal of on-site personnel. After the actual disposal is completed, the on-site personnel review the actual defect situation and feed back the review results to the system for continuous optimization and iteration of the defect model.

[0095] The present embodiment also provides a full-process intelligent disposal management system based on alarm signals, including a transmission detection module, which is used to use the Kettle data processing tool to monitor the transmission folder in real time, and once the alarm signal data file transmitted by the OCS system is found, it is imported into the database; a cleaning execution module, which is used to execute the cleaning program on the input data and extract the core information of the alarm signal; a matching generation module, which is used to build and continuously update the expert knowledge base, match the extracted core information in the expert knowledge base, and generate an expert knowledge model accordingly; a process splitting module, which is used to automatically split the disposal steps of the alarm signal into sub-processes according to the generated expert knowledge model; a judgment module, which is used to judge whether the automatic disposal criteria are met according to the matching results, and if so, it is automatically executed, and if not, it is executed after confirmation by the on-site personnel; a defect detection module, which is used to use the CART classification tree algorithm to predict potential defects based on historical alarm signal data and disposal results, and provide a decision-making basis for maintenance.

[0096] This embodiment also provides a computer device, which is suitable for the full-process intelligent disposal management method based on alarm signals, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the full-process intelligent disposal management method based on alarm signals as proposed in the above embodiment.

[0097] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0098] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for implementing a full-process intelligent disposal management method based on alarm signals as proposed in the above embodiment is implemented.

[0099] In summary, the present invention realizes real-time automatic perception and monitoring of alarm signal data files transmitted by the power grid monitoring system, thereby improving the timeliness and accuracy of abnormal monitoring; constructs a knowledge graph for the power grid field based on multi-source heterogeneous data fusion, automatically learns historical abnormal patterns, generates accurate expert handling knowledge models, and provides efficient and explainable abnormal handling decision support; transforms the expert handling model into an intelligent workflow, introduces a process intelligent management mechanism, supports adaptive adjustment of the process, parallel execution and abnormal handling, and improves the pertinence and executability of the handling plan; establishes a human-machine collaborative optimization mechanism, continuously optimizes and enriches the expert knowledge base, iteratively generates customized handling processes, and continuously improves the accuracy and high level of abnormal handling decisions.

[0100] Example 2

[0101] Reference Figure 1-2 , which is the second embodiment of the present invention, provides a full-process intelligent disposal management method based on alarm signals. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0102] Specifically, during the real-time monitoring and data import stage, the Kettle monitoring tool detects the OCS system transmission folder in real time and finds a new file "#0011 Main Transformer#2023-04-0214:05# Temperature rises too fast#", and imports the .dt file into the database.

[0103] Furthermore, a cleaning procedure is performed on the incoming data. A semantic rule set R is constructed, containing 100 rules; the data is semantically parsed to obtain a parsing tree T; the score function value is calculated to be 0.88, which is higher than the set threshold of 0.85, and the data is cleaned according to the T structure; the format after cleaning is: Nanshan Substation + Main Transformer

[0104] #0011+2023-04-0214:05+The temperature rises too quickly.

[0105] Furthermore, based on the core information "temperature rise too fast" matching in the knowledge base, the matching template {abnormal classification: "equipment overheating", treatment steps: [...]} is hit; context information is fused from the monitoring / working condition / equipment database, and the knowledge graph alignment objective function value after fusion is 0.92. The matching treatment step path is modeled as a workflow, the design state s is the generated process, and the action a is the editing operation; based on reinforcement learning, sub-processes are continuously generated / modified, and finally a branch process solution is generated: Sub-process 1: ["monitor temperature", "judge criticality"], Sub-process 2: ["on-site inspection", "find hot spots"], Sub-process 3: ["assess risk", "decide the next step"].

[0106] Furthermore, based on the model calculation, the comprehensive threat score TS is 0.78, which is lower than the threshold of 0.80. This case does not meet the automation conditions, so the human-computer interaction interface is started and the case is handed over to the on-duty personnel for disposal. The alarm signal is input into the CART defect prediction model, which predicts the possible existence of "on-site inspection omission" defects and gives a visual warning. After the on-duty personnel complete the on-site disposal, they feedback the existence of the above defects and use the feedback results to update the defect prediction model.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A full-process intelligent disposal management method based on alarm signals, characterized in that: include, Use Kettle data processing tool to monitor the transmission folder in real time. Once the alarm signal data file transmitted by the OCS system is found, it will be imported into the database. Perform cleaning procedures on incoming data to extract the core information of alarm signals; Build and continuously update the expert knowledge base, match the extracted core information in the expert knowledge base, and generate the expert knowledge model accordingly; According to the generated expert knowledge model, the handling steps of the alarm signal are automatically split into sub-processes by stage; Determine whether the automatic disposal criteria are met based on the matching results. If so, the process is automatically executed. If not, the process is executed after confirmation by on-site personnel. The CART classification tree algorithm is used to predict potential defects based on historical alarm signal data and disposal results, providing a decision-making basis for maintenance.

2. The full-process intelligent disposal management method based on alarm signals according to claim 1, characterized in that: The cleaning procedure for the incoming data includes the following steps: When performing the cleaning procedure on the incoming data, an adaptive data cleaning mechanism based on semantic rules is introduced; Based on domain ontology and background knowledge, a set of cleaning semantic rule sets R is constructed; For each piece of incoming data D, perform semantic analysis based on the cleaning semantic rule set R to obtain a semantic analysis tree T; Define a scoring function Score(T,R) to measure the score of the semantic parse tree T under the cleaning semantic rule R, and set the threshold parameter θ; If Score(T,R)>θ, data cleaning is performed according to the semantic structure of T; If Score(T,R)≤θ, the uncertain part of the semantic analysis is fed back, and the semantic rule R is corrected by manual review; Based on manual correction, the scoring function NewScore(T,R') is reconstructed; The incoming data D is cleaned again based on the latest scoring function NewScore(T,R').

3. The full-process intelligent disposal management method based on alarm signals according to claim 2, characterized in that: The specific formula of the scoring function Score(T,R) is as follows: Score(T,R)=λ1StructureSim(T,R)+λ2ConceptSim(T,R)+λ3ContextSim(T,R)+λ4RuleCoverageSim(T,R) Among them, StructureSim(T,R) represents the similarity score between the semantic parse tree T and the rule R at the structural level, and ConceptSim(T,R) represents the similarity score between the semantic parse tree T and the rule R at the conceptual level. ContextSim(T,R) represents the similarity score between the semantic parse tree T and the rule R at the context level, RuleCoverageSim(T,R) represents the coverage of the semantic parse tree T by the rule R, λ1, λ2, λ3, λ4 represent the weight parameters of each score, N T represents the node set in the semantic parse tree T, n represents the node, MLP represents the multi-layer perceptron network, σ represents the sigmoid function, E T Represents the edge set in the semantic parse tree T, e represents the edge, CosineSim() represents the calculation of the cosine similarity of two vectors, e t1 and e t2 represents the entities at both ends of the edge, Embed represents the entity vector representation (mapped to the vector representation in the low-dimensional dense vector space through the embedding technology), P T represents the path set in the semantic parse tree T, p represents the path, BLEU represents the commonly used evaluation index in natural language generation tasks, and p desc represents the path description, GenDesc represents the context generation model, E R Represents the edge set in rule R; The specific formula for reconstructing the scoring function NewScore(T,R') is as follows: NewScore(T,R')=αScore(T,R')+(1-α)SimWord(T,D) Among them, SimWord represents the bag-of-words similarity, α represents the coupling parameter, Score(T,R') represents the semantic matching degree score of the semantic parse tree T under the manually corrected new rule R', R' represents the manually corrected new rule, T represents the semantic parse tree, and D represents the stored data.

4. The full-process intelligent disposal management method based on alarm signals according to claim 1, characterized in that: The construction and continuous updating of the expert knowledge base comprises the following steps: Based on the existing manually annotated historical alarm data, the core information of the alarm signal is automatically extracted using natural language processing and machine learning technology; Integrate multi-source heterogeneous data from monitoring systems, working condition databases, and equipment archives into the expert knowledge base to provide comprehensive contextual information for abnormal analysis; Model the expert knowledge base as a knowledge graph for the power grid domain, effectively organize entities and their relationships, and support richer queries and reasoning; Using graph neural networks, we can automatically learn historical abnormal patterns on the knowledge graph, accurately match new alarm signals, and generate corresponding expert handling models. If the new abnormal alarm signal does not exist in the expert knowledge base, the expert database information is manually entered and matched again; For the expert treatment models generated by matching, experts will conduct manual review and feedback correction, and improve the expert knowledge base through continuous human-machine collaborative iterative optimization; The optimized expert handling model is combined with real-time monitoring data to dynamically generate personalized and explainable exception handling solutions.

5. The full-process intelligent disposal management method based on alarm signals according to claim 4, characterized in that: The process of fusing multi-source heterogeneous data from the monitoring system, the working condition database, and the equipment archives into the expert knowledge base includes the following steps: Construct an overall framework G including three knowledge graphs G1, G2, and G3 of monitoring, working conditions, and equipment; In each sub-graph G i The knowledge graph embedding model is trained separately to obtain the embedding representation vector v i ; Define the knowledge graph node alignment scoring function Align(v i ,v j )=δ(W·Concat(v i ,v j )), where δ represents a nonlinear function, W represents a weight matrix, Concat(v i ,v j ) represents the two vectors v i and v j For splicing, v i and v j Represents two different knowledge graphs G i and G j The node embedding vector of Maximize the global objective function of alignment score on all node pairs (i, j) and optimize the alignment matrix M; Under the optimized alignment matrix M, the features of each sub-graph are exchanged to achieve data fusion; The specific formula of the global objective function of the alignment score is as follows: max∑Align(v i ,in j )·M ij s.t.∑M ij =1,M ij ≥0 Where M represents the alignment matrix, i and j represent row and column variables respectively, and Align(v i ,v j ) represents the knowledge graph node alignment scoring function, v i and v j Represents two different knowledge graphs G i and G j The node embedding vector of .

6. The full-process intelligent disposal management method based on alarm signals according to claim 1, characterized in that: The process of automatically splitting the alarm signal handling steps into sub-processes according to stages includes the following steps: Utilize the rich semantic information contained in the knowledge graph and dynamically plan the optimal handling path based on the context of the alarm signal; Model the disposal path as an intelligent workflow, introduce process intelligence technology, and support adaptive adjustment, parallel execution, and exception handling of the process; The disposal process is regarded as a sequential decision-making problem, and reinforcement learning technology is used to automatically learn and optimize the best decomposition path based on a large number of historical cases; While splitting internal processes based on the expert knowledge base, orchestration and integration of cross-system processes are performed to enable the execution of disposal plans to cover a wider range of scenarios; For special and complex abnormal situations, experts are allowed to intervene manually and review, modify and supplement the pre-generated processes through visual modeling tools to generate customized disposal process solutions; New knowledge and new practices generated during the execution process are automatically summarized and fed back to the expert knowledge base through knowledge extraction technology, continuously optimizing and enriching the disposal process templates.

7. The full-process intelligent disposal management method based on alarm signals according to claim 6, characterized in that: The generation of a customized disposal process plan includes the following steps: The design state s is the currently generated process sequence, the action a is the node insertion / modification operation, and the reward function r(s,a) is defined as the process quality score after the new node is inserted; In the reinforcement learning framework, the state s, action a, and reward r(s,a) construct a Markov decision process to find the optimal generation strategy π * =argmax∑r(s,a); Based on the existing manual processing data, the strategy π is trained using a supervised learning algorithm. * Perform pre-training to build a data set of state-action pairs; In the human-computer interaction mode, the expert operation action a and feedback reward r are received in real time, and the optimization strategy π is continuously adjusted through the strategy iteration algorithm. * ; Build a model interpretability mechanism to explain the reasons for each step of the operation. If the expert does not understand or objects to a certain step, feedback and correction will be provided through the model interpretability mechanism. For the final strategy π of the optimization iteration * ,coded as a process generation model and integrated into a visual modeling tool to assist experts in process review and customization; The specific formula of the reward function r(s,a) is as follows: r(s,a)=w1SimStructure(s',s)+w2SimSemantic(s',s)+w3CostScore(s') Among them, s' represents the newly generated process after executing action a, s represents the reference high-quality process, SimStructure and SimSemantic are used to measure the similarity between the newly generated process s' and the reference high-quality process s at the structural and semantic levels, respectively, and w1, w2, and w3 represent the corresponding weights, respectively.

8. A full-process intelligent disposal management system based on alarm signals, based on the full-process intelligent disposal management method based on alarm signals according to any one of claims 1 to 7, characterized in that: Also includes, The transmission detection module is used to monitor the transmission folder in real time using the Kettle data processing tool. Once the alarm signal data file transmitted by the OCS system is found, it will be imported into the database; The cleaning execution module is used to execute the cleaning procedure on the incoming data and extract the core information of the alarm signal; The matching generation module is used to build and continuously update the expert knowledge base, match the extracted core information in the expert knowledge base, and generate the expert knowledge model accordingly; The process splitting module is used to automatically split the handling steps of the alarm signal into sub-processes according to the generated expert knowledge model; The judgment module is used to judge whether the automatic disposal criteria are met according to the matching results. If so, it is automatically executed; if not, it is executed after confirmation by the on-site personnel; The defect detection module uses the CART classification tree algorithm to predict potential defects based on historical alarm signal data and disposal results, providing a decision-making basis for maintenance.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the full-process intelligent disposal management method based on alarm signals described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the full-process intelligent disposal management method based on alarm signals described in any one of claims 1 to 7 are implemented.

Images