Intelligent operation and maintenance large model and public large model connection method

Abstract

The invention discloses a method for connecting an intelligent operation and maintenance large model and a public large model, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining equipment operation data from a sensor node through a preset data collection frame, and carrying out the preprocessing of the equipment operation data, thereby obtaining a first data set; performing abnormal state recognition on the first data set by using a convolutional neural network and an SVM classifier, extracting an abnormal data subset, and clustering to obtain an abnormal mode; acquiring historical equipment operation parameters and environmental parameters, and performing correlation analysis on the abnormal state, the historical equipment operation parameters and the environmental parameters by adopting a Bayesian network to obtain a potential reason set of the abnormality; constructing an equipment fault knowledge graph to perform path reasoning on reasons in the potential reason set, and generating a fault association mapping table; and a multi-objective optimization function is established based on the fault association mapping table and the real-time state data of the equipment, and an optimal disposal scheme combination is solved, so that the intelligent level of equipment fault diagnosis and operation and maintenance operation optimization is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a method for connecting a smart operation large model and a public large model. BACKGROUND

[0002] In the field of modern industry and equipment management, intelligent operation is considered as the core pillar to improve production efficiency and ensure equipment safety, and its importance is self-evident. With the increasing complexity of industrial equipment and the explosive growth of data, how to achieve precise monitoring and rapid response of equipment status through intelligent means has become a key direction for industry development. However, many current operation solutions still have significant limitations, mainly relying on manual experience for equipment anomaly detection and processing, lacking systematic analysis and optimization capabilities, resulting in slow fault response speed, low disposal efficiency, and difficulty in meeting the needs of complex scenarios.

[0003] Under this background, the field faces many challenges. The first challenge is how to quickly and accurately identify abnormal states from massive equipment operation data. Due to the diversity of data sources and the presence of noise, relying solely on traditional methods often fails to capture subtle abnormalities, leading to misjudgment or missed judgment. This problem further extends to the cause analysis after the occurrence of anomalies, as the lack of deep mining of equipment operation mechanisms and historical data often results in inaccurate root cause positioning, making subsequent treatment schemes difficult to address the problem. A deeper challenge is that even if a treatment scheme is generated, how to efficiently execute it in a complex on-site environment, especially in personnel scheduling and path planning, existing methods often ignore the dynamic matching of personnel skills and equipment distribution, resulting in resource waste and time delay.

[0004] Therefore, how to realize the intelligentization of the whole process from anomaly detection to root cause analysis, to treatment scheme generation and execution path optimization in equipment operation, has become a key problem to be solved. SUMMARY

[0005] The present application aims to at least solve one of the technical problems in the related art. To this end, one object of the present application is to provide a method for connecting a smart operation large model and a public large model, which improves the intelligent level of equipment fault diagnosis and operation optimization.

[0006] One aspect of the present application provides a method for connecting a smart operation large model and a public large model, comprising:

[0007] Step S100: obtaining equipment operation data from sensor nodes through a preset data acquisition framework, and pre-processing the equipment operation data to obtain a first data set;

[0008] Step S200: abnormal state recognition is performed on the first data set by using a convolutional neural network and an SVM classifier, an abnormal data subset is extracted, and an abnormal pattern is obtained by clustering;

[0009] Step S300: historical equipment operation parameters and environmental parameters are obtained, and a Bayesian network is used to perform correlation analysis on the abnormal state and the historical equipment operation parameters and environmental parameters, to obtain an abnormal potential cause set;

[0010] Step S400: a device fault knowledge graph is constructed to perform path reasoning on the causes in the potential cause set, and a fault correlation mapping table is generated;

[0011] Step S500: a multi-objective optimization function is established based on the fault correlation mapping table and real-time state data of the equipment, and an optimal treatment scheme combination is solved;

[0012] The specific method for obtaining the equipment operation data from the sensor nodes through the preset data acquisition framework and preprocessing the equipment operation data to obtain the first data set is as follows:

[0013] The specific method for obtaining the equipment operation parameters from N sensor nodes through the preset data acquisition framework, performing data cleaning on the original equipment operation parameters to obtain a cleaned data set, removing noise from the cleaned data set by using a time series smoothing algorithm to obtain a denoised first data set, and the first data set being composed of data samples is as follows:

[0014] The specific method for performing abnormal state recognition on the first data set by using a convolutional neural network and an SVM classifier, extracting an abnormal data subset, and obtaining an abnormal pattern by clustering is as follows:

[0015] Step S210: the denoised first data set is standardized by data preprocessing to obtain standardized data;

[0016] Step S220: according to the standardized data, a convolutional neural network model is used to perform feature extraction operation to extract a feature vector from the standardized data;

[0017] Step S230: the feature vector is input into a pre-trained SVM classifier to calculate an abnormal probability value corresponding to each data sample in the first data set;

[0018] Step S240: if the abnormal probability value is greater than a preset threshold, the data sample is determined to be in an abnormal state, and if the abnormal probability value is less than or equal to the preset threshold, the data sample is determined to be in a normal state, to obtain a state determination result;

[0019] Step S250: the data samples in the first data set are classified and screened by using the state determination result, the data samples corresponding to all abnormal states are taken as abnormal samples to form an abnormal data subset;

[0020] Step S260: clustering the abnormal samples in the abnormal data subset using the K-means clustering algorithm to obtain K different types of abnormal patterns;

[0021] The specific method of calculating the abnormal probability value corresponding to each data sample in the first data set by inputting the feature vector into the pre-trained SVM classifier is as follows: taking the feature vector extracted by the convolutional neural network model as the input of the SVM classifier, the label of the normal sample is 1, and the label of the abnormal sample is -1; the optimization objective of the SVM classifier is defined as finding a maximum interval hyperplane so that different categories of data samples can be correctly classified, the optimization objective of the SVM is solved to obtain the optimal Lagrange multiplier, and thus a trained SVM classifier is obtained; using the trained SVM classifier, for each data sample in the first data set, the distance of the data sample to the hyperplane is calculated, and the distance is mapped to an abnormal probability value through a sigmoid function;

[0022] The specific method of obtaining the historical equipment operation parameters and environment parameters, using the Bayesian network to perform correlation analysis on the abnormal state and the historical equipment operation parameters and environment parameters, and obtaining the set of potential causes of the abnormality is as follows:

[0023] Step S310: obtaining the historical equipment operation parameters and environment parameters from the database;

[0024] Step S320: constructing a Bayesian network, establishing the conditional probability relationship between the historical equipment operation parameters, environment parameters and abnormal state, and determining the causal relationship and probability distribution parameters between nodes;

[0025] Step S330: updating the conditional probability distribution of the Bayesian network using the abnormal data subset;

[0026] Step S340: for each abnormal sample in the abnormal data subset, calculating the posterior probability of the state node corresponding to the abnormal sample in the root cause analysis model;

[0027] Step S350: selecting the state node with the maximum posterior probability as the potential cause of the abnormality, and adding it to the set of potential causes;

[0028] The specific method for constructing a Bayesian network, establishing a conditional probability relationship between historical equipment operating parameters, environmental parameters and abnormal states, and determining the causal relationship and probability distribution parameters between each node is as follows: selecting variables related to the abnormal state of the equipment, the variables including historical equipment operating parameters, environmental parameters and abnormal states, determining the causal relationship between each variable based on expert knowledge, and determining the topological structure of the Bayesian network; initializing the probability distribution parameters of each node in the Bayesian network using a uniform distribution method; using the historical equipment operating parameters and environmental parameters, estimating the conditional probability distribution parameters of each node in the Bayesian network, wherein the conditional probability distribution parameters are estimated using a Bayesian estimation method;

[0029] The specific method of using the abnormal data subset to update the conditional probability distribution of the Bayesian network is as follows: using the abnormal samples in the abnormal data subset, merging them with historical abnormal samples to form a training data set, re-estimating the conditional probability distribution parameters of the Bayesian network, and obtaining an updated Bayesian network; for each new abnormal sample received, finding its corresponding state node in the Bayesian network, updating the conditional probability distribution parameters of the state node, and updating the conditional probability distribution parameters of the parent node of the state node;

[0030] The specific method of constructing the equipment fault knowledge graph to perform path reasoning on the causes in the potential cause set and generate the fault association mapping table is as follows:

[0031] Step S410: searching for a corresponding fault cause node in the equipment fault knowledge graph according to the cause in the potential cause set;

[0032] Step S420: Starting from the fault cause node, a breadth-first search algorithm is used to traverse the equipment fault knowledge graph to find the set of equipment component nodes directly connected to the fault cause node;

[0033] Step S430: For each component node in the device component node set, calculate its correlation with the fault cause node. If the correlation is greater than a preset threshold, mark the component node as a high-risk component.

[0034] Step S440: Obtain a parameter node set associated with a high-risk component;

[0035] Step S450: Calculate the support and confidence of abnormal states, high-risk components, and parameter nodes using an association rule mining algorithm;

[0036] Step S460: Generate a fault correlation mapping table between high-risk components, fault causes, parameters, correlations, and confidence levels;

[0037] The definition method of the device fault knowledge graph is: taking the fault reason, the device component and the parameter as a node, the parameter includes the device operation parameter and the environment parameter; taking the causal relationship between the nodes as an edge, the device obstacle knowledge graph is constructed;

[0038] The specific method of taking the fault reason node as the starting point and using the breadth-first search algorithm to traverse the device fault knowledge graph to find the device component node set directly connected with the fault reason node is:

[0039] Step S421: selecting the fault reason node as the starting point, adding it to the queue, and initializing the access node set as empty, which is used to record the accessed nodes;

[0040] Step S422: when the queue is not empty, the following steps are repeated: removing the head node of the queue, marking the node as accessed, and adding it to the access node set; traversing all neighbor nodes of the head node of the queue, if the neighbor node has not been accessed, adding the neighbor node to the queue;

[0041] Step S423: after the search is completed, the nodes in the access node set are the device component node set directly connected with the fault reason node;

[0042] The acquisition method of the parameter node set is: for the high-risk component node, finding the parameter node set directly connected with it through the adjacency relationship in the device fault knowledge graph; finding the nodes in the device obstacle knowledge graph that have a causal relationship with the high-risk component node, and adding them to the parameter node set.

[0043] The association rule mining algorithm uses the Apriori algorithm, converts the abnormal state node, the high-risk component node and the parameter node into transactional data, mines the frequent item set from the transactional data, calculates the support degree of each item set, and takes the item set with the support degree greater than or equal to the support degree threshold as the frequent item set; and calculates the confidence degree of each frequent item set;

[0044] The specific method of establishing a multi-objective optimization function based on the fault association mapping table and the device real-time state data to solve the optimal treatment scheme combination is:

[0045] Step S510: extracting the association relationship data from the fault association mapping table, and fusing it with the device real-time state data to form the input of the optimization problem;

[0046] Specifically, the device component, the fault reason, the association degree and the confidence degree are extracted from the fault association mapping table to form an association relationship data set; the device real-time state data at the current time is obtained from the device monitoring system, and the association relationship data and the device real-time state data are left-joined as the input of the optimization problem; the data after left-joining contains the device component, the fault reason, the association degree, the confidence degree and the device real-time state data;

[0047] Step S520: Establish a multi-objective optimization function with the optimization goal of minimizing treatment time and minimizing resource cost, and the decision variable of treatment scheme;

[0048] Step S530: Optimize and solve the treatment scheme by using a non-dominated sorting genetic algorithm to obtain an optimal treatment scheme combination;

[0049] Step S540: For each treatment scheme in the optimal treatment scheme combination, verify whether it meets the actual constraint condition, and if it meets the actual constraint condition, add it to the executable scheme set;

[0050] The actual constraint condition includes: each device component can only select one treatment scheme; and the usage of each resource does not exceed the maximum usage;

[0051] Step S550: Select the treatment scheme with the optimal comprehensive benefit from the executable scheme set as the final scheme.

[0052] One aspect of the present application provides a smart operation and maintenance large model and public large model coupling system, comprising:

[0053] A data preprocessing module is configured to obtain device operation data from a sensor node through a preset data acquisition framework, and preprocess the device operation data to obtain a first data set;

[0054] An abnormal state recognition module is configured to recognize an abnormal state of the first data set by using a convolutional neural network and an SVM classifier, extract an abnormal data subset, and cluster the abnormal data subset to obtain an abnormal pattern;

[0055] An abnormal reason analysis module is configured to obtain historical device operation parameters and environmental parameters, and perform correlation analysis on the abnormal state and the historical device operation parameters and the environmental parameters by using a Bayesian network to obtain a potential cause set of the abnormality;

[0056] A fault correlation reasoning module is configured to construct a device fault knowledge graph to perform path reasoning on the causes in the potential cause set, and generate a fault correlation mapping table;

[0057] An optimal scheme optimization module is configured to establish a multi-objective optimization function based on the fault correlation mapping table and real-time state data of the device, and solve an optimal treatment scheme combination.

[0058] The smart operation and maintenance large model and public large model coupling method provided by the present application has the following advantages compared with the prior art: BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 A method flowchart of the smart operation and maintenance large model and public large model coupling method provided by the present application;

[0060] Figure 2 Abnormal state identification method flowchart provided for the present application;

[0061] Figure 3 Abnormal reason analysis method flowchart provided for the present application;

[0062] Figure 4 Functional module diagram of the smart operation and maintenance large model and public large model connection system provided for the present application. DETAILED DESCRIPTION

[0063] For a better understanding of the present application, various aspects of the present application will be described in more detail with reference to the accompanying drawings. It is to be understood that the detailed description is merely descriptive of illustrative embodiments of the present application and is not intended to limit the scope of the present application in any way. Throughout the description, like reference numerals refer to like elements. The expression “and / or” includes any and all combinations of one or more of the associated listed items.

[0064] In the drawings, the size, dimensions, and shapes of the elements have been slightly adjusted for ease of illustration. The drawings are merely examples and are not drawn to scale. As used in this document, the terms “approximately,” “about,” and similar terms are used as terms of approximation and not as terms of degree, and are intended to account for the inherent deviations in a measuring or computing process. Additionally, in the present application, the order of the steps of the process described does not necessarily indicate the order in which the processes occur in actual operation, unless there is an explicit other limitation or it can be derived from the context.

[0065] It should also be understood that expressions such as “include”, “including”, “have”, “has”, “contain” and / or “containing” and the like, are open-ended terms that are used to indicate the presence of something, but do not exclude the presence of one or more other features, elements, components and / or combinations thereof. In addition, when expressions such as “at least one of” appear after a list of items, it modifies the entire list of items and does not merely modify the individual items in the list. Furthermore, when describing embodiments of the present application, the use of “may” indicates that one or more embodiments of the present application. Also, the term “exemplary” is intended to refer to an example or illustration.

[0066] Unless otherwise defined, all terms used in this document, including engineering and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. It should also be understood that, unless otherwise explicitly stated in the present application, words defined in common dictionaries should be interpreted as having meanings consistent with their meanings in the context of the relevant technology, and should not be interpreted in an idealized or overly formal sense.

[0067] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0068] Example 1

[0069] like Figure 1 As shown, the method for connecting the intelligent operation and maintenance big model and the public big model provided in this application includes:

[0070] Step S100: acquiring device operation data from sensor nodes through a preset data acquisition framework, and preprocessing the data to obtain a first data set;

[0071] The specific method of obtaining the device operation data from the sensor node through the preset data acquisition framework and preprocessing the data to obtain the first data set is:

[0072] Acquiring device operating parameters from N sensor nodes through a preset data acquisition framework, performing data cleaning on the original device operating parameters to obtain a cleaned data set, and removing noise from the cleaned data set using a time series smoothing algorithm to obtain a denoised first data set, the first data set consisting of data samples;

[0073] The equipment operating parameters can be expressed as X raw =x1,x2,...,x N , where x i represents the equipment operation data collected by the i-th sensor node;

[0074] The calculation formula for removing noise from the cleaned data set using the time series smoothing algorithm is: smooth (t) = a × x (t) + (1-a) × x smooth (t-1); where a is the smoothing coefficient, x(t) is the equipment operation data at time t in the cleaned data set, and x smooth (t-1) represents the smoothed equipment operation data at time t-1;

[0075] Exemplarily, firstly, through a data acquisition framework based on an Internet of Things architecture, device operation parameters are acquired in real time from 32 nodes of temperature sensors, vibration sensors, pressure sensors and the like distributed on a production line, a sampling frequency is set to 100 Hz, and 3200 data points are generated per second. The original data stream contains multi-dimensional parameters such as a temperature range of 25.3-89.7℃, a vibration amplitude of 0.02-2.15 mm, and a pressure value of 1.2-8.9 MPa. Due to the limitations of sensor accuracy and environmental electromagnetic interference, the original data has random noise fluctuations of ±0.5%. The system automatically starts a data preprocessing module, firstly performs outlier detection, identifies 147 outliers that are out of the normal range of 3 standard deviations, accounting for 0.046% of the total data amount, and automatically removes these abnormal data points through the Laplace criterion. Next, a moving average filtering algorithm is applied for time series smoothing processing, a window length of 15 sampling points is set, the weighted average value of the previous and subsequent 7 points is calculated for each data point, the weight coefficients are distributed in a Gaussian distribution mode, the center point weight is 0.25, and the weight decreases to 0.05 to both sides. After filtering, the standard deviation of the temperature data is reduced from 1.23℃ to 0.31℃, the vibration data noise amplitude is reduced by 68.2%, and the signal-to-noise ratio of the pressure data is improved to 42.7 dB. Finally, a first data set containing 30720 valid data points is formed, the data integrity reaches 99.54%, and a high-quality basic data source is provided for subsequent analysis.

[0076] Step S200: performing abnormal state recognition on the first data set by using a convolutional neural network and an SVM classifier, extracting an abnormal data subset, and clustering to obtain an abnormal pattern;

[0077] As shown in Figure 2 , the specific method of performing abnormal state recognition on the first data set by using a convolutional neural network and an SVM classifier, extracting an abnormal data subset, and clustering to obtain an abnormal pattern is as follows:

[0078] Step S210: performing standardization processing on the first data set after denoising through data preprocessing to obtain standardized data;

[0079] Step S220: performing feature extraction operation on the standardized data by using a convolutional neural network model to extract a feature vector from the standardized data;

[0080] The input of the convolutional neural network is standardized data, and the output is a feature vector of the standardized data. The convolutional neural network model comprises a plurality of convolutional layers and fully connected layers, and a max pooling layer and a Dropout layer are inserted between the convolutional layers and the fully connected layers. The max pooling layer is used for dimension reduction of features, and the Dropout layer is used to prevent overfitting of the convolutional neural network model during training. A cross-entropy loss function is used to calculate the loss of the convolutional neural network model, and the value of the cross-entropy loss function is minimized as the training target until a preset number of iterations is reached, and then the training is completed.

[0081] The convolutional neural network model is used as a public large model provided by the application to learn effective feature representations from device running data.

[0082] Step S230: inputting the feature vector into a pre-trained SVM classifier to calculate an abnormal probability value corresponding to each data sample in the first data set;

[0083] The specific method of inputting the feature vector into the pre-trained SVM classifier to calculate the abnormal probability value corresponding to each data sample in the first data set is as follows: taking the feature vector extracted by the convolutional neural network model as the input of the SVM classifier, taking the label of a normal sample as 1 and the label of an abnormal sample as -1; defining the optimization objective of the SVM classifier as finding a maximum interval hyperplane to correctly classify data samples of different categories, solving the optimization objective of the SVM, obtaining the optimal Lagrange multiplier, and thus obtaining the trained SVM classifier; using the trained SVM classifier, for each data sample in the first data set, calculating the distance of the data sample to the hyperplane, and mapping the distance to an abnormal probability value through a sigmoid function;

[0084] The application takes the feature vector extracted by the convolutional neural network model as the input of the SVM classifier, realizes the connection of the public large model and the operation and maintenance large model for anomaly detection, and realizes the diagnosis of abnormal data in the device operation and maintenance process.

[0085] Step S240: if the abnormal probability value is greater than a preset threshold, determining that the data sample is in an abnormal state, and if the abnormal probability value is less than or equal to the preset threshold, determining that the data sample is in a normal state, to obtain a state determination result;

[0086] The value of the preset threshold is set by a person skilled in the art according to experience.

[0087] Step S250: classifying and screening the data samples in the first data set using the state determination result, taking all data samples corresponding to the abnormal state as abnormal samples to form an abnormal data subset;

[0088] Step S260: Clustering the abnormal samples in the abnormal data subset using the K-means clustering algorithm to obtain K different types of abnormal patterns;

[0089] The specific method of clustering the abnormal samples in the abnormal data subset using the K-means clustering algorithm to obtain K different types of abnormal patterns is as follows:

[0090] Step S261: Randomly selecting K abnormal samples as initial clustering centers;

[0091] Step S262: Repeating the following steps until the clustering centers no longer change: for each abnormal sample, calculating its Euclidean distance to the K clustering centers; dividing the abnormal sample into the class where the nearest clustering center is located; updating the clustering center of each class based on the re-assigned abnormal samples to be the mean vector of all abnormal samples in the class;

[0092] Step S263: When the clustering centers no longer change, obtaining K clustering centers, each of which represents an abnormal pattern;

[0093] In actual application, the system first loads a pre-trained convolutional neural network anomaly detection model, which uses the ResNet-50 architecture and is trained on a historical data set containing 1 million normal samples and 200,000 abnormal samples. The model accuracy reaches 96.8%. For the first data set processed by wavelet denoising, the system automatically extracts a multi-dimensional feature vector, including statistical features such as mean 2.34, standard deviation 1.67, skewness 0.89, frequency domain features such as dominant frequency component 3.2 Hz, power spectral density peak 0.045, and time domain features such as peak factor 4.2, pulse factor 3.8, etc. Finally, a 128-dimensional feature vector is formed. The feature vector is then input into the SVM classifier for anomaly probability calculation. The SVM classifier uses a radial basis function kernel, with a penalty parameter C set to 10 and a kernel parameter gamma set to 0.001. The SVM classifier outputs an anomaly probability of 0.73, which exceeds the preset threshold of 0.65, and the system automatically determines that the current data state is abnormal. To ensure the accuracy of the determination, the system simultaneously runs an ensemble learning algorithm, combines the prediction results of the random forest and gradient boosting decision tree, and obtains the final anomaly probability of 0.76 through a weighted voting mechanism, further verifying the reliability of the anomaly determination. The system automatically classifies the identified abnormal data samples into the abnormal data subset, providing a data basis for subsequent fault diagnosis and early warning processing.

[0094] Step S300: Obtaining historical equipment operating parameters and environmental parameters, and using a Bayesian network to perform correlation analysis on the abnormal state and the historical equipment operating parameters and environmental parameters to obtain a set of potential causes of the abnormality;

[0095] As Figure 3As shown, the abnormal reason analysis method flowchart provided by the present application; the specific method for obtaining the historical equipment operation parameters and the environmental parameters, and correlatively analyzing the abnormal state and the historical equipment operation parameters and the environmental parameters by using the Bayesian network to obtain the potential reason set of the abnormality is:

[0096] Step S310: obtaining the historical equipment operation parameters and the environmental parameters from the database;

[0097] Step S320: constructing the Bayesian network, establishing the conditional probability relationship between the historical equipment operation parameters, the environmental parameters and the abnormal state, determining the causal relationship and the probability distribution parameters between the nodes;

[0098] The specific method for constructing the Bayesian network, establishing the conditional probability relationship between the historical equipment operation parameters, the environmental parameters and the abnormal state, and determining the causal relationship and the probability distribution parameters between the nodes is: selecting the variables related to the equipment abnormal state, the variables including the historical equipment operation parameters, the environmental parameters and the abnormal state, determining the causal relationship between the variables based on the expert knowledge, and determining the topological structure of the Bayesian network; initializing the probability distribution parameters of each node in the Bayesian network in a uniform distribution manner; using the historical equipment operation parameters and the environmental parameters to estimate the conditional probability distribution parameters of each node in the Bayesian network, and the conditional probability distribution parameters are estimated by using the Bayesian estimation method;

[0099] The Bayesian network model is constructed by the present application, which is used to describe the causal relationship between the equipment components, the failure modes and the operation parameters, and the abnormal detection result is used as the input of the Bayesian network to infer the most possible failure cause. The Bayesian network model is the core component of the operation and maintenance large model proposed by the present application, which is used for the diagnosis and positioning of the failure cause.

[0100] Step S330: updating the conditional probability distribution of the Bayesian network using the abnormal data subset;

[0101] The specific method for updating the conditional probability distribution of the Bayesian network using the abnormal data subset is: using the abnormal samples in the abnormal data subset, merging with the historical abnormal samples to form a training data set, re-estimating the conditional probability distribution parameters of the Bayesian network to obtain the updated Bayesian network; for the new abnormal sample received, finding the corresponding state node of the abnormal sample in the Bayesian network, updating the conditional probability distribution parameters of the state node, and updating the conditional probability distribution parameters of the parent nodes of the state node;

[0102] Step S340: calculating the posterior probability of the state node corresponding to each abnormal sample in the root cause analysis model in the abnormal data subset;

[0103] The calculation formula of the posterior probability is: Wherein, P(x|y) is a likelihood function, P(y) is a prior probability, P(x) is an evidence factor, and P(x, y) represents a joint probability distribution;

[0104] The evidence factor is calculated according to the formula: P(x) = ΣP(x, y) = ΣP(x|y) * P(y);

[0105] Step S350: Select the state node with the maximum posterior probability as the potential cause of the anomaly, and add it to the potential cause set;

[0106] For example, the system first acquires the historical operation records of the abnormal device in the past 30 days through a database query interface, including temperature, pressure, vibration frequency, and other key parameters, and extracts environmental parameter data such as humidity, ambient temperature, and voltage fluctuation in the corresponding time period. Based on the Bayesian network, these historical data are constructed into a conditional probability network, with device temperature anomaly as the target node and environmental temperature, device load, maintenance cycle, etc. as parent nodes. The system uses Bayes' theorem to calculate the conditional probability, for example, the probability of device temperature anomaly under the condition of environmental temperature exceeding 35 degrees Celsius is 0.72, and the probability of anomaly under the condition of continuous operation of the device exceeding 168 hours is 0.58. Through Markov chain Monte Carlo algorithm, the contribution weight of each factor to the current anomaly is calculated, and the result shows that the contribution of environmental factors is 0.45, the contribution of device aging is 0.32, and the contribution of improper operation is 0.23. The system further applies information entropy theory to quantify the uncertainty of each factor, and calculates that the information entropy value of environmental temperature fluctuation is 2.34 bits, and the information entropy value of device load change is 1.87 bits. Finally, through a probability threshold screening mechanism, factors with a related probability greater than 0.3 are included in the potential cause set, forming an analysis result including three main potential causes of environmental temperature anomaly, device overload operation, and cooling system failure, providing a quantitative basis for subsequent fault diagnosis.

[0107] Step S400: Construct a device fault knowledge graph to perform path reasoning on the causes in the potential cause set, and generate a fault correlation mapping table;

[0108] The specific method of constructing a device fault knowledge graph to perform path reasoning on the causes in the potential cause set, and generating a fault correlation mapping table is:

[0109] Step S410: According to the causes in the potential cause set, search for the corresponding fault cause node in the device fault knowledge graph;

[0110] The definition method of the device fault knowledge graph is: taking the fault reason, the device component and the parameter as nodes, the parameter including the device operation parameter and the environment parameter; taking the causal relationship between the nodes as edges, the device fault knowledge graph is constructed; the device fault knowledge graph is used for storing the device component, the fault mode, the maintenance scheme and other field knowledge, the fault reason diagnosed by the Bayesian network is mapped into the knowledge graph, and the related component and parameter can be found. The device fault knowledge graph provided in the application is another important component of the operation and maintenance big model, and is used for fault tracing and knowledge management.

[0111] Step S420: taking the fault reason node as the starting point, the breadth-first search algorithm is used to traverse the device fault knowledge graph, and a device component node set directly connected with the fault reason node is found;

[0112] The specific method of taking the fault reason node as the starting point, using the breadth-first search algorithm to traverse the device fault knowledge graph, and finding the device component node set directly connected with the fault reason node is:

[0113] Step S421: selecting the fault reason node as the starting point, adding it to the queue, and initializing the access node set as empty, which is used for recording the accessed nodes;

[0114] Step S422: when the queue is not empty, the following steps are repeated: removing the head node of the queue, marking the node as accessed, and adding it to the access node set; traversing all neighbor nodes of the head node, if the neighbor node has not been accessed, adding the neighbor node to the queue;

[0115] Step S423: after the search is completed, the nodes in the access node set are the device component node set directly connected with the fault reason node;

[0116] Step S430: for the component nodes in the device component node set, the correlation degree between the component nodes and the fault reason node is calculated, if the correlation degree is greater than a preset threshold, the component node is marked as a high-risk component;

[0117] The calculation formula of the correlation degree is: Wherein, P(c|f) is the probability of the device component node c appearing problems under the condition that the device component node f occurs, P(f|c) is the probability of the fault reason node f occurring under the condition that the device component node c appears problems, P(c) and P(f) are the prior probabilities of the device component node c and the device component node f respectively;

[0118] The preset threshold is set by the person skilled in the art according to experience.

[0119] Step S440: obtaining the parameter node set associated with the high-risk component;

[0120] The parameter node set acquisition method is: for high-risk component nodes, through the adjacency relationship in the device fault knowledge graph, find the parameter node set directly connected therewith; find the nodes in the device fault knowledge graph that have a causal relationship with the high-risk component nodes, and add them to the parameter node set.

[0121] Step S450: calculate the support and confidence of the abnormal state and the high-risk component and the parameter node by using the association rule mining algorithm;

[0122] The association rule mining algorithm uses the Apriori algorithm, converts the abnormal state node, the high-risk component node and the parameter node into transactional data, mines the frequent item set from the transactional data, calculates the support of each item set, and takes the item set with support greater than or equal to the support threshold as the frequent item set; calculate the confidence of each frequent item set;

[0123] Step S460: generate a fault association mapping table between the high-risk component, the fault cause, the parameter, the association degree and the confidence;

[0124] Specifically, based on the confidence and association degree of each frequent item set, a fault association mapping table is generated, which is used to explain the association pattern between the abnormal state and the high-risk component and the parameter, and provides decision support for fault diagnosis.

[0125] For example, based on the set of potential causes identified in advance, the system automatically calls the pre-constructed device fault knowledge graph for deep path reasoning analysis. Taking the bearing temperature anomaly as an example, the system traverses the node relationship in the knowledge graph, and identifies the device components related to the bearing through the breadth-first search algorithm, including the lubrication system, the cooling system, the transmission shaft and the sealing device. For each component, the system further extracts the associated operating parameters. The lubrication system is associated with three parameters: lubricating oil temperature, oil pressure and flow, and their normal operating thresholds are 45-65 degrees Celsius, 0.2-0.4 megapascals and 15-25 liters per minute, respectively. The cooling system is associated with cooling water temperature and flow parameters, and the normal range is 25-35 degrees Celsius and 80-120 liters per minute. The system uses the association rule mining algorithm to calculate the correlation coefficient between the parameters, and finds that the correlation coefficient between the bearing temperature and the lubricating oil temperature reaches 0.87, and the negative correlation coefficient between the bearing temperature and the cooling water flow is-0.72. Based on these analysis results, the system automatically generates a fault association mapping table, which contains structured data in six dimensions: fault phenomenon, potential cause, associated component, monitoring parameter, normal threshold and correlation weight. In this way, the system can provide accurate parameter monitoring targets and judgment basis for subsequent fault diagnosis, forming a complete fault analysis closed loop.

[0126] Step S500: Establish a multi-objective optimization function based on the fault correlation mapping table and the real-time state data of the equipment, and solve the optimal treatment scheme combination;

[0127] The specific method of establishing a multi-objective optimization function based on the fault correlation mapping table and the real-time state data of the equipment, and solving the optimal treatment scheme combination is:

[0128] Step S510: Extract the correlation data from the fault correlation mapping table, and fuse it with the real-time state data of the equipment to form the input of the optimization problem;

[0129] Specifically, the equipment components, fault causes, correlation degrees, and confidence degrees are extracted from the fault correlation mapping table to form a correlation data set. The real-time state data of the equipment at the current time is obtained from the equipment monitoring system, and the correlation data and the real-time state data of the equipment are left-joined as the input of the optimization problem. The data after left-joining contains the equipment components, fault causes, correlation degrees, confidence degrees, and real-time state data of the equipment.

[0130] Step S520: Establish a multi-objective optimization function with the minimum treatment time and the minimum resource cost as the optimization objectives, and the treatment scheme as the decision variable;

[0131] The present application constructs a multi-objective optimization model, takes the correlation data of the fault diagnosis result and the real-time state data as the input, generates the optimal treatment scheme, and uses the optimization model as the decision module of the operation and maintenance large model to generate specific operation and maintenance job instructions and scheduling schemes.

[0132] Step S530: Optimize and solve the treatment scheme by using the non-dominated sorting genetic algorithm to obtain the optimal treatment scheme combination;

[0133] Step S540: For each treatment scheme in the optimal treatment scheme combination, verify whether it meets the actual constraint condition, and if it meets the actual constraint condition, add it to the executable scheme set;

[0134] The actual constraint condition includes: each equipment component can only select one treatment scheme; and the use amount of each resource does not exceed the maximum use amount.

[0135] Step S550: Select the treatment scheme with the optimal comprehensive benefit from the executable scheme set as the final scheme;

[0136] The comprehensive benefit E j is calculated as: Wherein, w1 and w2 are weight coefficients of the two optimization objectives, f 1,max and f 2,max are maximum values of the treatment time and the resource cost, respectively, f1(X j ), f2(Xj ) are two optimization objectives, respectively, X j is a treatment scheme, X j is contained in a set of executable schemes;

[0137] Exemplarily, the system first calls the fault correlation mapping table, identifies that the transformer overheating fault is associated with the cooling system anomaly and the uneven load distribution with the correlation degrees of 0.85 and 0.72 respectively. Combined with the real-time device state monitoring data showing that the current transformer temperature is 89.5℃, the load rate is 92.3%, and the on-site resource distribution data includes 3 standby cooling devices, 2 groups of maintenance personnel, and 1 load transfer device, the system starts the genetic algorithm optimization engine. The algorithm initializes the population size to 50 individuals, each individual representing a treatment scheme combination, and the encoding method uses a binary string to represent the resource allocation strategy. After genetic operations with a crossover probability of 0.8 and a mutation probability of 0.02, the algorithm converges to the optimal solution after 100 iterations. The optimal scheme is determined to start 2 cooling devices to cool down, which is expected to take 45 minutes and cost 1200 yuan, while performing 30% load transfer which takes 25 minutes and costs 800 yuan, with a total treatment time of 45 minutes and a total cost of 2000 yuan. The algorithm reaches the convergence condition at the 78th generation, and the fitness value increases from the initial 0.42 to 0.89, meeting the requirements of the time constraint of 60 minutes and the cost constraint of 2500 yuan, forming an optimal treatment strategy that takes into account efficiency and economy.

[0138] Embodiment 2

[0139] As Figure 4 shown, the intelligent operation and maintenance large model provided by the present application is connected with a public large model system, which comprises:

[0140] A data preprocessing module is configured to obtain device operation data from sensor nodes through a preset data acquisition framework, and pre-process the device operation data to obtain a first data set;

[0141] An abnormal state identification module is configured to identify abnormal states of the first data set by using a convolutional neural network and an SVM classifier, extract an abnormal data subset, and cluster the abnormal data subset to obtain an abnormal pattern;

[0142] An abnormal reason analysis module is configured to obtain historical device operation parameters and environmental parameters, and perform correlation analysis on the abnormal state, the historical device operation parameters, and the environmental parameters by using a Bayesian network to obtain a set of potential causes of the abnormality;

[0143] A fault correlation reasoning module is configured to construct a device fault knowledge graph to perform path reasoning on the causes in the set of potential causes, and generate a fault correlation mapping table;

[0144] An optimal scheme optimization module is configured to establish a multi-objective optimization function based on the fault correlation mapping table and real-time device state data, and solve an optimal treatment scheme combination.

[0145] In addition, parts of the above technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of corresponding technical solutions in the prior art are not described in detail to avoid excessive repetition.

[0146] The specific embodiments described above further illustrate the objects, technical solutions, and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for connecting a smart operation and maintenance big model with a public big model, characterized in that: include: Acquire device operation data from sensor nodes through a preset data acquisition framework, and pre-process the data to obtain a first data set; Using a convolutional neural network and a SVM classifier to identify abnormal states of the first data set, extracting abnormal data subsets and clustering them to obtain abnormal patterns; Obtain historical equipment operating parameters and environmental parameters, use Bayesian networks to correlate abnormal conditions with historical equipment operating parameters and environmental parameters, and obtain a set of potential causes of abnormalities; Build a knowledge graph of equipment failures to perform path reasoning on the causes in the potential cause set and generate a fault association mapping table; A multi-objective optimization function is established based on the fault association mapping table and real-time equipment status data to solve the optimal combination of treatment solutions.

2. The method for connecting the intelligent operation and maintenance big model with the public big model according to claim 1, characterized in that: The specific method of obtaining the device operation data from the sensor node through the preset data acquisition framework and preprocessing the data to obtain the first data set is: Device operating parameters are obtained from N sensor nodes through a preset data acquisition framework, the original device operating parameters are cleaned to obtain a cleaned data set, and a time series smoothing algorithm is used to remove noise from the cleaned data set to obtain a denoised first data set, which is composed of data samples.

3. The method for connecting the intelligent operation and maintenance big model with the public big model according to claim 2, characterized in that: The specific method of using the convolutional neural network and the SVM classifier to identify abnormal states of the first data set, extract abnormal data subsets and cluster them to obtain abnormal patterns is: Performing standardization on the first data set after denoising through data preprocessing to obtain standardized data; Based on the standardized data, a convolutional neural network model is used to perform a feature extraction operation to extract a feature vector from the standardized data; The feature vector is input into a pre-trained SVM classifier to calculate an abnormal probability value corresponding to each data sample in the first data set; If the abnormal probability value is greater than the preset threshold, the data sample is determined to be in an abnormal state. If the abnormal probability value is less than or equal to the preset threshold, it is determined to be in a normal state, and a state determination result is obtained; Using the state determination result to classify and screen the data samples in the first data set, taking the data samples corresponding to all abnormal states as abnormal samples to form an abnormal data subset; The K-means clustering algorithm is used to cluster the abnormal samples in the abnormal data subset to obtain K different types of abnormal patterns.

4. The method for connecting the intelligent operation and maintenance big model with the public big model according to claim 3 is characterized in that: The specific method of obtaining historical equipment operating parameters and environmental parameters, and using a Bayesian network to perform correlation analysis between abnormal states and historical equipment operating parameters and environmental parameters to obtain a set of potential causes of abnormalities is as follows: Obtain historical equipment operating parameters and environmental parameters from the database; Construct a Bayesian network to establish the conditional probability relationship between historical equipment operating parameters, environmental parameters and abnormal conditions, and determine the causal relationship and probability distribution parameters between each node; Update the conditional probability distribution of the Bayesian network using the subset of outlier data; For each abnormal sample in the abnormal data subset, calculate the posterior probability of its corresponding state node in the root cause analysis model; The state node with the largest posterior probability is selected as the potential cause of the anomaly and added to the potential cause set.

5. The method for connecting the intelligent operation and maintenance big model with the public big model according to claim 4, characterized in that: The specific method of obtaining historical equipment operating parameters and environmental parameters, and using a Bayesian network to perform correlation analysis between abnormal states and historical equipment operating parameters and environmental parameters to obtain a set of potential causes of abnormalities is as follows: Obtain historical equipment operating parameters and environmental parameters from the database; Construct a Bayesian network to establish the conditional probability relationship between historical equipment operating parameters, environmental parameters and abnormal conditions, and determine the causal relationship and probability distribution parameters between each node; Update the conditional probability distribution of the Bayesian network using the subset of outlier data; For each abnormal sample in the abnormal data subset, calculate the posterior probability of its corresponding state node in the root cause analysis model; The state node with the largest posterior probability is selected as the potential cause of the anomaly and added to the potential cause set.

6. The method for connecting the intelligent operation and maintenance big model with the public big model according to claim 5, characterized in that: The specific method of constructing the equipment fault knowledge graph to perform path reasoning on the causes in the potential cause set and generate the fault association mapping table is as follows: According to the causes in the potential cause set, search for the corresponding fault cause node in the equipment fault knowledge graph; Starting from the fault cause node, a breadth-first search algorithm is used to traverse the equipment fault knowledge graph to find the set of equipment component nodes directly connected to the fault cause node; For each component node in the device component node set, calculate its correlation with the fault cause node. If the correlation is greater than a preset threshold, mark the component node as a high-risk component. Get the parameter node set associated with high-risk components; Use association rule mining algorithms to calculate the support and confidence of abnormal states and high-risk components and parameter nodes; Generate a fault correlation mapping table between high-risk components, fault causes, parameters, correlation and confidence.

7. The method for connecting the intelligent operation and maintenance big model with the public big model according to claim 6, characterized in that: The device failure knowledge graph is defined as follows: using failure causes, device components, and parameters as nodes, where the parameters include device operating parameters and environmental parameters; and using the causal relationships between nodes as edges to construct a device failure knowledge graph.

8. The method for connecting the intelligent operation and maintenance big model with the public big model according to claim 7, characterized in that: The specific method of traversing the equipment fault knowledge graph using a breadth-first search algorithm starting from the fault cause node to find the set of equipment component nodes directly connected to the fault cause node is as follows: Select the node causing the fault as the starting point, add it to the queue, and initialize the visited node set to be empty to record the nodes that have been visited; When the queue is not empty, repeat the following steps: remove the first node, mark the node as visited, and add it to the visited node set; traverse all neighbor nodes of the first node, and if the neighbor node has not been visited, add the neighbor node to the queue; After the search is completed, the nodes in the access node set are the device component node set directly connected to the fault cause node.

9. The method for connecting the intelligent operation and maintenance big model with the public big model according to claim 8, characterized in that: The specific method of establishing a multi-objective optimization function based on the fault association mapping table and the real-time status data of the equipment to solve the optimal treatment solution combination is as follows: Extract association data from the fault association mapping table and fuse it with the real-time status data of the equipment to form the input of the optimization problem; Taking minimizing disposal time and minimizing resource cost as optimization objectives and disposal options as decision variables, a multi-objective optimization function is established; The non-dominated sorting genetic algorithm is used to optimize the treatment plan and obtain the optimal treatment plan combination; For each treatment plan in the optimal treatment plan combination, verify whether it meets the actual constraints. If so, add it to the executable plan set; The disposal plan with the best comprehensive benefits is selected from the set of executable plans as the final plan.

10. A system for connecting a smart operation and maintenance big model with a public big model, which is implemented based on the method for connecting a smart operation and maintenance big model with a public big model according to any one of claims 1 to 9, characterized in that: include: A data preprocessing module is used to obtain device operation data from the sensor node through a preset data acquisition framework, and preprocess the data to obtain a first data set; An abnormal state identification module is used to identify abnormal states of the first data set using a convolutional neural network and an SVM classifier, extract abnormal data subsets and cluster them to obtain abnormal patterns; The abnormality cause analysis module is used to obtain historical equipment operating parameters and environmental parameters, and use Bayesian networks to correlate abnormal conditions with historical equipment operating parameters and environmental parameters to obtain a set of potential causes of abnormalities; The fault association reasoning module is used to build a knowledge graph of equipment faults, perform path reasoning on the causes in the potential cause set, and generate a fault association mapping table; The optimal solution optimization module is used to establish a multi-objective optimization function based on the fault association mapping table and the real-time status data of the equipment to solve the optimal treatment solution combination.

Images