Fault prediction method and system for intelligent accessory monitoring platform

Abstract

The invention belongs to the technical field of fault prediction, and discloses a fault prediction method and system for an intelligent accessory monitoring platform. The method comprises the following steps: an intelligent accessory monitoring platform constructs a knowledge graph, a fault prediction model and a fault maintenance strategy generation model; performing knowledge mapping on the real-time monitoring log by using a knowledge graph to obtain a semantic enhanced real-time monitoring log; performing fault prediction by using a fault prediction model according to the real-time operation monitoring data and the semantic enhanced real-time monitoring log to obtain a real-time fault prediction result; generating a fault maintenance strategy by using a fault maintenance strategy generation model according to the real-time fault prediction result to obtain a real-time fault maintenance strategy; and visually displaying the real-time fault prediction result and the real-time fault maintenance strategy of the attachment. According to the invention, the problems of insufficient fault prediction accuracy, insufficient data processing capability and low intelligent degree in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault prediction, and in particular relates to a fault prediction method and system for an intelligent accessory monitoring platform. Background Art

[0002] With the development of intelligent technology, intelligent accessory monitoring platforms play an important role in industrial production, aerospace and other fields. In the intelligent accessory monitoring platform, accessories are various auxiliary devices or components connected to the main equipment or system, including sensors and actuators, communication equipment, data processing units, mechanical parts, software and firmware, etc. In the fields of aerospace and industrial production, these accessories may include specially designed components to meet the needs of specific applications, such as aircraft engine monitoring systems, sensor arrays of industrial robots, etc. By integrating and managing these accessories, the intelligent accessory monitoring platform realizes real-time monitoring, status analysis, fault diagnosis and predictive maintenance of the entire system or equipment, thereby improving the operating reliability and maintenance efficiency of systems or equipment in fields such as industrial production and aerospace.

[0003] The existing fault prediction technology of the intelligent attachment monitoring platform has the following defects: 1) Insufficient fault prediction accuracy: Existing fault prediction technologies often rely on historical data and statistical analysis, but these models may not be able to accurately predict all potential fault conditions in complex and changing working environments; 2) Insufficient data processing capabilities: Existing technologies may have low computational efficiency when processing large amounts of real-time monitoring data, resulting in insufficient real-time performance of fault prediction; 3) Low intelligence: Existing technologies fail to effectively utilize domain knowledge and accessory manuals, resulting in insufficient authenticity in fault prediction. Fault repair strategy generation often relies on preset rules and lacks an intelligent decision-making process, making it impossible to provide optimal repair solutions for different fault situations. Summary of the invention

[0004] In order to solve the problems of insufficient fault prediction accuracy, insufficient data processing capability and low intelligence in the prior art, the present invention aims to provide a fault prediction method and system for an intelligent accessory monitoring platform.

[0005] The technical solution adopted by the present invention is: A fault prediction method for an intelligent accessory monitoring platform comprises the following steps: The intelligent accessory monitoring platform builds a knowledge graph based on the accessory manual and uses artificial intelligence algorithms to build a fault prediction model and a fault repair strategy generation model; The intelligent attachment monitoring platform collects the real-time operation monitoring data and real-time monitoring logs of attachments, and uses the knowledge graph to perform knowledge mapping on the real-time monitoring logs to obtain semantically enhanced real-time monitoring logs; The intelligent attachment monitoring platform uses the fault prediction model to perform fault prediction based on real-time operation monitoring data and semantically enhanced real-time monitoring logs, and obtains real-time fault prediction results; The intelligent accessory monitoring platform generates fault repair strategies based on real-time fault prediction results and uses the fault repair strategy generation model to obtain real-time fault repair strategies; The intelligent accessory monitoring platform visualizes the real-time fault prediction results and real-time fault repair strategies of accessories.

[0006] Furthermore, the intelligent accessory monitoring platform constructs a knowledge graph based on the accessory manual and uses an artificial intelligence algorithm to construct a fault prediction model and a fault repair strategy generation model, including the following steps: The intelligent attachment monitoring platform builds a knowledge graph based on the attachment manual using pre-trained named entity and entity relationship extraction models; Collecting a number of historical operation monitoring data and a number of historical monitoring logs of the attachment, and preprocessing them to obtain a number of preprocessed historical operation monitoring data and a number of preprocessed historical monitoring logs; Use the knowledge graph to perform knowledge mapping on several pre-processed historical monitoring logs to obtain several semantically enhanced historical monitoring logs; Based on some historical operation monitoring data and some semantically enhanced historical monitoring logs, a deep learning algorithm is used to build a fault prediction model and generate some historical fault prediction results. Based on several historical fault prediction results, an enhanced reinforcement learning algorithm is used to build a fault maintenance strategy generation model.

[0007] Furthermore, the named entity and entity relationship extraction model is constructed based on the BERT-Double CRF algorithm, and the named entity and entity relationship extraction model includes a text feature extraction module constructed based on the BERT algorithm, a named entity extraction module constructed based on the CRF algorithm, and an entity relationship extraction module constructed based on the CRF algorithm, and the text feature extraction module is connected to the named entity extraction module and the entity relationship extraction module respectively; The fault prediction model is built based on the GCN-LSTM-Attention-MLP algorithm, and the fault prediction model includes a graph structure feature extraction module built based on the GCN algorithm, a sequence feature extraction module built based on the LSTM algorithm, an attention weight module built based on the Attention mechanism, and a fault prediction module built based on the MLP algorithm; The fault repair strategy generation model is constructed based on the ICPO-DQN algorithm, and the encryption update strategy generation model includes a parameter initialization module constructed based on the ICPO algorithm and a reinforcement learning module constructed based on the DQN algorithm. The reinforcement learning module is provided with an intelligent agent, a deep Q network and an experience replay pool. The intelligent agent is connected to the deep Q network and the experience replay pool respectively, and the parameter initialization module is connected to the deep Q network.

[0008] Furthermore, the intelligent attachment monitoring platform uses the pre-trained named entity and entity relationship extraction model according to the attachment manual to build a knowledge graph, including the following steps: The intelligent attachment monitoring platform performs word segmentation, stop word removal, and professional terminology unification on the attachment manual to obtain the pre-processed attachment manual text data; Using the named entity and entity relationship extraction model, named entity and entity relationship extraction is performed on the preprocessed attachment manual text data to obtain several knowledge named entities and several knowledge entity relationships; According to several knowledge named entities and several knowledge entity relationships, a knowledge graph is constructed to obtain the corresponding knowledge graph.

[0009] Furthermore, using the knowledge graph, knowledge mapping is performed on several pre-processed historical monitoring logs to obtain several semantically enhanced historical monitoring logs, including the following steps: Use the named entity and entity relationship extraction model to extract named entities and entity relationships from the preprocessed historical monitoring logs to obtain several historical log named entities and several historical log entity relationships; Map the knowledge named entities of the knowledge graph to the historical log named entities, and associate the knowledge entity relationships with the corresponding historical log named entities to obtain the initial semantically enhanced historical monitoring logs; The historical log entity relationships and associated knowledge entity relationships of the same historical log named entity are deleted repeatedly to obtain the final semantically enhanced historical monitoring log; Traverse all the pre-processed historical monitoring logs to obtain several final semantically enhanced historical monitoring logs.

[0010] Furthermore, based on a number of historical operation monitoring data and a number of semantically enhanced historical monitoring logs, a deep learning algorithm is used to construct a fault prediction model and generate a number of historical fault prediction results, including the following steps: Randomly associate some historical operation monitoring data with some semantically enhanced historical monitoring logs, and add corresponding real labels to obtain some fault prediction samples; Several fault prediction samples are divided into model training set and model test set in a ratio of 7:3, and the GCN-LSTM-Attention-MLP algorithm is used to build the initial fault prediction model; Input the model training set, optimize the initial fault prediction model, obtain the optimized fault prediction model, and generate several historical fault prediction results; The model test set is input to perform model testing on the optimized fault prediction model to obtain the test accuracy. If the test accuracy is greater than the accuracy threshold, the final fault prediction model is output.

[0011] Furthermore, based on several historical fault prediction results, an enhanced reinforcement learning algorithm is used to construct a fault repair strategy generation model, which includes the following steps: Use the ICPO-DQN algorithm to build an initial fault repair strategy generation model; the initial fault repair strategy generation model includes a parameter initialization module and an initial reinforcement learning module; Taking minimizing the mean square error as the optimization goal, the parameter initialization module is used to initialize the parameters of the deep Q network of the initial reinforcement learning module to obtain an optimized reinforcement learning module; According to several historical fault prediction states of historical fault prediction results, a state space of an optimized reinforcement learning module is defined, and according to several preset fault repair actions, an action space of an optimized reinforcement learning module is defined; Define the reward function of the optimized reinforcement learning module based on the impact of the preset fault repair action on the historical fault prediction status; Based on the state space, action space and reward function, several historical fault prediction results are input, the intelligent agent of the optimized reinforcement learning module is trained, the final reinforcement learning module is obtained, and several historical fault maintenance strategy generation experiences are generated; The parameter initialization module and the final reinforcement learning module are integrated to obtain the final fault repair strategy generation model, and several historical fault repair strategy generation experiences are stored in the experience replay pool.

[0012] Furthermore, the intelligent attachment monitoring platform uses the fault prediction model to perform fault prediction based on the real-time operation monitoring data and the semantically enhanced real-time monitoring log, and obtains the real-time fault prediction result, including the following steps: Intelligent attachment monitoring platform, which inputs real-time operation monitoring data and semantically enhanced real-time monitoring logs into the fault prediction model; Use the graph structure feature extraction module to extract the real-time graph structure features of the real-time monitoring log after semantic enhancement, and use the sequence feature extraction module to extract the real-time sequence features of the real-time operation monitoring data; According to the preset attention weight value, the attention weight module is used to perform weighted fusion on the real-time graph structure features and the real-time sequence features to obtain the real-time weighted fusion features; According to the real-time weighted fusion features, the fault prediction module is used to perform fault prediction and obtain real-time fault prediction results.

[0013] Furthermore, the intelligent accessory monitoring platform generates a fault repair strategy based on the real-time fault prediction result using a fault repair strategy generation model to obtain a real-time fault repair strategy, including the following steps: The intelligent accessory monitoring platform analyzes the real-time fault prediction results to obtain a number of real-time fault prediction states, and updates the state space of the fault maintenance strategy generation model according to the real-time fault prediction states to obtain an updated state space; According to the real-time fault prediction results, the experience playback pool of the fault maintenance strategy generation model is searched to obtain the real-time fault maintenance strategy generation experience that matches successfully, and according to the real-time fault maintenance strategy generation experience, the action space of the fault maintenance strategy generation model is updated to obtain an updated action space; Connecting the updated state space to the input layer of the deep Q network of the fault repair strategy generation model, and connecting the updated action space to the output layer of the deep Q network of the fault repair strategy generation model to obtain an updated deep Q network; According to the reward function, the reward value of each real-time fault repair alternative action in the updated action space for each real-time fault prediction state in the updated state space is obtained, and the intelligent agent is used to control the updated deep Q network to generate the Q value of each real-time fault repair alternative action in the updated action space; According to the reward value, the intelligent agent is used to iteratively update the Q value of the real-time fault repair alternative action to obtain an updated Q value until the number of iterations reaches the iteration number threshold, and a greedy strategy is used to select the real-time fault repair alternative action with the highest updated Q value as the real-time fault repair execution action, and all real-time fault repair execution actions are integrated to obtain the real-time fault repair strategy.

[0014] A fault prediction system for an intelligent accessory monitoring platform is used to implement a fault prediction method for the intelligent accessory monitoring platform. The system is applied to the intelligent accessory monitoring platform, and the system includes an initialization unit, a knowledge mapping unit, a fault prediction unit, a fault repair strategy generation unit and a visualization unit connected in sequence.

[0015] The beneficial effects of the present invention are: The present invention provides a fault prediction method and system for an intelligent accessory monitoring platform. The method uses an artificial intelligence algorithm to build a fault prediction model, which can more accurately identify and predict potential fault conditions and maintain a high prediction accuracy even in a complex and changeable industrial environment. The method combines real-time operation monitoring data and real-time monitoring logs to improve the comprehensiveness of fault prediction data, extract more valuable fault information, and perform knowledge mapping on real-time monitoring logs, which can more accurately capture the correlation between accessories and fault propagation paths, further improving the accuracy and efficiency of fault prediction. The method performs automated and efficient data processing based on artificial intelligence models, which can quickly process a large amount of real-time data and ensure the real-time nature of fault prediction. A more comprehensive and in-depth knowledge graph is built through the accessory manual, which makes full use of domain knowledge and expert experience in the accessory manual, making fault prediction and maintenance strategy generation more accurate and targeted. The method uses an artificial intelligence algorithm to build a fault maintenance strategy generation model, which can intelligently generate fault maintenance strategies according to fault prediction conditions and provide optimal maintenance plans, thereby improving the degree of intelligence and realizing a fault prediction response mechanism.

[0016] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of the fault prediction method of the intelligent accessory monitoring platform in the present invention.

[0018] Figure 2 It is a structural block diagram of the fault prediction system of the intelligent accessory monitoring platform in the present invention. DETAILED DESCRIPTION

[0019] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments.

[0020] Embodiment 1: like Figure 1 As shown, this embodiment provides a fault prediction method for an intelligent accessory monitoring platform, comprising the following steps: S1: The intelligent accessory monitoring platform builds a knowledge graph based on the accessory manual and uses artificial intelligence algorithms to build a fault prediction model and a fault repair strategy generation model, including the following steps: S1-1: The intelligent attachment monitoring platform uses the pre-trained named entity and entity relationship extraction model to build a knowledge graph according to the attachment manual, including the following steps: S1-1-1: Intelligent attachment monitoring platform, which performs word segmentation, stop word removal, and professional terminology unification on the attachment manual to obtain the pre-processed attachment manual text data; S1-1-2: Use the named entity and entity relationship extraction model to extract named entities and entity relationships from the preprocessed attachment manual text data to obtain several knowledge named entities and several knowledge entity relationships; The named entity and entity relationship extraction model is constructed based on the Bidirectional Encoder Representations from Transformers (BERT)-Double Conditional Random Fields (CRF) algorithm, and the named entity and entity relationship extraction model includes a text feature extraction module constructed based on the BERT algorithm, a named entity extraction module constructed based on the CRF algorithm, and an entity relationship extraction module constructed based on the CRF algorithm. The text feature extraction module is connected to the named entity extraction module and the entity relationship extraction module respectively; The BERT module can capture the deep semantic information in the preprocessed attachment manual text data, which is very useful for identifying different types of entities and extracting the relationship between entities. The CRF module can consider the dependency between adjacent labels, which can help the model learn the sequence dependency of entity labels, thereby improving the accuracy of named entity labeling. In the entity relationship labeling task, the CRF module handles the relationship classification between entity pairs, especially when dealing with multiple entities and relationships, to achieve entity relationship labeling. S1-1-3: Based on a number of knowledge named entities and a number of knowledge entity relationships, a knowledge graph is constructed to obtain a corresponding knowledge graph, including the following steps: S1-1-3-1: Take the knowledge named entity as the first node of the knowledge graph, and the corresponding knowledge entity relationship as the first edge of the knowledge graph; S1-1-3-2: Construct a knowledge graph framework based on all first nodes, and improve the knowledge graph framework based on the edges between each first node and other first nodes to obtain the corresponding knowledge graph; The knowledge graph includes knowledge such as equipment structure, function, parameters, and failure mode. The knowledge named entities and knowledge entity relationships in the knowledge graph can provide rich contextual information for monitoring logs. S1-2: Collecting some historical operation monitoring data and some historical monitoring logs of the attachment, and preprocessing them to obtain some preprocessed historical operation monitoring data and some preprocessed historical monitoring logs; The preprocessing here includes cleaning the data to remove noise and irrelevant information, standardizing or normalizing the data to maintain consistency, and processing missing values ​​and outliers, which improves the quality of the data and provides data support for subsequent model construction; Operation monitoring data includes operation monitoring indicators (including current, voltage, power, etc.) and operation monitoring values ​​of accessories, and operation environment monitoring indicators (including temperature, humidity, sunshine intensity, etc.) and operation environment monitoring values ​​of accessories. Monitoring logs include platform operation logs (platform operation status, fault records, maintenance records, etc.) and platform operation logs (usually including timestamp, event type, equipment status, operator behavior, etc.) of the intelligent accessory monitoring platform. S1-3: Use the knowledge graph to perform knowledge mapping on several pre-processed historical monitoring logs to obtain several semantically enhanced historical monitoring logs, including the following steps: S1-3-1: Use the named entity and entity relationship extraction model to extract named entities and entity relationships from the preprocessed historical monitoring logs to obtain several historical log named entities (including log named entities of equipment structure, function, parameter, failure mode, etc.) and several historical log entity relationships (for example, if a certain operation may cause a certain component of the equipment to fail, then the operation log named entity and the failure mode log named entity have a strong entity relationship); S1-3-2: Map the knowledge named entities of the knowledge graph to the historical log named entities, and associate the knowledge entity relationships with the corresponding historical log named entities to obtain the initial semantically enhanced historical monitoring logs; Through mapping, the original information in the monitoring log is given additional semantic information, which comes from the structured knowledge in the knowledge graph. The original simple log entries become semantic entities and relationships rich in contextual information. In addition, the monitoring log is transformed from the original low-dimensional data to high-dimensional data with graph structure information, which improves the monitoring log's ability to represent fault conditions. S1-3-3: Delete duplicates of the historical log entity relationships and associated knowledge entity relationships of the same historical log named entity to obtain the final semantically enhanced historical monitoring log; S1-3-4: traverse all the pre-processed historical monitoring logs to obtain several final semantically enhanced historical monitoring logs; S1-4: Based on some historical operation monitoring data and some semantically enhanced historical monitoring logs, a deep learning algorithm is used to build a fault prediction model and generate some historical fault prediction results, including the following steps: S1-4-1: Randomly associate some historical operation monitoring data with some semantically enhanced historical monitoring logs, and add corresponding real labels to obtain some fault prediction samples; S1-4-2: Divide several fault prediction samples into model training set and model test set in a ratio of 7:3, and use the graph convolutional network (GCN)-long short-term memory network (LSTM)-attention-multilayer perceptron (MLP) algorithm to build an initial fault prediction model; The fault prediction model is built based on the GCN-LSTM-Attention-MLP algorithm, and the fault prediction model includes a graph structure feature extraction module built based on the GCN algorithm, a sequence feature extraction module built based on the LSTM algorithm, an attention weight module built based on the Attention mechanism, and a fault prediction module built based on the MLP algorithm; The GCN network propagates features on the semantically enhanced historical monitoring logs through operations similar to convolution, extracts node features of several log named entities and knowledge named entities in the semantically enhanced historical monitoring logs, edge features of log entity relationships and knowledge entity relationships between nodes, and forms complex, high-dimensional global graph structure features; the LSTM network can learn to retain important information in long time series and ignore unimportant information through the gate mechanism, which makes the LSTM network have significant advantages when processing sequence data. The LSTM network can capture long-term dependencies in the data and extract deep sequence features of operation monitoring data; the Attention mechanism performs weighted fusion of graph structure features and sequence features, enhances the model's attention to important features, and improves the accuracy and efficiency of model prediction; the MLP network, as a fully connected network, can perform accurate and efficient label prediction based on fused features; S1-4-3: Input the model training set, optimize the initial fault prediction model, obtain the optimized fault prediction model, and generate several historical fault prediction results; S1-4-4: Input the model test set, perform model testing on the optimized fault prediction model, and obtain the test accuracy. If the test accuracy is greater than the accuracy threshold, output the final fault prediction model; S1-5: Based on several historical fault prediction results, an enhanced reinforcement learning algorithm is used to build a fault repair strategy generation model, including the following steps: S1-5-1: Use the Improved Crested Porcupine Optimizer (ICPO)-Deep Q Network (DQN) algorithm to build an initial fault repair strategy generation model; the initial fault repair strategy generation model includes a parameter initialization module and an initial reinforcement learning module; The fault repair strategy generation model is constructed based on the ICPO-DQN algorithm, and the encryption update strategy generation model includes a parameter initialization module constructed based on the ICPO algorithm and a reinforcement learning module constructed based on the DQN algorithm. The reinforcement learning module is provided with an intelligent agent, a deep Q network and an experience replay pool. The intelligent agent is connected to the deep Q network and the experience replay pool respectively, and the parameter initialization module is connected to the deep Q network. The parameter initialization module is used to optimize the initial network parameters of the deep Q network, avoiding the defect of the deep Q network being sensitive to initial values, improving the training speed of the deep Q network, and further improving the training efficiency of the fault maintenance strategy generation model and the generation quality of the fault debugging strategy; the reinforcement learning module is an algorithm that combines deep learning and reinforcement learning. It is mainly used to solve decision-making problems with high-dimensional input space. It can generate corresponding fault maintenance strategies based on the input fault prediction results combined with the experience playback mechanism; S1-5-2: Taking minimizing the mean square error as the optimization goal, use the parameter initialization module to initialize the parameters of the deep Q network of the initial reinforcement learning module to obtain an optimized reinforcement learning module, including the following steps: S1-5-2-1: Take minimizing the mean square error value as the optimization goal, and set the parameter initialization module's fitness function according to the optimization goal; The formula is:

[0021] In the formula, is the fitness function; is the mean square error function; For the The true output value of samples; For the The ideal output value of samples; is the sample indicator amount; N is the total number of samples; S1-5-2-2: Encode the initial network parameters of the deep Q network into individual vectors of the parameter initialization module, and set the algorithm parameters of the parameter initialization module; the algorithm parameters include the ICPO population parameters and the maximum number of iterations of the ICPO algorithm; S1-5-2-3: Based on the individual vector and ICPO population parameters, the Circle chaotic mapping sequence is used to initialize the ICPO population to obtain an initial ICPO population containing several initial ICPO individuals; The formula is:

[0022] In the formula, is the initial ICPO individual of the Circle chaos map, corresponding to an initial solution; is the randomly generated initial ICPO individual; It is the ICPO individual indicator; S1-5-2-4: Introduce a cyclic population reduction mechanism to limit the number of individuals in the ICPO population parameters and obtain updated ICPO population parameters for the next iteration; The formula is:

[0023] In the formula, For the The number of individuals in the ICPO population parameter of the iteration; For the The number of individuals in the ICPO population parameter of the iteration; is the minimum number of individuals in the ICPO population parameter; Evaluate arguments for functions; Evaluate loop parameters for a function; Evaluate loop parameters for the maximum function; t is the indicator of the number of iterations; S1-5-2-5: Calculate the initial fitness value of the initial ICPO individuals in the initial ICPO population according to the fitness function; S1-5-2-6: According to the initial fitness value and the updated ICPO population parameters, the first defense strategy, the second defense strategy, the third defense strategy and the fourth defense strategy are used to update the initial ICPO population to obtain an updated ICPO population; The formula for the first defense strategy is:

[0024] In the formula, For the updated ICPO individuals within the first defense range; The initial ICPO individual within the first defense range; is a random number based on normal distribution; is a random value in the interval [0,1]; It is the optimal solution within the first defense range; is the vector generated between the true optimal solution within the first defense range and the optimal solution randomly selected from the ICPO population; It is the ICPO individual indicator; is the iteration indicator; The formula for the second defense strategy is:

[0025] In the formula, For the updated ICPO individuals within the second defense range; The initial ICPO individual within the second defense range; is the search upper limit vector of the second defense range; is a random value in the interval [0,1]; Respectively Initial ICPO individuals; All are [1, ]; is the vector generated between the true optimal solution within the second defense range and the optimal solution randomly selected from the ICPO population; The formula for the third defense strategy is:

[0026] In the formula, For the updated ICPO individuals within the third defense range; The initial ICPO individual within the third defense range; is the search upper limit vector of the third defense range; Respectively Initial ICPO individuals; is [1, ]; is the odor diffusion factor; It is a defense factor; Control parameters for search direction; The formula for the fourth defense strategy is:

[0027] In the formula, For the updated ICPO individuals within the fourth defense range; The initial ICPO individual within the fourth defense range; It is the optimal solution within the fourth defense range; All are random values ​​in the interval [0,1]; It is a defense factor; Control parameters for search direction; is the average force affecting the search direction; is the convergence speed factor;

[0028] In the formula, is the convergence speed factor; tanh(.) is the hyperbolic tangent function; a max , a minare the maximum and minimum values ​​of the convergence factor, respectively; λ is the deceleration rate parameter, k is the decreasing cycle parameter, λ =-2 π , k = π ; is the iteration indicator; is the maximum number of iterations; S1-5-2-7: Use the dynamic reverse learning algorithm to perform dynamic reverse learning on the updated ICPO population to generate a dynamic reverse ICPO population; The formula is:

[0029] In the formula, It is a dynamically reversed ICPO individual; is the decreasing inertia coefficient; are the maximum and minimum values ​​of the vector space respectively; For the updated ICPO individuals; S1-5-2-8: According to the fitness function, calculate the fitness values ​​of all ICPO individuals in the updated ICPO population and the dynamically reversed ICPO population, take the ICPO individual with the minimum fitness value as the optimal individual, and retain the optimal individual; S1-5-2-9: If the number of iterations reaches the maximum number of iterations or the fitness value of the optimal individual meets the requirements, the individual encoding vector of the optimal individual is decoded to obtain the optimal initial network parameters of the deep Q network; The initial network parameters include the neuron data of the deep Q network, the initial neuron weights, the initial neuron threshold, and the learning rate. S1-5-2-10: Initialize the parameters of the deep Q network of the initial reinforcement learning module according to the optimal initial network parameters to obtain an optimized reinforcement learning module; S1-5-3: Based on several historical fault prediction states of historical fault prediction results, define the state space of the optimized reinforcement learning module, and based on several preset fault repair actions, define the action space of the optimized reinforcement learning module; S1-5-4: Define the reward function of the optimized reinforcement learning module based on the impact of the preset fault repair action on the historical fault prediction status; S1-5-5: Based on the state space, action space and reward function, several historical fault prediction results are input, the intelligent agent of the optimized reinforcement learning module is trained, the final reinforcement learning module is obtained, and several historical fault maintenance strategy generation experiences are generated; S1-5-6: Integrate the parameter initialization module and the final reinforcement learning module to obtain the final fault repair strategy generation model, and store several historical fault repair strategy generation experiences in the experience replay pool; S2: Intelligent attachment monitoring platform collects the real-time operation monitoring data and real-time monitoring logs of attachments, and uses knowledge graphs to perform knowledge mapping on real-time monitoring logs to obtain semantically enhanced real-time monitoring logs, including the following steps: S2-1: collect the real-time operation monitoring data and real-time monitoring log of the attachment, and perform preprocessing to obtain the preprocessed real-time operation monitoring data and the preprocessed real-time monitoring log; The preprocessing here includes normalizing the real-time operation monitoring data, mapping the real-time operation monitoring data to the magnitude of 0-1, eliminating the data magnitude difference, and converting the format of the real-time monitoring log to generate a data format that can be recognized by the subsequent model; S2-2: Use the named entity and entity relationship extraction model to extract named entities and entity relationships from the preprocessed real-time monitoring logs to obtain a number of real-time log named entities and a number of real-time log entity relationships; S2-3: Map the knowledge named entities of the knowledge graph to the real-time log named entities, and associate the knowledge entity relationships with the corresponding real-time log named entities to obtain the initial semantically enhanced real-time monitoring logs; S2-4: De-duplicate the real-time log entity relationship and the associated knowledge entity relationship of the same real-time log named entity to obtain the final semantically enhanced historical monitoring log; S3: The intelligent attachment monitoring platform uses the fault prediction model to perform fault prediction based on the real-time operation monitoring data and the real-time monitoring log after semantic enhancement, and obtains the real-time fault prediction results, including the following steps: S3-1: Intelligent attachment monitoring platform, which inputs real-time operation monitoring data and semantically enhanced real-time monitoring logs into the fault prediction model; S3-2: Use the graph structure feature extraction module to extract the real-time graph structure features of the real-time monitoring log after semantic enhancement, and use the sequence feature extraction module to extract the real-time sequence features of the real-time operation monitoring data; S3-3: According to the preset attention weight value, use the attention weight module to perform weighted fusion on the real-time graph structure features and the real-time sequence features to obtain the real-time weighted fusion features; S3-4: Based on the real-time weighted fusion features, a fault prediction module is used to perform fault prediction and obtain a real-time fault prediction result; Fault prediction results include fault probability prediction results (the probability or confidence of each possible fault, relative risk score of fault occurrence, etc.), fault type prediction results (such as mechanical failure, electrical failure, software failure of accessories, etc.), fault location prediction results (predicting the specific location or component where the fault may occur, etc.), fault time prediction results (predicting the time point or time period when the fault may occur, remaining service life, etc.) and fault impact prediction results (the degree of impact of the fault on system performance, the safety risk and economic loss assessment caused by the fault, etc.); S4: The intelligent accessory monitoring platform generates a fault repair strategy based on the real-time fault prediction result using a fault repair strategy generation model to obtain a real-time fault repair strategy, including the following steps: S4-1: The intelligent accessory monitoring platform analyzes the real-time fault prediction results to obtain a number of real-time fault prediction states, and updates the state space of the fault maintenance strategy generation model according to the real-time fault prediction states to obtain an updated state space; The fault prediction status includes the fault probability prediction status, the fault type prediction status, the fault location prediction status, the fault time prediction status and the fault impact prediction status; In this embodiment, possible specific status values ​​are provided, the fault probability prediction status is "50%", the fault type prediction status is "1", indicating that a mechanical fault has occurred in the accessory, the fault location prediction status is "XXX-YYY-S", indicating the specific location and specific accessory number of the mechanical fault of the accessory, the fault time prediction status is "20", indicating that the service life is 20 days remaining, and the fault impact prediction status is "1", indicating that the fault impact is small; S4-2: According to the real-time fault prediction result, the experience playback pool of the fault repair strategy generation model is searched to obtain the real-time fault repair strategy generation experience that matches successfully, and according to the real-time fault repair strategy generation experience, the action space of the fault repair strategy generation model is updated to obtain an updated action space; S4-3: Connect the updated state space to the input layer of the deep Q network of the fault repair strategy generation model, and connect the updated action space to the output layer of the deep Q network of the fault repair strategy generation model to obtain an updated deep Q network; S4-4: according to the reward function, obtaining the reward value of each real-time fault repair alternative action in the updated action space for each real-time fault prediction state in the updated state space, and using the intelligent agent to control the updated deep Q network to generate the Q value of each real-time fault repair alternative action in the updated action space; Possible / implemented actions for fault repair include diagnostic inspection actions (detection of fault location, regular inspection, etc.), component replacement actions (replacement of components or assemblies that are predicted to fail, preparation of replacement components for quick replacement, etc.), parameter adjustment actions (modification of operating parameters of equipment or accessories to mitigate the impact of failure, reconfiguration of system settings to optimize performance, etc.), software update actions (installation of software patches or updates to resolve predicted software-related failures, resetting or reprogramming of control systems, etc.), cleaning and maintenance actions (cleaning equipment, removing dirt or blockages that may cause failures, performing routine maintenance operations such as lubrication, tightening bolts, etc.), and repair or recalibration actions (repair of damaged components, recalibration of sensors or instruments to ensure accurate readings, etc.); S4-5: According to the reward value, use the intelligent agent to iteratively update the Q value of the real-time fault maintenance alternative action to obtain an updated Q value, until the number of iterations reaches the iteration number threshold, and use the greedy strategy to select the real-time fault maintenance alternative action with the highest updated Q value as the real-time fault maintenance execution action, and integrate all real-time fault maintenance execution actions to obtain the real-time fault maintenance strategy; S5: Intelligent accessory monitoring platform, which visualizes the real-time fault prediction results and real-time fault repair strategies of accessories.

[0030] The present invention provides a fault prediction method and system for an intelligent accessory monitoring platform. The method uses an artificial intelligence algorithm to build a fault prediction model, which can more accurately identify and predict potential fault conditions and maintain a high prediction accuracy even in a complex and changeable industrial environment. The method combines real-time operation monitoring data and real-time monitoring logs to improve the comprehensiveness of fault prediction data, extract more valuable fault information, and perform knowledge mapping on real-time monitoring logs, which can more accurately capture the correlation between accessories and fault propagation paths, further improving the accuracy and efficiency of fault prediction. The method performs automated and efficient data processing based on artificial intelligence models, which can quickly process a large amount of real-time data and ensure the real-time nature of fault prediction. A more comprehensive and in-depth knowledge graph is built through the accessory manual, which makes full use of domain knowledge and expert experience in the accessory manual, making fault prediction and maintenance strategy generation more accurate and targeted. The method uses an artificial intelligence algorithm to build a fault maintenance strategy generation model, which can intelligently generate fault maintenance strategies according to fault prediction conditions and provide optimal maintenance plans, thereby improving the degree of intelligence and realizing a fault prediction response mechanism.

[0031] Embodiment 2: like Figure 2As shown, this embodiment provides a fault prediction system for an intelligent accessory monitoring platform, which is used to implement a fault prediction method for an intelligent accessory monitoring platform. The system is applied to the intelligent accessory monitoring platform, and the system includes an initialization unit, a knowledge mapping unit, a fault prediction unit, a fault repair strategy generation unit, and a visualization unit connected in sequence; An initialization unit is used to construct a knowledge graph based on the accessory manual and use an artificial intelligence algorithm to build a fault prediction model and a fault repair strategy generation model; The knowledge mapping unit is used to collect the real-time operation monitoring data and real-time monitoring log of the attachment, and use the knowledge graph to perform knowledge mapping on the real-time monitoring log to obtain the real-time monitoring log after semantic enhancement; A fault prediction unit is used to perform fault prediction based on real-time operation monitoring data and real-time monitoring logs after semantic enhancement using a fault prediction model to obtain real-time fault prediction results; A fault repair strategy generation unit is used to generate a fault repair strategy based on the real-time fault prediction result using a fault repair strategy generation model to obtain a real-time fault repair strategy; The visualization unit is used to visualize the real-time fault prediction results and real-time fault repair strategies of the accessories.

[0032] The present invention is not limited to the above optional implementations, and anyone can derive other various forms of products under the enlightenment of the present invention. The above specific implementations should not be understood as limiting the scope of protection of the present invention. The scope of protection of the present invention should be based on the definition in the claims, and the description can be used to interpret the claims.

Claims

1. A fault prediction method for an intelligent accessory monitoring platform, characterized in that: The steps include: The intelligent accessory monitoring platform builds a knowledge graph based on the accessory manual and uses artificial intelligence algorithms to build a fault prediction model and a fault repair strategy generation model; The intelligent attachment monitoring platform collects the real-time operation monitoring data and real-time monitoring logs of attachments, and uses the knowledge graph to perform knowledge mapping on the real-time monitoring logs to obtain semantically enhanced real-time monitoring logs; The intelligent attachment monitoring platform uses the fault prediction model to perform fault prediction based on real-time operation monitoring data and semantically enhanced real-time monitoring logs, and obtains real-time fault prediction results; The intelligent accessory monitoring platform generates fault repair strategies based on real-time fault prediction results and uses the fault repair strategy generation model to obtain real-time fault repair strategies; The intelligent accessory monitoring platform visualizes the real-time fault prediction results and real-time fault repair strategies of accessories.

2. The fault prediction method of the intelligent accessory monitoring platform according to claim 1, characterized in that: The intelligent accessory monitoring platform builds a knowledge graph based on the accessory manual and uses artificial intelligence algorithms to build a fault prediction model and a fault repair strategy generation model, including the following steps: The intelligent attachment monitoring platform builds a knowledge graph based on the attachment manual using pre-trained named entity and entity relationship extraction models; Collecting a number of historical operation monitoring data and a number of historical monitoring logs of the attachment, and preprocessing them to obtain a number of preprocessed historical operation monitoring data and a number of preprocessed historical monitoring logs; Use the knowledge graph to perform knowledge mapping on several pre-processed historical monitoring logs to obtain several semantically enhanced historical monitoring logs; Based on some historical operation monitoring data and some semantically enhanced historical monitoring logs, a deep learning algorithm is used to build a fault prediction model and generate some historical fault prediction results. Based on several historical fault prediction results, an enhanced reinforcement learning algorithm is used to build a fault maintenance strategy generation model.

3. The fault prediction method of the intelligent accessory monitoring platform according to claim 2, characterized in that: The named entity and entity relationship extraction model is constructed based on the BERT-Double CRF algorithm, and the named entity and entity relationship extraction model includes a text feature extraction module constructed based on the BERT algorithm, a named entity extraction module constructed based on the CRF algorithm, and an entity relationship extraction module constructed based on the CRF algorithm, and the text feature extraction module is connected to the named entity extraction module and the entity relationship extraction module respectively; The fault prediction model is constructed based on the GCN-LSTM-Attention-MLP algorithm, and the fault prediction model includes a graph structure feature extraction module constructed based on the GCN algorithm, a sequence feature extraction module constructed based on the LSTM algorithm, an attention weight module constructed based on the Attention mechanism, and a fault prediction module constructed based on the MLP algorithm; The fault repair strategy generation model is constructed based on the ICPO-DQN algorithm, and the encryption update strategy generation model includes a parameter initialization module constructed based on the ICPO algorithm and a reinforcement learning module constructed based on the DQN algorithm. The reinforcement learning module is provided with an intelligent agent, a deep Q network and an experience replay pool. The intelligent agent is connected to the deep Q network and the experience replay pool respectively, and the parameter initialization module is connected to the deep Q network.

4. The fault prediction method of the intelligent accessory monitoring platform according to claim 3, characterized in that: The intelligent attachment monitoring platform uses the pre-trained named entity and entity relationship extraction model to build a knowledge graph according to the attachment manual, including the following steps: The intelligent attachment monitoring platform performs word segmentation, stop word removal, and professional terminology unification on the attachment manual to obtain the pre-processed attachment manual text data; Using the named entity and entity relationship extraction model, named entity and entity relationship extraction is performed on the preprocessed attachment manual text data to obtain several knowledge named entities and several knowledge entity relationships; According to several knowledge named entities and several knowledge entity relationships, a knowledge graph is constructed to obtain the corresponding knowledge graph.

5. The fault prediction method of the intelligent accessory monitoring platform according to claim 3, characterized in that: Using the knowledge graph, knowledge mapping is performed on several pre-processed historical monitoring logs to obtain several semantically enhanced historical monitoring logs, including the following steps: Use the named entity and entity relationship extraction model to extract named entities and entity relationships from the preprocessed historical monitoring logs to obtain several historical log named entities and several historical log entity relationships; Map the knowledge named entities of the knowledge graph to the historical log named entities, and associate the knowledge entity relationships with the corresponding historical log named entities to obtain the initial semantically enhanced historical monitoring logs; The historical log entity relationships and associated knowledge entity relationships of the same historical log named entity are deleted repeatedly to obtain the final semantically enhanced historical monitoring log; Traverse all the pre-processed historical monitoring logs to obtain several final semantically enhanced historical monitoring logs.

6. The fault prediction method of the intelligent accessory monitoring platform according to claim 3, characterized in that: Based on some historical operation monitoring data and some semantically enhanced historical monitoring logs, a deep learning algorithm is used to build a fault prediction model and generate some historical fault prediction results, including the following steps: Randomly associate some historical operation monitoring data with some semantically enhanced historical monitoring logs, and add corresponding real labels to obtain some fault prediction samples; Several fault prediction samples are divided into model training set and model test set in a ratio of 7:3, and the GCN-LSTM-Attention-MLP algorithm is used to build the initial fault prediction model; Input the model training set, optimize the initial fault prediction model, obtain the optimized fault prediction model, and generate several historical fault prediction results; The model test set is input, and the optimized fault prediction model is tested to obtain the test accuracy. If the test accuracy is greater than the accuracy threshold, the final fault prediction model is output.

7. The fault prediction method of the intelligent accessory monitoring platform according to claim 3, characterized in that: Based on several historical fault prediction results, an enhanced reinforcement learning algorithm is used to build a fault repair strategy generation model, which includes the following steps: Use the ICPO-DQN algorithm to build an initial fault repair strategy generation model; the initial fault repair strategy generation model includes a parameter initialization module and an initial reinforcement learning module; Taking minimizing the mean square error as the optimization goal, the parameter initialization module is used to initialize the parameters of the deep Q network of the initial reinforcement learning module to obtain an optimized reinforcement learning module; According to several historical fault prediction states of historical fault prediction results, a state space of an optimized reinforcement learning module is defined, and according to several preset fault repair actions, an action space of an optimized reinforcement learning module is defined; Define the reward function of the optimized reinforcement learning module based on the impact of the preset fault repair action on the historical fault prediction status; Based on the state space, action space and reward function, several historical fault prediction results are input, the intelligent agent of the optimized reinforcement learning module is trained, the final reinforcement learning module is obtained, and several historical fault maintenance strategy generation experiences are generated; The parameter initialization module and the final reinforcement learning module are integrated to obtain the final fault repair strategy generation model, and several historical fault repair strategy generation experiences are stored in the experience replay pool.

8. The fault prediction method of the intelligent accessory monitoring platform according to claim 6, characterized in that: The intelligent attachment monitoring platform uses the fault prediction model to perform fault prediction based on the real-time operation monitoring data and the real-time monitoring log after semantic enhancement, and obtains the real-time fault prediction results, including the following steps: Intelligent attachment monitoring platform, which inputs real-time operation monitoring data and semantically enhanced real-time monitoring logs into the fault prediction model; Use the graph structure feature extraction module to extract the real-time graph structure features of the real-time monitoring log after semantic enhancement, and use the sequence feature extraction module to extract the real-time sequence features of the real-time operation monitoring data; According to the preset attention weight value, the attention weight module is used to perform weighted fusion on the real-time graph structure features and the real-time sequence features to obtain the real-time weighted fusion features; According to the real-time weighted fusion features, the fault prediction module is used to perform fault prediction and obtain real-time fault prediction results.

9. The fault prediction method of the intelligent accessory monitoring platform according to claim 7, characterized in that: The intelligent accessory monitoring platform generates a fault repair strategy based on the real-time fault prediction results using a fault repair strategy generation model to obtain a real-time fault repair strategy, including the following steps: The intelligent accessory monitoring platform analyzes the real-time fault prediction results to obtain a number of real-time fault prediction states, and updates the state space of the fault maintenance strategy generation model according to the real-time fault prediction states to obtain an updated state space; According to the real-time fault prediction results, the experience playback pool of the fault maintenance strategy generation model is searched to obtain the real-time fault maintenance strategy generation experience that matches successfully, and according to the real-time fault maintenance strategy generation experience, the action space of the fault maintenance strategy generation model is updated to obtain an updated action space; Connecting the updated state space to the input layer of the deep Q network of the fault repair strategy generation model, and connecting the updated action space to the output layer of the deep Q network of the fault repair strategy generation model to obtain an updated deep Q network; According to the reward function, the reward value of each real-time fault repair alternative action in the updated action space for each real-time fault prediction state in the updated state space is obtained, and the intelligent agent is used to control the updated deep Q network to generate the Q value of each real-time fault repair alternative action in the updated action space; According to the reward value, the intelligent agent is used to iteratively update the Q value of the real-time fault repair alternative action to obtain an updated Q value until the number of iterations reaches the iteration number threshold, and a greedy strategy is used to select the real-time fault repair alternative action with the highest updated Q value as the real-time fault repair execution action, and all real-time fault repair execution actions are integrated to obtain the real-time fault repair strategy.

10. A fault prediction system for an intelligent accessory monitoring platform, used to implement the fault prediction method for an intelligent accessory monitoring platform as claimed in any one of claims 1 to 9, characterized in that: The system is applied to an intelligent accessory monitoring platform, and the system comprises an initialization unit, a knowledge mapping unit, a fault prediction unit, a fault repair strategy generation unit and a visualization unit which are connected in sequence.

Images