Asset operation and maintenance decision management platform and management method based on data fusion

Abstract

The application relates to the technical field of operation and maintenance management, in particular to an asset operation and maintenance decision management platform and method based on data fusion, which comprises the following steps: collecting asset life cycle data and constructing a standardized asset data set; using ontology modeling and knowledge extraction technology to construct a dynamic asset knowledge graph containing asset attributes, topological relationships and operating states; training a deep learning model to realize asset fault diagnosis, constructing an asset operation and maintenance risk detection algorithm, performing real-time analysis on asset multidimensional data, identifying asset safety risk levels, and triggering corresponding operation and maintenance strategies; using natural language generation technology to convert the diagnosed asset faults and generated operation and maintenance strategies into natural language and push them to operation and maintenance personnel for execution. The method fuses multi-source heterogeneous data and realizes intelligent operation and maintenance decision of the asset life cycle.

Description

Technical Field

[0001] This application relates to the field of operation and maintenance management technology, specifically to an asset operation and maintenance decision management platform and management method based on data fusion. Background Technology

[0002] Traditional IT operations and maintenance (O&M) rely heavily on human experience and manual operation, resulting in inefficiency, slow response times, and a high risk of errors. O&M personnel need to log into different management systems to query asset information, monitor performance metrics, check fault alarms, and then make decisions and take O&M measures based on experience. This manual O&M model struggles to cope with the ever-increasing scale and complexity of IT assets, and cannot achieve rapid and accurate fault diagnosis and risk prediction.

[0003] Furthermore, the entire lifecycle management of IT assets involves various stages such as asset procurement, deployment, configuration, monitoring, maintenance, optimization, and decommissioning, generating massive amounts of multi-source heterogeneous data. However, this data often lacks correlation and integration, making it difficult to form a comprehensive asset view and knowledge base. The fragmentation of multi-source heterogeneous data leads to a lack of data support for IT operations and maintenance decisions, hindering the realization of intelligent lifecycle management of assets.

[0004] In view of this, this application proposes an asset operation and maintenance decision management platform and management method based on data fusion. Summary of the Invention

[0005] To achieve the above objectives, this application provides an asset operation and maintenance decision management platform and management method based on data fusion, the specific technical solution of which is as follows:

[0006] Data fusion-based asset operation and maintenance decision management methods include:

[0007] Through a unified operation and maintenance platform, multi-source heterogeneous data throughout the entire asset lifecycle are collected, and the data is cleaned, transformed, and integrated to form a standardized asset dataset.

[0008] By utilizing ontology modeling and knowledge extraction techniques, a dynamically updated asset knowledge graph is constructed based on the fused asset dataset, including multi-dimensional information such as asset attributes, topological relationships, and operational status.

[0009] Train an asset fault diagnosis model based on a deep learning model, learn asset fault patterns from an asset knowledge graph, and the asset fault diagnosis model detects whether the current asset has a fault based on the asset operation and maintenance strategy.

[0010] An asset operation and maintenance risk detection algorithm is constructed. Through real-time analysis of multi-dimensional asset data, the asset security risk coefficient is identified and the asset security risk level is determined. Based on the asset security risk level, the corresponding asset operation and maintenance strategy is triggered.

[0011] By using natural language generation technology, asset faults and asset operation and maintenance strategies identified based on knowledge graphs are converted into natural language and pushed to operation and maintenance management personnel, who then implement the operation and maintenance measures.

[0012] Preferably, the assets include: switches, routers, network security devices, network terminals, data center proxy servers, virtual machines, operating systems, databases, middleware, containers, IP addresses, and domain name information;

[0013] Collect various parameter information of the assets, including: static attribute data of the assets, dynamic monitoring data of the assets, and operation and maintenance data of the assets;

[0014] Data preprocessing is performed on the collected data, including: missing value handling, noise removal, outlier detection, and data normalization.

[0015] All collected data are aggregated, and the JS divergence between data points is used as the data priority sorting method to resolve the problem of data conflicts for the same indicator between multiple data sources.

[0016] Preferably, the asset ontology is modeled to construct an ontology model covering the entire lifecycle management of assets, including: asset concepts, asset relationships, and asset attributes, and the asset attribute values ​​are associated with asset entities; the asset concepts, asset relationships, and asset attributes are formally defined using an ontology description language to construct the asset ontology model;

[0017] Based on the constructed IT asset ontology model, structured asset knowledge is extracted from multi-source heterogeneous datasets, including entity recognition, relation extraction, attribute association, and ontology matching.

[0018] An asset knowledge graph is constructed through steps including knowledge representation learning, knowledge fusion, knowledge graph storage, and knowledge graph updating.

[0019] Preferably, the constructed asset knowledge graph is converted into a low-dimensional dense vector representation, which is used as the input to the fault diagnosis model. At the same time, the collected asset dynamic data is also used as the input to the fault diagnosis model. The knowledge graph representation learning algorithm of the TransE model is used to learn the embedding vector of each asset node and relation edge.

[0020] A fault diagnosis model based on the Graph Attention Network (GAT) model and the Long Short-Term Memory (LSTM) model is constructed. The fault diagnosis model includes an asset embedding layer, a graph attention layer, an LSTM layer, and a fault classification layer.

[0021] Preferably, based on the asset risk assessment results and historical operation and maintenance records, the fault types of the assets are labeled to form training samples for training the fault diagnosis model;

[0022] The fault diagnosis model is trained through end-to-end supervised learning using training samples to learn the model parameters. When the fault diagnosis model is trained, the trained fault diagnosis model is used to predict the fault type and output the fault diagnosis result based on the asset risk level and the predicted fault type.

[0023] Preferably, multi-dimensional data analysis is performed on the asset data to obtain static asset data, dynamic asset data, and asset knowledge graph data; an asset relationship adjacency matrix is ​​constructed based on the asset knowledge graph data.

[0024] By combining features from static and dynamic asset data, an asset security risk assessment feature matrix is ​​constructed.

[0025] Based on the asset relationship adjacency matrix, a random walk algorithm is used to calculate the correlation risk coefficient between assets.

[0026] Preferably, an asset operation and maintenance risk detection algorithm is constructed based on the correlation risk coefficient between assets to calculate the comprehensive security risk score of the assets;

[0027] The comprehensive safety risk score is normalized to obtain the risk probability value of the asset.

[0028] Based on the comprehensive risk probability value of the assets, a risk level classification threshold is set, and the assets are divided into four risk levels: low risk, medium risk, high risk, and severe risk. Different operation and maintenance strategies are assigned to each risk level.

[0029] Preferably, the asset fault diagnosis results are semantically parsed and represented as fault knowledge vectors; the asset risk assessment results are semantically parsed and represented as risk knowledge vectors; faulty assets and risky assets are queried from the asset knowledge graph, and the query results are represented by comprehensive asset knowledge vectors.

[0030] By using a pre-trained language model, with fault knowledge vectors, risk knowledge vectors, and comprehensive asset knowledge vectors as inputs, the system outputs textual solutions for operation and maintenance decisions.

[0031] Preferably, the written decision plan is pushed to the operation and maintenance management personnel, who can then edit and modify the plan through a human-computer interaction interface and implement operation and maintenance measures on the assets.

[0032] The asset operation and maintenance decision management platform based on data fusion is used to implement the asset operation and maintenance decision management method based on data fusion, including: a data acquisition module, a knowledge graph module, an asset fault diagnosis module, an operation and maintenance risk detection module, and an operation and maintenance execution module.

[0033] The data acquisition module collects multi-source heterogeneous data throughout the entire asset lifecycle through a unified operation and maintenance platform, and performs data cleaning, transformation and fusion to form a standardized asset dataset.

[0034] The knowledge graph module utilizes ontology modeling and knowledge extraction techniques to construct a dynamically updated asset knowledge graph based on the fused asset dataset, including multi-dimensional information such as asset attributes, topological relationships, and operational status.

[0035] The asset fault diagnosis module is used to train an asset fault diagnosis model based on a deep learning model, learn asset fault patterns from an asset knowledge graph, and detect whether the current asset has a fault according to the asset operation and maintenance strategy.

[0036] The operation and maintenance risk detection module is used to construct an asset operation and maintenance risk detection algorithm. Through real-time analysis of multi-dimensional asset data, it identifies the asset security risk coefficient, determines the asset security risk level, and triggers the corresponding asset operation and maintenance strategy according to the asset security risk level.

[0037] The operation and maintenance execution module uses natural language generation technology to convert asset faults and asset operation and maintenance strategies identified based on knowledge graphs into natural language and pushes them to operation and maintenance management personnel, who then execute the operation and maintenance measures.

[0038] The beneficial effects of this application are as follows: By collecting, cleaning, transforming and integrating data throughout the entire lifecycle of assets, this application establishes a standardized and high-quality asset dataset, providing a reliable data foundation for subsequent intelligent operation and maintenance.

[0039] This application constructs a dynamic knowledge graph containing multi-dimensional asset information, forming a global view of assets, and providing a comprehensive, accurate, and real-time asset semantic knowledge base for intelligent decision-making.

[0040] This application utilizes knowledge graph and deep learning technologies to train an asset fault diagnosis model, which can automatically, quickly, and accurately detect asset faults based on asset status and historical fault patterns.

[0041] This application constructs an asset operation and maintenance risk detection algorithm, which continuously assesses the asset security risk level by analyzing multi-dimensional asset data in real time, and realizes the proactive identification and early warning of operation and maintenance risks.

[0042] This application utilizes natural language generation technology to automatically generate highly readable fault diagnosis and operation and maintenance strategy reports, and pushes them to operation and maintenance personnel, thereby improving operation and maintenance efficiency and executability. Attached Figure Description

[0043] Figure 1 A flowchart of the asset operation and maintenance decision management method based on data fusion provided for this application;

[0044] Figure 2 A flowchart of the asset fault diagnosis method based on the data fusion-based asset operation and maintenance decision management method provided in this application;

[0045] Figure 3 A flowchart of the operation and maintenance risk assessment method for the asset operation and maintenance decision management method based on data fusion provided in this application;

[0046] Figure 4 The operation and maintenance decision generation method process of the asset operation and maintenance decision management method based on data fusion provided in this application;

[0047] Figure 5 The structural diagram of the asset operation and maintenance decision management platform based on data fusion provided in this application. Detailed Implementation

[0048] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0049] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0050] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of this application. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that excludes other embodiments.

[0051] Example 1

[0052] Reference Figures 1 to 4 This is the first embodiment of the present application, which provides an asset operation and maintenance decision management method based on data fusion.

[0053] Step 1: Collect multi-source heterogeneous data throughout the entire asset lifecycle through a unified operation and maintenance platform, and perform data cleaning, transformation and fusion to form a standardized asset dataset.

[0054] Specifically, the assets include: switches, routers, network security equipment, network terminals, data center proxy servers, virtual machines, operating systems, databases, middleware, containers, IP addresses, and domain name information.

[0055] By deploying various sensors, smart gateways, and other IoT devices on assets, various parameter information of the assets is collected, including: static asset attribute data, dynamic asset monitoring data, and asset operation and maintenance data. The static asset attribute data includes: asset ID, type, specifications, manufacturer, purchase time, and location information. The dynamic asset monitoring data includes: asset working status, performance indicators (e.g., efficiency, output, energy consumption), environmental parameters (e.g., temperature, humidity, vibration), and health metrics (e.g., failure rate, maintenance rate). The operation and maintenance data includes: asset start-up and shutdown records, parameter settings, maintenance, and defect handling operation and maintenance events. The data collection frequency can be set to real-time, second-level, minute-level, or hourly, depending on the characteristics of different asset types and monitoring parameters.

[0056] The collected heterogeneous data from multiple sources is transmitted through wired or wireless communication networks; at the same time, metadata information of the data is recorded, including data source, collection time and original format, to provide data traceability capability for subsequent data processing.

[0057] Because multi-source heterogeneous data may have missing values, outliers, and inconsistent data quality issues, data preprocessing is required, including: missing value handling, noise removal, outlier detection, and data normalization.

[0058] Missing value handling: Missing values ​​are estimated through record deletion, interpolation, and model prediction; Noise removal: High-frequency noise interference is removed using signal processing techniques such as Kalman filtering and wavelet transform; Outlier detection: Outliers are identified using statistical methods or machine learning algorithms such as distance-based clustering and density analysis; Data normalization: Data from different sources and in different formats are identified according to unified naming conventions and encoding rules to improve data interoperability.

[0059] By utilizing database technology and big data platforms, the cleaned and transformed standardized data is integrated, stored, and managed to form a database of asset subject domains. Through the key attributes of asset ID and timestamp, data of the same asset from different sources are linked to form a complete asset status record. Data of the same indicator from multiple sources is aggregated and merged to improve data quality and reliability. For conflicting data between multiple data sources, consistency rules are defined to resolve the issues, and the consistency rules are implemented using a priority sorting method.

[0060] In the data fusion process, the consistency rule for multi-source heterogeneous data is measured by calculating the JS divergence between data sources, and the formula is as follows:

[0061]

[0062] in, Represents probability distribution and JS divergence between It is a probability distribution and Uniformly mixed distribution, , For probability distribution Compared to The KL divergence between them is calculated using the following formula: , For parameters, It is a probability distribution The probability value in the i-th category, It is a mixed probability distribution Similarly, the probability value for the i-th category... Represents probability distribution Compared to KL divergence, Used to calculate distribution Compared to Information gain.

[0063] JS divergence measures the similarity between two probability distributions, with a value between 0 and 1. The smaller the JS divergence, the closer the distributions between the data sources are, and the higher the data consistency. By arranging the JS divergence values ​​in ascending order as a priority sequence, high-priority data indicators are selected as representative data for the same indicator, thus resolving data conflicts between multiple data sources.

[0064] Step 2: Using ontology modeling and knowledge extraction techniques, construct a dynamically updated asset knowledge graph based on the fused asset dataset, including multi-dimensional information such as asset attributes, topological relationships, and operational status.

[0065] Asset ontology is modeled. An ontology is a formalized and explicit specification of knowledge in a specific domain, composed of elements such as concepts, relationships, attributes, and axioms. For the IT asset operation and maintenance domain, an ontology model covering the entire asset lifecycle management is constructed, including: asset concepts, asset relationships, and asset attributes. The asset concepts are used to abstract IT assets of different granularities and types into a unified ontology concept, forming an asset classification system, including the concepts of network devices such as switches and routers, computing devices such as servers and virtual machines, and application concepts such as software, middleware, and databases. The asset relationships are used to define the semantic associations between assets, such as composition relationships, connection relationships, dependency relationships, and location relationships. The asset attributes are used to characterize the intrinsic characteristics of assets, including the asset's brand, model, configuration parameters, and health status.

[0066] Associate asset attribute values ​​with asset entities; use ontology description languages ​​(such as OWL, RDF) to formally define asset concepts, asset relationships and asset attributes, and build an IT asset ontology model.

[0067] Based on the constructed IT asset ontology model, structured asset knowledge is extracted from multi-source heterogeneous datasets, including entity recognition, relation extraction, attribute association, and ontology matching.

[0068] The entity recognition includes: using Named Entity Recognition (NER) technology to identify asset entities from unstructured text data (e.g., equipment logs and maintenance work orders) and mapping them to ontology concepts; where the NER task can be implemented based on machine learning models such as Conditional Random Fields (CRF) or Recurrent Neural Networks (RNN); the relation extraction includes: using rule-based, template-based, or machine learning methods to extract semantic relationships between asset entities from text, such as the correlation between equipment connection topology and faults; commonly used relation extraction models include PCNN or Transformer models; the attribute association includes: associating asset monitoring data streams with asset entities to form real-time attribute value updates; for example, mapping server CPU utilization and memory usage monitoring indicators to the corresponding attributes of server entities; the ontology matching includes: using ontology matching technology to semantically associate extracted asset instances with ontology concepts to achieve data and ontology fusion; the extracted asset entities and relational knowledge elements are stored in the form of triples to form a knowledge base. ,express Head entity, Indicates the tail entity. It represents the relationship between entities, where an entity is a mapping of an asset in a triple.

[0069] Since asset knowledge originates from multi-source heterogeneous data and changes dynamically in real time, knowledge fusion and reasoning are required to form a high-quality, consistent asset knowledge graph.

[0070] An asset knowledge graph is constructed through steps including knowledge representation learning, knowledge fusion, knowledge graph storage, and knowledge graph updating.

[0071] Knowledge Representation Learning: Employing knowledge representation models, such as the TransE or TransR models, asset entities and relationships are embedded into a low-dimensional semantic space, ensuring that semantically similar entities are close in distance within the embedding space. Knowledge Fusion: Integrating knowledge representation learning, predicate logic reasoning, and rule-based reasoning, conflict resolution, consistency verification, and new knowledge reasoning are performed on RDF knowledge constrained by domain ontology. Knowledge Graph Storage: Utilizing a graph database to store the asset knowledge graph. Knowledge Graph Update: Real-time updates of attribute values ​​in the asset knowledge graph based on changes in state parameters during asset operation; simultaneously, the knowledge graph also supports periodic batch updates, integrating changes to asset master data and externally introduced industry knowledge bases.

[0072] The asset knowledge graph obtained by fusion reasoning formally represents the static attributes, dynamic states, and topological semantic relationships between assets, providing rich semantic and high-quality knowledge support for intelligent operation and maintenance decision-making. The construction of the knowledge graph can significantly improve the accuracy and efficiency of tasks such as defect cause analysis, fault diagnosis, and risk prediction.

[0073] During the knowledge graph construction process, the semantic similarity between asset entities is calculated using the TransE model; for any triple in the asset knowledge graph... Learning triples through the TransE model Low-dimensional vector representation in the embedding space This makes the head entity embedding vector via relation vector Translation and tail entity embedding vector As close as possible, that is: .

[0074] The TransE model employs an interval-based ranking loss function. Perform optimized training:

[0075]

[0076] in It is the set of all real triples in the asset knowledge graph. This is the set of negative sample triples obtained by randomly replacing the head or tail entities. It is the negative sample head entity embedding vector. It is the embedding vector of the negative sample tail entity. This represents the distance between positive and negative samples. Measuring the Euclidean distance between entity embeddings This indicates the ReLU operation, used to obtain the non-negative part of the value.

[0077] The asset entity and relation embedding vectors learned through the TransE model can be used to calculate the semantic similarity between entities. The semantic similarity is calculated using the cosine similarity of the embedding vectors.

[0078]

[0079] Based on triplet Low-dimensional vector representation is performed in the embedding space to support knowledge services such as similar asset retrieval and fault propagation analysis on the knowledge graph.

[0080] Step 3: Train an asset fault diagnosis model based on a deep learning model, learn asset fault patterns from the asset knowledge graph, and detect whether the current asset has a fault according to the asset operation and maintenance strategy.

[0081] The constructed asset knowledge graph is converted into a low-dimensional dense vector representation, which serves as the input to the fault diagnosis model. The TransE model's knowledge graph representation learning algorithm is used to learn the embedding vectors of each asset node and relation edge. For any asset node in the knowledge graph... To obtain its embedding vector For the relationship edge of asset nodes To obtain its embedding vector ,in is the dimension of the embedded vector.

[0082] The collected asset dynamic data is also used as input to the fault diagnosis model; for any asset node, the monitoring indicators of the asset node are extracted. ,in To monitor the number of indicators, For timestamps.

[0083] A fault diagnosis model based on a Graph Attention Network (GAT) model and a Long Short-Term Memory Network (LSTM) model is constructed; the model includes:

[0084] Asset Embedding Layer: Embedding asset knowledge graphs into vectors This serves as the initial characteristic representation of an asset node.

[0085] Graph Attention Layer: Aggregates neighbor information of asset nodes through an attention mechanism and updates the feature representation of the nodes; for asset nodes... Its attention coefficient Represents asset nodes right The importance is calculated using the following formula:

[0086]

[0087] in, For asset nodes The set of neighboring nodes, This is the weight matrix. and Represents asset nodes and Embedded vector, For attention vectors, This represents the transpose of a vector. This represents a vector concatenation operation. This indicates the computation of the concatenated feature vector and the attention vector. The similarity between them For activation function, It is any node different from the asset node. and The vector representation of the neighboring asset nodes.

[0088] LSTM layer: Real-time monitoring data for assets Temporal modeling is performed to capture the dynamic changes in asset status; the input to the LSTM layer is the asset node. The updated feature representation and real-time monitoring data are concatenated, and the output is the hidden state of the asset node. :

[0089]

[0090] in, For nodes Feature representation updated after graph attention layer Indicates that the LSTM layer is in The hidden state at any given moment.

[0091] Fault classification layer: based on the output of the LSTM layer Through a fully connected layer and a softmax function, asset nodes are output. At any moment Fault types :

[0092]

[0093]

[0094] in, Represents asset nodes The probability distribution of different types of failures occurring over time. and These are the weight matrix and bias vector of the classification layer, respectively. Indicates taking The index with the highest probability is used as the predicted fault category.

[0095] Based on asset risk assessment results and historical operation and maintenance records, the failure types of assets are labeled to form training samples. Fault types can be categorized into multiple granularities, such as hardware faults, software faults, network faults, and security faults.

[0096] Using the cross-entropy loss function and the Adam optimizer, end-to-end supervised learning is performed on the fault diagnosis model on the training set to learn the model parameters. , , and .

[0097] For newly collected asset data, the trained fault diagnosis model is used to predict the fault type; based on the asset risk level and the predicted fault type, the fault diagnosis results are output, including the fault asset ID, fault type, fault time and fault location information.

[0098] The fault diagnosis results are fed back to the constructed operation and maintenance strategy, and corresponding maintenance operations are carried out on assets that are predicted to fail, such as fault isolation, anomaly analysis, configuration adjustment and security hardening. At the same time, the diagnosis results are updated to the asset knowledge graph to form a knowledge loop of asset faults.

[0099] This step utilizes asset knowledge graph embedding and real-time monitoring data, combined with graph attention networks and LSTM, to construct an asset fault diagnosis model. This model can fully utilize the static attributes, dynamic states, and topological association information of assets to achieve real-time prediction and location of asset faults, providing a basis for intelligent operation and maintenance decisions for assets.

[0100] Step 4: Construct an asset operation and maintenance risk detection algorithm. Through real-time analysis of multi-dimensional asset data, identify asset security risk coefficients and determine asset security risk levels. Trigger corresponding asset operation and maintenance strategies based on asset security risk levels.

[0101] Acquire static asset data by extracting asset static data information from the asset knowledge graph, including asset type, model, specifications, deployment location, and configuration parameters, which constitute the asset static data features. ,in This is the dimension of static asset data.

[0102] Acquire dynamic asset data: Through the asset data acquisition interface, obtain real-time performance monitoring indicators of the assets, including CPU utilization, memory usage, disk I / O, network traffic, and anomaly logs, to form a time-series dataset. ,in For the amount of dynamic asset data, For timestamps.

[0103] Obtain asset knowledge graph data, extract topological connections, business dependencies, and historical fault associations between assets from the asset knowledge graph, and construct an asset relationship adjacency matrix. ,in The total number of assets. Represents asset nodes Asset Nodes There is a relationship between them; otherwise, the value is 0.

[0104] Based on static asset data and asset dynamic data We will construct an asset security risk assessment indicator system, which includes multiple dimensions such as asset importance, configuration compliance, performance anomaly degree, and event severity.

[0105] Calculate the importance weight of each indicator by using the relationships between indicators and historical risk events in the asset knowledge graph. Considering the degree centrality and betweenness centrality graph characteristics of indicator nodes, as well as the correlation strength between indicators and historical risk events.

[0106] For static asset data indicators, their weights are determined through expert experience or historical data analysis; for dynamic monitoring indicators, time series analysis methods can be used to predict future trends and anomalies, and the weights can be dynamically adjusted.

[0107] static asset data and asset dynamic data By combining features, a feature matrix for asset security risk assessment is constructed. .

[0108] Based on asset relationship adjacency matrix The random walk algorithm is used to calculate the correlation risk coefficient between assets. The random walk algorithm steps include: defining the transition probability matrix. Considering the semantic weights of the associated edges in the knowledge graph, we obtain the weighted transition probability matrix. The steady-state distribution of asset nodes is obtained through multiple iterative calculations. ,use Represents asset nodes and The associated risk coefficient.

[0109] Taking into account the asset's own characteristics and associated risks, an asset operation and maintenance risk detection algorithm is constructed to calculate the asset's comprehensive security risk score:

[0110]

[0111] in, This represents the static data dimension of assets. Indicates the quantity of dynamic asset data. Asset security risk assessment feature matrix No. The characteristic values ​​of the column, For subscript index parameters, and For asset nodes and Knowledge graph embedding vectors, The weighting coefficients for associated risks, This is used to measure the impact of the associated risk of assets on the overall risk score. For example, if the correlation between assets is weak, such as independent equipment, a smaller β value (e.g., 0.1-0.3) is set; if the correlation between assets is strong (e.g., network equipment, production equipment with close dependencies), a larger β value (e.g., 0.5-0.9) is set.

[0112] Comprehensive security risk score Normalization is performed to obtain the risk probability value of the asset. The Sigmoid function is used for mapping:

[0113]

[0114] in, The steepness of the normalized curve. The threshold for risk scoring can be set based on historical data statistics, calculating the risk of assets that have historically experienced failures. Value, as For example, if the average number of failed assets that have occurred in the past 1000 data points is... If the value is 65, then... Set to 65.

[0115] Based on the overall risk probability value of the asset Set risk level classification thresholds Assets are classified into four levels: low risk, medium risk, high risk, and severe risk.

[0116] The classification criteria include: when When classified as low risk, When classified as medium risk; When classified as high risk; It is classified as a serious risk. The probability threshold for risk level can be adjusted according to the security requirements and risk preferences of the business system.

[0117] For different asset security risk levels, corresponding operation and maintenance strategies are formulated in advance to form a mapping relationship between risk levels and strategies.

[0118] Operation and maintenance strategies include, but are not limited to: adjusting monitoring frequency, setting alarm thresholds, isolation and protection, access control, data backup, and emergency plans. The formulation of strategies needs to comprehensively consider factors such as the importance of assets, the degree of risk impact, and business continuity requirements, and balance security and availability.

[0119] Real-time monitoring of changes in asset security risk levels; when the risk level exceeds a preset threshold or a sudden change occurs, the corresponding operation and maintenance strategy is automatically triggered.

[0120] The strategy is executed by sending it to the asset management domain through a unified automated operation and maintenance platform, calling relevant interfaces to execute the strategy, and monitoring the execution status and results.

[0121] Based on feedback from strategy execution, the effectiveness of the strategy and the improvement in asset risk status are evaluated, forming a closed-loop feedback mechanism for strategy optimization.

[0122] This step fully utilizes multi-source data such as static asset data, dynamic monitoring, and knowledge graphs, and proposes a comprehensive risk scoring model based on dynamic allocation of indicator weights and random walks, which can comprehensively characterize the asset's own risks and related propagation risks. On this basis, risk-driven adaptive operation and maintenance decisions are realized through risk level classification and strategy matching, and a continuous optimization mechanism is introduced to continuously improve the intelligent level of asset security operation and maintenance.

[0123] Step 5: Integrate asset knowledge graph information with fault diagnosis and risk assessment results, use natural language generation technology to generate a comprehensive asset operation and maintenance decision plan, and push it to operation and maintenance management personnel for execution.

[0124] Semantic parsing is performed on the asset fault diagnosis results to extract asset fault information, including faulty assets, fault types, fault symptoms, and scope of impact, which is represented as a fault knowledge vector. Semantic parsing is also performed on the asset risk assessment results to extract asset risk information, including risky assets, risk levels, risk factors, and potential impacts, which is represented as a risk knowledge vector. Attributes, relationships, and contextual information related to faulty and risky assets are queried from the asset knowledge graph, and static asset attributes and dynamic monitoring data are integrated to form a comprehensive asset knowledge vector.

[0125] By using a pre-trained language model, with fault knowledge vectors, risk knowledge vectors, and comprehensive asset knowledge vectors as inputs, an operation and maintenance decision-making scheme is output.

[0126] The decision-making plan is pushed to the operation and maintenance management personnel; the operation and maintenance management personnel can edit and modify the plan through the human-computer interaction interface, input supplementary explanations, and form an executable operation and maintenance plan.

[0127] Operations and maintenance management personnel break down the operations and maintenance plan into multiple specific operations and maintenance tasks, which are then assigned to relevant technical personnel for execution.

[0128] During the execution of operation and maintenance tasks, asset status data is collected in real time through IoT devices, monitoring tools, and other means, and information such as operation logs and change records for each task is recorded to evaluate the effectiveness of the decision-making plan.

[0129] Example 2

[0130] Reference Figure 5 The second embodiment of this application provides an asset operation and maintenance decision management platform based on data fusion.

[0131] The platform includes: a data acquisition module, a knowledge graph module, an asset fault diagnosis module, an operation and maintenance risk detection module, and an operation and maintenance execution module.

[0132] The data acquisition module collects multi-source heterogeneous data throughout the entire asset lifecycle through a unified operation and maintenance platform, and performs data cleaning, transformation and fusion to form a standardized asset dataset.

[0133] The knowledge graph module utilizes ontology modeling and knowledge extraction techniques to construct a dynamically updated asset knowledge graph based on the fused asset dataset, including multi-dimensional information such as asset attributes, topological relationships, and operational status.

[0134] The asset fault diagnosis module is used to train an asset fault diagnosis model based on a deep learning model, learn asset fault patterns from an asset knowledge graph, and detect whether the current asset has a fault according to the asset operation and maintenance strategy.

[0135] The operation and maintenance risk detection module is used to construct an asset operation and maintenance risk detection algorithm. Through real-time analysis of multi-dimensional asset data, it identifies the asset security risk coefficient, determines the asset security risk level, and triggers the corresponding asset operation and maintenance strategy based on the asset security risk level.

[0136] The operation and maintenance execution module uses natural language generation technology to convert asset faults and asset operation and maintenance strategies identified based on knowledge graphs into natural language and pushes them to operation and maintenance management personnel, who then execute the operation and maintenance measures.

[0137] Example 3

[0138] This application provides a storage medium storing a computer program thereon. When the computer program is executed by a processor, it performs the steps of the method described above. Through the above technical solution, when the computer program is executed by a processor, it performs the method in any optional implementation of the above embodiments to achieve the following functions: acquiring multiple sets of historical electricity consumption data; dividing the historical electricity consumption data and outputting specific electricity consumption groups; calculating the historical electricity consumption data and outputting a moving average; analyzing the moving average and outputting the device type and peak time; acquiring the current time and real-time electricity consumption; analyzing and calculating the real-time electricity consumption based on the current time, peak time, and device type; and outputting abnormal electricity consumption information based on the calculation results.

[0139] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0140] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0141] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0142] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of this application without departing from the spirit and scope of protection of the claims. All of these variations are within the protection scope of this application.

Claims

1. An asset operation and maintenance decision management method based on data fusion, characterized in that, include: Through a unified operation and maintenance platform, multi-source heterogeneous data throughout the entire asset lifecycle are collected, and the data is cleaned, transformed, and integrated to form a standardized asset dataset. By utilizing ontology modeling and knowledge extraction techniques, a dynamically updated asset knowledge graph is constructed based on the fused asset dataset, including multi-dimensional information such as asset attributes, topological relationships, and operational status. An asset operation and maintenance risk detection algorithm is constructed. Through real-time analysis of multi-dimensional asset data, the asset security risk coefficient is identified and the asset security risk level is determined. Based on the asset security risk level, the corresponding asset operation and maintenance strategy is triggered. Perform multi-dimensional data analysis on asset data to obtain static asset data, dynamic asset data, and asset knowledge graph data; Construct an asset relationship adjacency matrix based on asset knowledge graph data; By combining features from static and dynamic asset data, an asset security risk assessment feature matrix is ​​constructed. Based on the asset relationship adjacency matrix, a random walk algorithm is used to calculate the association risk coefficient between assets; Based on the correlation risk coefficient between assets, an asset operation and maintenance risk detection algorithm is constructed to calculate the comprehensive security risk score of the assets; The comprehensive safety risk score is normalized to obtain the risk probability value of the asset. Based on the comprehensive risk probability value of the assets, a risk level classification threshold is set, and the assets are divided into four risk levels: low risk, medium risk, high risk, and severe risk. Different operation and maintenance strategies are assigned to each risk level. For different asset security risk levels, corresponding asset operation and maintenance strategies are formulated in advance to form a mapping relationship between risk levels and strategies; Real-time monitoring of changes in asset security risk levels; when the risk level exceeds a preset threshold or a sudden change occurs, the corresponding asset operation and maintenance strategy is automatically triggered. Train an asset fault diagnosis model based on a deep learning model, learn asset fault patterns from an asset knowledge graph, and the asset fault diagnosis model detects whether the current asset has a fault based on the asset operation and maintenance strategy. For newly acquired asset data, the trained fault diagnosis model is used to predict the fault type. Based on the asset security risk level and the predicted fault type, output fault diagnosis results, including fault asset ID, fault type, fault time and fault location information; The fault diagnosis results are fed back to the constructed operation and maintenance strategy, and corresponding maintenance operations are carried out on assets that are predicted to fail. By using natural language generation technology, asset faults and asset operation and maintenance strategies identified based on knowledge graphs are converted into natural language and pushed to operation and maintenance management personnel, who then implement the operation and maintenance measures.

2. The asset operation and maintenance decision management method based on data fusion according to claim 1, characterized in that, The assets include: switches, routers, network security equipment, network terminals, data center proxy servers, virtual machines, operating systems, databases, middleware, containers, IP addresses, and domain name information; Collect various parameter information of the assets, including: static attribute data of the assets, dynamic monitoring data of the assets, and operation and maintenance data of the assets; Data preprocessing is performed on the collected data, including: missing value handling, noise removal, outlier detection, and data normalization. All collected data are aggregated, and the JS divergence between data points is used as the data priority sorting method to resolve the problem of data conflicts for the same indicator between multiple data sources.

3. The asset operation and maintenance decision management method based on data fusion according to claim 2, characterized in that, The asset ontology is modeled to construct an ontology model covering the entire lifecycle management of assets, including: asset concepts, asset relationships, and asset attributes, and the asset attributes are associated with asset entities; the asset concepts, asset relationships, and asset attributes are formally defined using an ontology description language to construct the asset ontology model. Based on the constructed IT asset ontology model, structured asset knowledge is extracted from multi-source heterogeneous datasets, including entity recognition, relation extraction, attribute association, and ontology matching. An asset knowledge graph is constructed through steps including knowledge representation learning, knowledge fusion, knowledge graph storage, and knowledge graph updating.

4. The asset operation and maintenance decision management method based on data fusion according to claim 3, characterized in that, The constructed asset knowledge graph is converted into a low-dimensional dense vector representation, which is used as the input to the fault diagnosis model. The collected dynamic asset data is also used as the input to the fault diagnosis model. The knowledge graph representation learning algorithm of the TransE model is used to learn the embedding vector of each asset node and relation edge. Construct a fault diagnosis model based on the Graph Attention Network (GAT) model and the Long Short-Term Memory (LSTM) model; The fault diagnosis model includes: an asset embedding layer, a graph attention layer, an LSTM layer, and a fault classification layer.

5. The asset operation and maintenance decision management method based on data fusion according to claim 4, characterized in that, Based on the asset security risk assessment results and historical operation and maintenance records, the fault types of assets are labeled to form training samples for training the fault diagnosis model. The fault diagnosis model is trained through end-to-end supervised learning using training samples to learn the model parameters. When the fault diagnosis model is trained, the trained fault diagnosis model is used to predict the fault type. Based on the asset safety risk level and the predicted fault type, the fault diagnosis result is output.

6. The asset operation and maintenance decision management method based on data fusion according to claim 5, characterized in that, Semantic parsing is performed on the asset fault diagnosis results, and the asset fault results are represented as fault knowledge vectors; Semantic parsing is performed on the asset security risk assessment results to represent asset security risks as risk knowledge vectors; Query faulty assets and risky assets from the asset knowledge graph, and represent the query results using a comprehensive asset knowledge vector; By using a pre-trained language model, with fault knowledge vectors, risk knowledge vectors, and comprehensive asset knowledge vectors as inputs, the system outputs textual solutions for operation and maintenance decisions.

7. The asset operation and maintenance decision management method based on data fusion according to claim 6, characterized in that, The written decision plan is pushed to the operations and maintenance management personnel, who can then edit and modify the plan through a human-computer interaction interface and implement operations and maintenance measures on the assets.

8. A data fusion-based asset operation and maintenance decision management platform, used to implement the data fusion-based asset operation and maintenance decision management method according to any one of claims 1 to 7, characterized in that, include: The system includes a data acquisition module, a knowledge graph module, an asset fault diagnosis module, an operation and maintenance risk detection module, and an operation and maintenance execution module. The data acquisition module collects multi-source heterogeneous data throughout the entire asset lifecycle through a unified operation and maintenance platform, and performs data cleaning, transformation and fusion to form a standardized asset dataset. The knowledge graph module utilizes ontology modeling and knowledge extraction techniques to construct a dynamically updated asset knowledge graph based on the fused asset dataset, including multi-dimensional information such as asset attributes, topological relationships, and operational status. The operation and maintenance risk detection module is used to construct an asset operation and maintenance risk detection algorithm. Through real-time analysis of multi-dimensional asset data, it identifies the asset security risk coefficient, determines the asset security risk level, and triggers the corresponding asset operation and maintenance strategy according to the asset security risk level. The asset fault diagnosis module is used to train an asset fault diagnosis model based on a deep learning model, learn asset fault patterns from an asset knowledge graph, and detect whether the current asset has a fault according to the asset operation and maintenance strategy. The operation and maintenance execution module uses natural language generation technology to convert asset faults and asset operation and maintenance strategies identified based on knowledge graphs into natural language and pushes them to operation and maintenance management personnel, who then execute the operation and maintenance measures.

Images