CN-122022084-A - Intelligent operation and maintenance decision method and system based on multi-element heterogeneous data fusion

CN122022084ACN 122022084 ACN122022084 ACN 122022084ACN-122022084-A

Abstract

The invention relates to the field of data processing and information technology operation and maintenance, and discloses an intelligent operation and maintenance decision method and system based on multi-element heterogeneous data fusion, wherein the method is adaptive to the operation of an artificial intelligent optimization operating system, an artificial intelligent middleware and a function library and comprises the steps of constructing a dynamic causal graph reflecting the causal relation between a system state and an environmental factor based on a causal structure learning algorithm; constructing a multi-agent game decision model based on the dynamic causal graph, and respectively setting a global operation and maintenance target and a local demand of a component as a leader and follower strategy of a game; and solving game balance by utilizing multi-agent reinforcement learning, and outputting an optimal joint operation and maintenance decision instruction. The system is composed of a data acquisition preprocessing module, a multi-mode feature fusion module, a causal structure learning module, a game decision solving module and an execution monitoring module. The method and the system can accurately extract the causal logic by fusing causal inference and game theory, improve the robustness and the interpretability of decisions, and realize global optimal collaborative configuration.

Inventors

OU WEISHENG
ZHOU YA
LI QUN

Assignees

珠海德茵电气有限公司

Dates

Publication Date: 20260512
Application Date: 20260416

Claims (10)

1. The intelligent operation and maintenance decision method based on the multi-element heterogeneous data fusion is characterized by being adaptive to the operation of an artificial intelligent optimization operating system, an artificial intelligent middleware and a function library, and comprises the following steps of: s1, acquiring a multi-element heterogeneous data stream of a system to be monitored; S2, preliminarily screening the multi-element heterogeneous fusion characteristics by using a constraint-based path constraint algorithm to generate a dynamic causal graph reflecting the physical logic of the system in real time; S3, constructing a multi-agent game decision model based on the dynamic causal graph, defining a hardware component or a functional module in the system to be monitored as a controlled agent, endowing the controlled agent with a local utility function, defining an operation and maintenance global optimization target as a leader agent, guiding decision behaviors of the controlled agent by the leader agent through formulating a global strategy vector, and establishing a state transfer constraint between the controlled agents based on the dynamic causal graph, wherein a Stark primary game relationship is formed between the leader agent and the controlled agent; S4, initializing strategy network and value network parameters of each agent, solving Nash equilibrium state of the multi-agent game decision model by utilizing a multi-agent reinforcement learning algorithm, and outputting an optimal joint operation and maintenance decision instruction.
2. The intelligent operation and maintenance decision method based on multi-element heterogeneous data fusion according to claim 1, wherein the multi-element heterogeneous data stream comprises structured time sequence data and unstructured text data, the structured time sequence data covers structured detection and recognition result data output by computer audio visual software and biological feature recognition software, normalization processing and sliding window feature extraction are performed on the structured time sequence data to generate continuous state vectors reflecting equipment performance features, entity recognition and event extraction are performed on the unstructured text data to convert the unstructured text data into discrete intervention signal features, the continuous state vectors and the intervention signal features are input into a cross-modal feature fusion network adapting to an artificial intelligent optimization operating system, an artificial intelligent middleware and a function library, information of different dimensions is mapped into an embedded space of uniform dimension through nonlinear transformation inside the cross-modal feature fusion network, and multi-element heterogeneous fusion features representing real-time operation states of the system to be monitored are output.
3. The intelligent operation and maintenance decision method based on multi-component heterogeneous data fusion according to claim 1, wherein the condition independence relation among nodes is determined by executing high-order condition independence test among feature pairs to generate an initial undirected graph, a score-based greedy equivalent search algorithm is applied to score and optimize the initial undirected graph, directional edge directions among nodes are determined by maximizing Bayesian information criterion scores to form an initial causal structure, a graph neural network is introduced to dynamically update the initial causal structure, a nonlinear evolution rule of node features along with time change is captured by using an attention mechanism, and a dynamic causal graph reflecting system physical logic in real time is generated.
4. The intelligent operation and maintenance decision method based on multi-component heterogeneous data fusion according to claim 1, wherein the leader agent outputs the global strategy vector of a preset dimension according to the current state of the dynamic causal graph, each controlled agent as a follower outputs action selection based on the local utility function after receiving the global strategy vector, and the system outputs reward signals according to joint actions and iteratively updates parameters by using a gradient descent algorithm until the strategies of all agents reach a balance point where higher profits can not be obtained by changing own strategies without changing the strategies of other agents.
5. The intelligent operation and maintenance decision method based on multi-component heterogeneous data fusion according to claim 2, wherein the obtained structured time sequence data comprises central processing unit utilization rate, memory occupancy rate, input and output waiting time, equipment real-time operation temperature, equipment defect detection structured data output by computer visual and audio sense software and operation and maintenance right verification structured data output by biological feature recognition software; The unstructured text data comprise equipment operation log information, historical maintenance work order records, system state description recorded by a management terminal, image identification marking text output by computer visual and audio sense software and identity identification event text output by biological feature identification software.
6. The intelligent operation and maintenance decision method based on multi-element heterogeneous data fusion according to claim 2, wherein the process of generating the multi-element heterogeneous fusion features comprises the steps of constructing a cross-modal feature fusion network based on a gating circulation unit, adapting an artificial intelligent optimization operating system, an artificial intelligent middleware and a function library to operate, performing feature compression and weight distribution on the received continuous state vector and the interference signal features through the gating circulation unit, and projecting original information with heterogeneous formats into the embedded space with fixed dimensions to realize integrated characterization on real-time operation states.
7. The intelligent operation and maintenance decision method based on multi-component heterogeneous data fusion according to claim 3, wherein the process of applying the greedy equivalent search algorithm based on score comprises starting from a blank graph structure, attempting to increase a directed edge capable of improving a model score in each round of iteration, entering a pruning stage when the score stops increasing and attempting to remove redundant directed edges, and performing operations of increasing edges, deleting edges or reversing edges in a graph space until the greedy equivalent graph structure is converged on a locally optimal equivalent class graph structure; the model score is determined by applying a penalty term to the number of edges in the graph based on evaluating the model's ability to interpret the observed data.
8. The intelligent operation and maintenance decision method based on multi-element heterogeneous data fusion according to claim 3, wherein the process of dynamically updating the initial causal structure by using a graph neural network comprises the steps of inputting the multi-element heterogeneous fusion characteristic as a node attribute into a space-time graph convolution network, capturing the evolution trend of a single performance index in a preset time window by using a time convolution layer, aggregating the characteristic information of neighbor nodes by using a space-time graph convolution layer based on the connection strength of causal edges, and processing the output result of the space-time graph convolution network by using a residual connection and normalization layer to generate a dynamic node characteristic representation comprising prediction performance.
9. The intelligent operation and maintenance decision method based on multi-component heterogeneous data fusion according to claim 4, wherein the multi-agent reinforcement learning algorithm adopts a depth deterministic strategy gradient algorithm with a centralized training and distributed execution architecture; And in the execution stage, each intelligent agent outputs the optimal joint operation and maintenance decision instruction in real time based on the local observation data and the received global strategy vector.
10. The intelligent operation and maintenance decision system based on multi-element heterogeneous data fusion is characterized by being used for realizing the method of any one of claims 1 to 9, adapting to the deployment and operation of an artificial intelligent optimizing operation system, an artificial intelligent middleware and a function library, belonging to an artificial intelligent application software development technology landing carrier, and comprising a data acquisition and preprocessing module, a multi-mode feature fusion module and a multi-mode feature fusion module, wherein the data acquisition and preprocessing module is configured to acquire and synchronize multi-element heterogeneous data streams of a system to be monitored, the multi-element heterogeneous data streams comprise structured time sequence data and unstructured text data, perform normalization and feature extraction on the structured time sequence data, and perform entity recognition and event extraction on the unstructured text data; The system comprises a causal structure learning module, a game decision solving module, a game decision making module and a multi-element heterogeneous fusion module, wherein the causal structure learning module is configured to construct a dynamic causal graph by utilizing a causal inference algorithm, the causal structure learning module is connected with the multi-element feature fusion module, the causal structure learning module comprises a knowledge base checking unit and an execution monitoring module, the knowledge base checking unit is used for injecting a preset priori logic relationship into a causal structure searching process, checking the causal discovery process by applying punishment weights to directed edges which do not accord with physical common sense, the game decision solving module is used for receiving the dynamic causal graph and the multi-element heterogeneous fusion feature, a plurality of controlled agent units and a leader control unit are configured to establish a game model based on the dynamic causal graph and solve a Nash equilibrium state by utilizing a multi-agent reinforcement learning algorithm so as to output an optimal joint operation and maintenance decision command, and the execution monitoring module is configured to receive the command output by the game decision solving module and act on physical equipment, and simultaneously collect and transmit the feedback result after execution to the data acquisition and preprocessing module in real time so as to form a closed loop control circuit.

Description

Intelligent operation and maintenance decision method and system based on multi-element heterogeneous data fusion Technical Field The invention belongs to the field of data processing and information technology operation and maintenance, and particularly relates to an intelligent operation and maintenance decision method and system based on multi-element heterogeneous data fusion. Background With the deep integration of industrial Internet and big data technology, intelligent operation and maintenance has become a core foundation for guaranteeing efficient and stable operation of large-scale complex systems such as smart factories, modern data centers and smart city infrastructure. By integrating massive industrial sensors, equipment logs and business data, the system realizes the spanning development from traditional manual inspection to automatic monitoring and digital management, and greatly improves the perceptibility and management efficiency of the equipment running state in a complex environment. The intelligent operation and maintenance decision technology based on multi-element heterogeneous data fusion is the core focus of research in the field at present. The technology aims to establish a comprehensive information model reflecting the overall operation of the system by carrying out deep integration and association analysis on data with different dimensions and formats such as equipment operation parameters, time sequence performance indexes, alarm texts, historical work order records and the like. The basic principle is that an abnormal mode is identified from mass data flow by using a data-driven algorithm, and the fault risk is predicted, so that automatic operation and maintenance strategy generation is driven to reduce the unplanned shutdown frequency and optimize the system resource allocation. Existing intelligent operation and maintenance technologies still face a number of challenges in practical applications. The existing system generally relies on a data-driven association analysis model excessively, is basically only subjected to statistical correlation among captured data, is extremely easy to be influenced by external environment variables or random artificial interference to generate false association, and fails to sufficiently integrate multidimensional operation and maintenance association data output by computer visual sense software and biological feature recognition software, so that obvious limitations exist on data sources and feature dimensions, and the system frequently outputs wrong decision instructions. Due to the lack of deep mining of physical world real-world causality, models tend to fall into dilemma of knowing and not knowing the complex failure of complex systems, resulting in very poor robustness of decision suggestions and lack of necessary interpretability. The existing decision mechanism often ignores dynamic coordination and benefit game among all components, and is difficult to realize multi-objective optimal configuration of dimensions such as energy consumption, service life and the like on the premise of guaranteeing global service level. The problems commonly cause the serious deficiency of the capability of the existing operation and maintenance system in the aspects of precisely stripping causal logic, coping with unseen fault scenes and realizing global optimal decision in complex environments, and are also difficult to adapt to the landing requirement of artificial intelligent application software development. Disclosure of Invention The invention aims to provide an intelligent operation and maintenance decision method and system based on multi-element heterogeneous data fusion, which can solve the problems in the background technology. Aiming at the core technical defects that when the existing intelligent operation and maintenance system processes complex industrial environment or large-scale computing clusters, environment interference and false association are difficult to strip due to excessive dependence on statistical correlation analysis, depth adaptation with an artificial intelligent optimization operating system, an artificial intelligent middleware and a function library is not realized, multidimensional operation and maintenance data of computer vision software and biological feature recognition software cannot be fully fused, and further decision robustness is insufficient and interpretability is poor when the complex faults are faced, the invention provides a brand new technical framework combining causal inference depth map learning and Stark-berg game decision, belongs to an artificial intelligent application software development technical innovation scheme, and aims to realize accurate extraction of data to causal logic under complex dynamic environment and construct a globally optimal collaborative decision system. The technical scheme adopted by the invention is that 1, a multi-element heterogeneous data stream of a system to be