Intelligent factory operation and maintenance management system based on multi-dimensional data driving

Abstract

The invention discloses an intelligent factory operation and maintenance management system based on multi-dimensional data driving, and relates to the technical field of intelligent operation and maintenance management, and the system comprises a knowledge graph construction module which constructs a factory operation and maintenance knowledge graph through an RDF standard format based on factory operation and maintenance characteristics; the model construction module is used for constructing an intelligent operation and maintenance prediction model based on a deep reinforcement learning model and a long-short-term memory network model; the fault risk value prediction module is used for obtaining a fault risk prediction value through an intelligent operation and maintenance prediction model based on the factory operation and maintenance knowledge graph; and the risk assessment and early warning module generates an early warning notice and formulates maintenance measures based on the fault risk prediction value. According to the invention, by integrating multi-dimensional data acquisition and preprocessing, intelligent feature extraction, knowledge graph construction and an intelligent operation and maintenance prediction model fusing deep reinforcement learning and a long and short term memory network, the efficiency and reliability of factory operation are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent operation and maintenance management technology, and in particular to an intelligent factory operation and maintenance management system driven by multi-dimensional data. Background Art

[0002] With the continuous advancement of industrialization, plant operation and maintenance, as an important part of ensuring the efficient operation of enterprises, has received increasing attention. Plant operation and maintenance management involves multiple links such as equipment installation, commissioning, operation, maintenance, and overhaul. Its goal is to ensure the long-term efficient and stable operation of equipment and avoid economic losses caused by downtime due to failures. In recent years, with the continuous development of information and digital technologies, multi-dimensional data-driven intelligent plant operation and maintenance management has gradually become a key means to improve operation and maintenance efficiency and prediction accuracy. The application of advanced technologies such as big data technology, Internet of Things technology, and artificial intelligence algorithms has enabled plant operation and maintenance management to gradually shift from traditional manual experience to data-based intelligent decision-making. In particular, the introduction of technologies such as deep learning, natural language processing, and principal component analysis (PCA) has not only improved the analysis capabilities of multi-dimensional plant operation and maintenance data, but also provided strong support for the prediction of equipment failures and the optimization of maintenance strategies.

[0003] However, although many current intelligent plant operation and maintenance management systems have made progress in prediction accuracy and fault warning through data collection and analysis, existing technologies still have many limitations. First, traditional operation and maintenance management systems mostly rely on a single data source, lack effective fusion and intelligent analysis of multi-dimensional data, and are difficult to fully reflect the operating status and potential risks of equipment. Secondly, although existing intelligent prediction models can make predictions based on historical data, they often ignore the deep correlation analysis of multi-source heterogeneous data, especially in dynamic behavior prediction and complex relationship modeling, and are difficult to handle the complex interactions and long-term dependencies between system status and behavior during plant operation and maintenance. This makes the accuracy and real-time performance of current prediction models insufficient, and cannot fully meet the needs of increasingly complex industrial environments. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an intelligent factory operation and maintenance management system driven by multidimensional data to solve the problems of the traditional operation and maintenance management system in terms of multidimensional data fusion and deep correlation analysis.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: The present invention provides an intelligent plant operation and maintenance management system driven by multidimensional data, which includes a data acquisition and preprocessing module, which collects multidimensional plant operation and maintenance data and preprocesses the multidimensional plant operation and maintenance data; a feature extraction module, which extracts features based on the preprocessed multidimensional plant operation and maintenance data through natural language processing technology and principal component analysis algorithm to obtain plant operation and maintenance features; a knowledge graph construction module, which constructs a plant operation and maintenance knowledge graph through the RDF standard format based on the plant operation and maintenance features; a model construction module, which constructs an intelligent operation and maintenance prediction model based on a deep reinforcement learning model and a long short-term memory network model; a fault risk value prediction module, which obtains a fault risk prediction value through an intelligent operation and maintenance prediction model based on the plant operation and maintenance knowledge graph; a risk assessment and early warning module, which generates an early warning notification and formulates maintenance measures based on the fault risk prediction value.

[0007] As a preferred solution of the intelligent plant operation and maintenance management system based on multi-dimensional data drive described in the present invention, wherein: the multi-dimensional plant operation and maintenance data includes vibration frequency, power consumption, operation log and fault report; The preprocessing of the multi-dimensional plant operation and maintenance data includes preliminary cleaning, format conversion and standardization.

[0008] As a preferred solution of the intelligent factory operation and maintenance management system based on multidimensional data drive described in the present invention, wherein: based on the pre-processed multidimensional factory operation and maintenance data, feature extraction is performed through natural language processing technology and principal component analysis algorithm to obtain factory operation and maintenance features. The specific steps are as follows: Based on the pre-processed multi-dimensional factory operation and maintenance data, factory text data and factory numerical data are extracted through data analysis methods; Through the word embedding method, the factory affairs text data is converted into word vector form; Based on the factory affairs text data in the form of word vectors, the global context feature vector of the factory affairs text data is obtained by aggregation through aggregation method; Based on the factory numerical data, the covariance matrix is ​​obtained by statistical analysis; The covariance matrix is ​​decomposed by the eigenvalue decomposition method to extract the principal component eigenvalues ​​and corresponding principal component eigenvectors in the factory numerical data; Based on the order of the extracted principal component eigenvalues, the principal component eigenvector corresponding to the largest principal component eigenvalue is used as the characteristic principal component; Map the factory affairs numerical data to the characteristic principal components to obtain the factory affairs characteristics after dimensionality reduction; Based on the global context feature vector of factory affairs text data and the factory affairs features after dimensionality reduction, the factory affairs operation and maintenance features are obtained by connecting them in vector order through the splicing method.

[0009] As a preferred solution of the intelligent factory operation and maintenance management system based on multidimensional data drive described in the present invention, wherein: based on the factory operation and maintenance characteristics, the factory operation and maintenance knowledge graph is constructed through the RDF standard format, and the specific steps are as follows: Based on the factory operation and maintenance characteristics, the factory operation and maintenance entity types, factory operation and maintenance entity attributes, and factory operation and maintenance entity relationships of the knowledge graph are defined through feature analysis and semantic abstraction methods to obtain the semantic structure of the factory operation and maintenance knowledge graph; Based on the factory affairs text data and factory affairs numerical data in the multi-dimensional factory affairs operation and maintenance data, they are mapped into attribute values ​​of factory affairs operation and maintenance entities through natural language processing methods and data analysis methods; Based on the semantic structure of the plant operation and maintenance knowledge graph and the attribute values ​​of the plant operation and maintenance entities, the plant operation and maintenance knowledge graph triples are constructed using the RDF standard format; Based on the plant operation and maintenance knowledge graph triples, it is organized into an RDF graph structure through GraphDB to obtain the plant operation and maintenance knowledge graph.

[0010] As a preferred solution of the intelligent factory operation and maintenance management system based on multi-dimensional data drive described in the present invention, wherein: the intelligent operation and maintenance prediction model is constructed based on the deep reinforcement learning model and the long short-term memory network model, and the specific steps are as follows: Through graph embedding method and time series annotation method, the historical plant operation and maintenance knowledge graph is converted into time series data format; The model is based on the long short-term memory network model; The input layer of the LSTM model receives the historical plant operation and maintenance knowledge graph converted into a time series data format; LSTM layer uses memory cells and gate mechanisms. The LSTM model selectively remembers and forgets information, capturing long-term dependencies in time series. The fully connected layer extracts features and reduces the dimension based on the output of the LSTM layer and maps it to the output layer; The output layer outputs preliminary prediction values; The deep reinforcement learning layer optimizes the decision-making process by using the reward function through interactive learning with the environment, adjusts the parameters of the deep reinforcement learning model, and optimizes the output results of the LSTM model; The deep reinforcement learning output layer outputs the final prediction value; Construct a preliminary intelligent operation and maintenance prediction model; Through the back propagation algorithm and reinforcement learning algorithm, the preliminary intelligent operation and maintenance prediction model is trained, evaluated and optimized, and finally an intelligent operation and maintenance prediction model is constructed.

[0011] As a preferred solution of the intelligent plant operation and maintenance management system based on multidimensional data drive described in the present invention, the preliminary intelligent operation and maintenance prediction model is trained, evaluated and optimized through the back propagation algorithm and the reinforcement learning algorithm, and finally the intelligent operation and maintenance prediction model is constructed. The specific steps are as follows: The historical plant operation and maintenance knowledge graph converted into time series data format is divided into a training set and a validation set through random partitioning. The prediction performance of the LSTM model is evaluated by minimizing the loss function based on the training set, and the weights and parameters of the LSTM model are optimized through the back-propagation algorithm; Through the prediction value output by the LSTM model, the deep reinforcement learning model uses the policy gradient method to adjust the decision process according to the reward signal, adjust the output of the LSTM model, and optimize the fault risk prediction accuracy; Use the validation set to evaluate the prediction performance of the preliminary intelligent operation and maintenance prediction model by analyzing the accuracy, recall, and F1-score, and fine-tune the reward function, LSTM layer parameters, and reinforcement learning strategy based on the evaluation results; The k-fold cross-validation method is used to verify the initially trained intelligent operation and maintenance prediction model, and finally the intelligent operation and maintenance prediction model is output.

[0012] As a preferred solution of the intelligent plant operation and maintenance management system based on multidimensional data drive described in the present invention, wherein: based on the plant operation and maintenance knowledge graph, the fault risk prediction value is obtained through the intelligent operation and maintenance prediction model, and the specific steps are as follows: Based on the plant operation and maintenance knowledge graph converted into time series data format, the time series data is processed through the long short-term memory network model, and the deep reinforcement learning model is combined with the plant operation and maintenance knowledge graph to capture the complex dynamic relationships in plant operation and maintenance; Based on complex dynamic relationships, a multidimensional data-driven approach is used to obtain the failure risk prediction value at each time point. The expression is: ; in, Indicates time point The failure risk prediction value of represents the weight of the deep reinforcement learning model, Indicates that the DRL model is based on time points Factory Operation and Maintenance Knowledge Graph and time point Factory operation and maintenance status The predicted value of the output, represents the weight of the long short-term memory network model, Indicates that the LSTM model is based on time points To the time point The initial prediction value of the plant operation and maintenance knowledge graph output converted into time series data format, Indicates the length of the time period, Indicates time point Operation and maintenance feedback data, An index variable representing a time point.

[0013] As a preferred solution of the intelligent factory operation and maintenance management system based on multidimensional data drive described in the present invention, the LSTM model is based on time points. To the time point The initial prediction value of the plant operation and maintenance knowledge graph output in the time series data format is converted into the following specific steps: The LSTM model analyzes the plant operation and maintenance knowledge graph converted into a time series data format to predict the plant operation and maintenance status at the current moment, and obtains the preliminary prediction value by weighted averaging the prediction results at historical moments. The expression is: ; in, represents the number of historical moments, An index variable representing a historical moment, Indicates The weight coefficient of the LSTM model at each historical moment, express arrive The time interval is converted into a plant operation and maintenance knowledge graph in the time series data format. Indicates time point Move forward time plus time period , Indicates time point Move forward A moment.

[0014] As a preferred solution of the intelligent factory operation and maintenance management system based on multi-dimensional data drive described in the present invention, the DRL model is based on time points. Factory Operation and Maintenance Knowledge Graph and time point Factory operation and maintenance status The output prediction value, the specific steps are as follows, The deep reinforcement learning model uses the plant operation and maintenance knowledge graph, combined with the complex dynamic relationship between status and behavior in historical data, to continuously improve the prediction ability of plant operation and maintenance behavior through the decision optimization process. It also captures the dynamic relationship between plant operation and maintenance status and behavior by analyzing the multi-dimensional plant operation and maintenance data at past moments, and obtains the prediction value by weighted averaging the prediction results at historical moments. The expression is: ; in, Indicates The weighted coefficient of the DRL model at each historical moment, Indicates time point The knowledge graph of factory operation and maintenance at all times, Indicates time point The factory operation and maintenance status at all times.

[0015] As a preferred solution of the intelligent plant operation and maintenance management system based on multi-dimensional data drive described in the present invention, wherein: based on the fault risk prediction value, the early warning notification is generated and maintenance measures are formulated. The specific steps are as follows: Based on historical failure risk data, cluster analysis is used to set the upper risk threshold. and the lower risk threshold ; Based on the failure risk prediction value , passing the upper risk threshold and the lower risk threshold , assess the failure risk level of plant operation and maintenance; when When , it indicates that the failure risk of plant operation and maintenance is in the low failure risk zone; when When , it indicates that the failure risk of plant operation and maintenance is in the medium failure risk zone; when When , it indicates that the failure risk of plant operation and maintenance is in the high failure risk zone; Based on the fault risk level of plant operation and maintenance, real-time tracking and decision support of equipment operation status and potential fault risks are carried out, and early warning notifications are automatically generated; Based on the fault risk prediction value and early warning notification, and combined with the equipment operating environment and equipment status, formulate corresponding maintenance measures.

[0016] The beneficial effects of the present invention are: by integrating multi-dimensional data collection and preprocessing, intelligent feature extraction, knowledge graph construction, and an intelligent operation and maintenance prediction model that integrates deep reinforcement learning and long short-term memory networks, comprehensive monitoring of the factory operation and maintenance status and accurate fault risk prediction are achieved. It not only improves the automation and intelligence level of operation and maintenance management, but also effectively prevents equipment failures and reduces unplanned downtime through early warning and optimized maintenance measures, thereby significantly improving the efficiency and reliability of factory operations, reducing maintenance costs, and enhancing the stability and safety of the production system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 Schematic diagram of the intelligent factory operation and maintenance management system driven by multi-dimensional data in Example 1; Figure 2 This is a schematic diagram of constructing a factory operation and maintenance knowledge graph in Example 1. DETAILED DESCRIPTION

[0019] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0022] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides an intelligent factory operation and maintenance management system based on multi-dimensional data drive, including the following steps: The data acquisition and preprocessing module collects multi-dimensional plant operation and maintenance data and pre-processes the multi-dimensional plant operation and maintenance data.

[0023] Multi-dimensional plant operation and maintenance data includes vibration frequency, power consumption, operation logs and fault reports.

[0024] Preprocessing of multi-dimensional plant operation and maintenance data includes preliminary cleaning, format conversion and standardization.

[0025] It should be noted that the multi-dimensional plant operation and maintenance data is preprocessed through preliminary cleaning, format conversion and standardization preprocessing methods. The specific process is as follows: Preliminary cleaning: First, identify and remove outliers and noise in the multi-dimensional plant operation and maintenance data, handle missing values ​​- use interpolation methods to fill in the gaps or reasonably estimate according to business logic, and perform deduplication operations to eliminate duplicate records to ensure that the final data set is accurate, complete and highly consistent; Format conversion: Parse and convert multi-dimensional plant operation and maintenance data from different sources and in various original formats (such as CSV, JSON, XML or database records) into an internal standard format to ensure that all data items have consistent structures and clear semantics, and encode non-numeric multi-dimensional plant operation and maintenance data (such as text labels or category information) into numerical form; Standardization: Adjust the distribution of multidimensional plant operation and maintenance data of each feature to a common scale. Usually, the statistical parameters of each feature, such as the mean and standard deviation, are calculated, and then techniques such as minimum-maximum normalization or Z-score normalization are applied to map the values ​​to a specific interval or convert them to a standard normal distribution with zero mean and unit variance, thereby eliminating the dimensional effect.

[0026] The feature extraction module extracts features based on the pre-processed multi-dimensional plant operation and maintenance data through natural language processing technology and principal component analysis algorithm to obtain plant operation and maintenance features.

[0027] Based on the preprocessed multi-dimensional factory operation and maintenance data, factory affairs text data and factory affairs numerical data are extracted through data analysis methods.

[0028] It should be noted that specific parsing algorithms and rules are used to conduct in-depth analysis of the standardized multi-dimensional plant operation and maintenance data to identify and separate plant text data containing descriptive information (such as natural language descriptions in operation logs and fault reports) and plant numerical data of quantitative indicators (such as temperature readings, vibration frequencies and other measurement values).

[0029] The factory affairs text data is converted into word vector form through word embedding method.

[0030] It should be noted that a pre-trained or self-trained word embedding model (such as Word2Vec, GloVe, or FastText) is used to map each word to a real number vector of fixed length in a high-dimensional space, namely a word vector.

[0031] Based on the factory affairs text data in the form of word vectors, aggregation is performed through an aggregation method to obtain the global context feature vector of the factory affairs text data.

[0032] It should be noted that, first, for each word in each text record, its corresponding word vector is used for representation; then, a suitable aggregation strategy is selected, such as a simple average or weighted average of the word vectors of all words, using maximum / minimum pooling to select the most representative word vector features, or using a more complex attentive pooling method to assign different weights to words according to their importance in the sentence; finally, a fixed-length global context feature vector that can reflect the overall semantics and contextual information of the entire text paragraph is generated through the selected aggregation method.

[0033] Based on the factory numerical data, the covariance matrix is ​​obtained by statistical analysis.

[0034] It should be noted that the covariance between each pair of factory numerical data features is calculated, which measures the degree of correlation between the changes in the two eigenvalues; then, a square matrix is ​​constructed, in which the value at each position reflects the covariance between the corresponding two features, and the elements on the diagonal represent the variance of each feature itself; the final covariance matrix is ​​obtained; It should also be noted that the factory numerical data features come from the quantitative measurements in the pre-processed multi-dimensional factory operation and maintenance data set, such as vibration frequency, temperature, and power consumption.

[0035] The covariance matrix is ​​decomposed by the eigenvalue decomposition method to extract the principal component eigenvalues ​​and corresponding principal component eigenvectors in the factory affairs numerical data.

[0036] It should be noted that, first, the covariance matrix is ​​calculated based on the preprocessed factory numerical data; then, a mathematical algorithm (such as power iteration, QR algorithm or Jacobi method) is applied to perform eigenvalue decomposition on the covariance matrix to solve a set of eigenvalues ​​and their corresponding eigenvectors, where the eigenvalues ​​represent the amount of data variation explained by each principal component, and the eigenvectors indicate the projection method of the data on the directions of these principal components.

[0037] The extracted principal component eigenvalues ​​are sorted by size, and the principal component eigenvector corresponding to the largest principal component eigenvalue is taken as the characteristic principal component.

[0038] It should be noted that after completing the eigendecomposition of the covariance matrix, a set of eigenvalues ​​and their corresponding eigenvectors are obtained; then, they are sorted in descending order according to the size of the eigenvalues, because larger eigenvalues ​​represent more variation information in the data; and then the eigenvector with the largest eigenvalue is selected, which indicates the direction with the largest variance in the data set, that is, the first principal component.

[0039] The factory affairs numerical data are mapped to the characteristic principal components to obtain the factory affairs characteristics after dimensionality reduction.

[0040] It should be noted that, first, the characteristic decomposition of the covariance matrix is ​​used to find the characteristic principal components representing the main variation directions of the data; then, the most important one or several principal components are selected (usually those eigenvectors corresponding to the largest eigenvalues); then, the data is projected onto the new principal component coordinate axis by performing matrix multiplication on the original factory numerical data matrix and the selected principal component vector; finally, the factory characteristics after dimensionality reduction are obtained.

[0041] Based on the global context feature vector of factory affairs text data and the factory affairs features after dimensionality reduction, the factory affairs operation and maintenance features are obtained by connecting them in vector order through the splicing method.

[0042] It should be noted that, first, ensure that the global context feature vector reflecting the text content and the numerical feature vector after dimensionality reduction through principal component analysis have been obtained; then, in a predetermined order (usually text features first and then numerical features or vice versa), these two vectors are horizontally spliced ​​together to form a longer comprehensive feature vector; the final plant operation and maintenance features are obtained.

[0043] The knowledge graph construction module constructs the factory operation and maintenance knowledge graph based on the factory operation and maintenance characteristics through the RDF standard format.

[0044] Based on the characteristics of factory operation and maintenance, the factory operation and maintenance entity types, factory operation and maintenance entity attributes and factory operation and maintenance entity relationships of the knowledge graph are defined through feature parsing and semantic abstraction methods to obtain the semantic structure of the factory operation and maintenance knowledge graph.

[0045] It should be noted that, first, the extracted plant operation and maintenance features are deeply analyzed to identify representative entity types (such as equipment, operators, maintenance events, etc.); then, the key attributes of each entity (such as equipment model, operating status, fault code, etc.) are determined, and high-level concepts and categories are extracted through semantic abstraction; then, based on the interactions and associations between entities in the multidimensional plant operation and maintenance data, the relationships between entities are defined (such as "equipment failure", "operator performs maintenance", etc.); finally, these entity types, attributes and relationships are organized in accordance with standard formats such as RDF (Resource Description Framework) to construct a knowledge graph semantic structure that can express the complex relationships and dynamic changes in the field of plant operation and maintenance.

[0046] Based on the factory affairs text data and factory affairs numerical data in the multi-dimensional factory affairs operation and maintenance data, they are mapped into attribute values ​​of factory affairs operation and maintenance entities through natural language processing methods and data analysis methods.

[0047] It should be noted that, first, natural language processing technology is used to parse the factory affairs text data, extract key information and convert it into structured semantic labels or classifications (such as fault type, operating instructions, etc.); at the same time, statistical analysis is performed on the factory affairs numerical data, and descriptive statistics (such as mean, standard deviation) or other characteristic values ​​are calculated to quantify equipment performance or operating status; then, the information extracted from the text and numerical values ​​is associated with the corresponding factory affairs operation and maintenance entity as its attribute value, such as assigning a specific fault code or average energy consumption to a certain equipment entity.

[0048] Based on the semantic structure of the factory operation and maintenance knowledge graph and the attribute values ​​of the factory operation and maintenance entities, the factory operation and maintenance knowledge graph triples are constructed using the RDF standard format.

[0049] It should be noted that, first, the entities and their relationships are determined according to the defined semantic structure, and then each entity and its attribute values ​​are represented as resources (Subject) in RDF; then, appropriate predicates (Predicate) are selected or defined to describe the relationship between entities or the attributes of entities; finally, the attribute values ​​of the entities are taken as objects (Object), organized in the form of "subject-predicate-object" triples, and encoded using RDF standard syntax (such as Turtle, N-Triples, etc.), thereby forming a series of knowledge graph triples that describe objects and their relationships in the field of plant operations and maintenance.

[0050] Based on the plant operation and maintenance knowledge graph triples, GraphDB is used to organize them into an RDF graph structure to obtain the plant operation and maintenance knowledge graph.

[0051] It should be noted that, first, the constructed RDF triples are imported into a graph database such as GraphDB; then, GraphDB automatically parses these triples and organizes them into a coherent graph structure based on the relationship between the subject, predicate, and object, with entities as nodes and relationships as edges; then, the query language provided by GraphDB (such as SPARQL) is used for data retrieval and complex pattern matching to verify and optimize the graph structure; finally, the complete plant operation and maintenance knowledge graph is accessed and displayed through visualization tools or API interfaces.

[0052] The model building module builds an intelligent operation and maintenance prediction model based on the deep reinforcement learning model and the long short-term memory network model.

[0053] Through graph embedding method and time series labeling method, the historical factory operation and maintenance knowledge graph is converted into time series data format.

[0054] It should be noted that, first, graph embedding technology (such as GraphSAGE, Node2Vec, etc.) is used to convert the nodes and edges in the knowledge graph into low-dimensional vector representations to capture the semantic information of entities and their relationships; then, these embedded vectors are sorted according to timestamps or the order of events, and time series annotations are added to identify the time point of each state; then, the embedded vectors with time labels are serialized to form a continuous time series dataset.

[0055] The model is based on the long short-term memory network model; The input layer of the LSTM model receives the historical plant operation and maintenance knowledge graph converted into a time series data format; LSTM layer uses memory cells and gate mechanisms. The LSTM model selectively remembers and forgets information, capturing long-term dependencies in time series. The fully connected layer extracts features and reduces the dimension based on the output of the LSTM layer and maps it to the output layer; The output layer outputs preliminary prediction values; The deep reinforcement learning layer optimizes the decision-making process by using the reward function through interactive learning with the environment, adjusts the parameters of the deep reinforcement learning model, and optimizes the output results of the LSTM model; The deep reinforcement learning output layer outputs the final prediction value; Construct a preliminary intelligent operation and maintenance prediction model; Through the back propagation algorithm and reinforcement learning algorithm, the preliminary intelligent operation and maintenance prediction model is trained, evaluated and optimized, and finally an intelligent operation and maintenance prediction model is constructed.

[0056] It should be noted that the benefits of building an intelligent operation and maintenance prediction model based on the long short-term memory network (LSTM model) model combined with the deep reinforcement learning (DRL) model are: the LSTM model can effectively capture the long-term dependencies in time series data and handle the complex patterns that change over time in plant operation and maintenance; while deep reinforcement learning continuously optimizes the decision-making process through interaction with the environment, improving the accuracy of fault prediction and maintenance strategies. The combination of the two not only enhances the model's ability to learn historical data and improves the accuracy of predicting future fault risks, but also can adaptively adjust maintenance measures to achieve more intelligent and automated operation and maintenance management, thereby effectively reducing equipment failure rates and maintenance costs.

[0057] Through the back propagation algorithm and reinforcement learning algorithm, the preliminary intelligent operation and maintenance prediction model is trained, evaluated and optimized, and finally an intelligent operation and maintenance prediction model is constructed.

[0058] Through the random partitioning method, the historical plant operation and maintenance knowledge graph converted into time series data format is divided into a training set and a validation set.

[0059] It should be noted that, first, ensure that the time series data is arranged in chronological order and maintains integrity; then, according to the predetermined ratio (such as 70% training set and 30% validation set), use random sampling method to select samples from the entire data set, while ensuring that the continuity and context dependency of the time series are not destroyed; then, when dividing, care must be taken to avoid future data leakage into the training set. The usual practice is to extract data from an earlier time period as a training set and data from a later time period as a validation set; finally, generate independent training and validation sets.

[0060] The prediction performance of the LSTM model is evaluated by minimizing the loss function based on the training set, and the weights and parameters of the LSTM model are optimized through the back-propagation algorithm.

[0061] It should be noted that the training data is input into the LSTM model for forward propagation, and the loss value (such as mean square error or cross entropy) between the LSTM model output and the true label is calculated; then the back propagation algorithm is used to update the weight matrix, bias term and gate unit parameters of the LSTM model according to the gradient of the loss relative to each parameter to gradually reduce the loss function.

[0062] Through the predicted value output by the LSTM model, the deep reinforcement learning model uses the policy gradient method to adjust the decision-making process according to the reward signal, adjust the output of the LSTM model, and optimize the fault risk prediction accuracy.

[0063] It should be noted that the LSTM model first generates preliminary fault risk prediction values ​​based on the time series data converted from the historical plant operation and maintenance knowledge graph; then, these prediction values ​​are input into the deep reinforcement learning (DRL) model as environmental states. The DRL model performs actions according to the current strategy and receives reward or penalty signals based on the actual operation and maintenance results; then, the DRL model uses the policy gradient method to adjust its decision-making strategy parameters according to the reward signal, and at the same time feeds the optimized decision back to the LSTM model, prompting the LSTM to update its weights and parameters through the back-propagation algorithm to correct the prediction output.

[0064] Using the validation set, the prediction performance of the preliminary intelligent operation and maintenance prediction model was evaluated by analyzing the accuracy, recall rate, and F1-score. The reward function, parameters of the LSTM layer, and reinforcement learning strategy were fine-tuned based on the evaluation results.

[0065] It should be noted that when using the validation set to evaluate the performance of the preliminary intelligent operation and maintenance prediction model, the validation data is first input into the trained model to generate a fault risk prediction result, and compared with the actual fault labels in the validation set, the accuracy (the proportion of correct predictions), recall (the proportion of correct predictions that are actually positive) and F1-score (the harmonic mean of accuracy and recall) are calculated to comprehensively evaluate the predictive ability of the preliminary intelligent operation and maintenance prediction model. Based on the results of these evaluation indicators, the memory units and gate mechanism parameters of the LSTM layer are fine-tuned to optimize its ability to capture long-term dependencies in time series; at the same time, the reward function and other related parameters in the reinforcement learning strategy are adjusted to improve the decision-making process.

[0066] The k-fold cross-validation method is used to verify the initially trained intelligent operation and maintenance prediction model, and finally the intelligent operation and maintenance prediction model is output.

[0067] It should be noted that the multidimensional plant operation and maintenance data set is divided into k mutually exclusive subsets; one subset is selected as the validation set each time, and the remaining data is used to evaluate the performance of the initially trained intelligent operation and maintenance prediction model; for each subset, the performance indicators of the initially trained intelligent operation and maintenance prediction model on the validation set are recorded, such as accuracy, recall rate and F1-score; the results of all k validations are summarized, and the average performance is calculated to evaluate the overall stability and generalization ability of the initially trained intelligent operation and maintenance prediction model; the configuration or parameter setting of the initially trained intelligent operation and maintenance prediction model is selected according to the cross-validation results, and finally the verified and optimized intelligent operation and maintenance prediction model is output.

[0068] The fault risk value prediction module obtains the fault risk prediction value based on the plant operation and maintenance knowledge graph through the intelligent operation and maintenance prediction model.

[0069] Based on the plant operation and maintenance knowledge graph converted into time series data format, the time series data is processed through the long short-term memory network model, and the deep reinforcement learning model is combined with the plant operation and maintenance knowledge graph to capture the complex dynamic relationships in plant operation and maintenance.

[0070] It should be noted that, first, the plant operation and maintenance knowledge graph is converted into a time series data format to reflect the operation and maintenance status that changes over time; then, the LSTM model is used to analyze these time series data to capture long-term dependencies and historical patterns; at the same time, the DRL model makes decisions based on the current and past operation and maintenance status, and continuously optimizes its strategy through interaction with the environment; the LSTM model and the DRL model work together, the LSTM model provides accurate time series predictions, and the DRL model adjusts the operation and maintenance strategy based on the prediction results, and the two work together to identify and adapt to complex dynamic changes.

[0071] Based on complex dynamic relationships, a multidimensional data-driven approach is used to obtain the failure risk prediction value at each time point. The expression is: ; in, Indicates time point The failure risk prediction value of represents the weight of the deep reinforcement learning model, Indicates that the DRL model is based on time points Factory Operation and Maintenance Knowledge Graph and time point Factory operation and maintenance status The predicted value of the output, represents the weight of the long short-term memory network model, Indicates that the LSTM model is based on time points To the time point The initial prediction value of the plant operation and maintenance knowledge graph output converted into time series data format, Indicates the length of the time period, Indicates time point Operation and maintenance feedback data, An index variable representing a time point.

[0072] It should be noted that the operation and maintenance feedback data is collected and integrated from multiple sources of information such as equipment operating status, fault alarms, maintenance activity results, and performance indicators through real-time monitoring systems and databases that store historical data related to equipment operation and maintenance; Operation and maintenance feedback data includes real-time or historical records of equipment operating status, fault alarms, maintenance activity results, and performance indicators; It should be noted that the factory operation and maintenance status refers to a set of data that reflects the operating status of factory equipment at a specific point in time, including but not limited to the equipment's operating parameters (such as temperature, pressure, vibration frequency, etc.), operating status (such as startup, shutdown, maintenance), fault information, and environmental conditions; It should also be noted that the factory operation and maintenance status comes from multiple data sources, mainly including physical parameters collected in real time by the sensor network, operation logs recorded by the equipment’s own monitoring system, manual inspection reports entered by maintenance personnel, and historical fault and maintenance records.

[0073] LSTM model is based on time points To the time point The preliminary prediction value of the plant operation and maintenance knowledge graph output converted into time series data format.

[0074] The LSTM model analyzes the plant operation and maintenance knowledge graph converted into a time series data format to predict the plant operation and maintenance status at the current moment, and obtains the preliminary prediction value by weighted averaging the prediction results at historical moments. The expression is: ; in, represents the number of historical moments, An index variable representing a historical moment, Indicates The weight coefficient of the LSTM model at each historical moment, express arrive The time interval is converted into a plant operation and maintenance knowledge graph in the time series data format. Indicates time point Move forward time plus time period , Indicates time point Move forward A moment.

[0075] It should be noted that the calculation process of the above expression is as follows: First, for the past time points, each of which corresponds to a historical time period. The time series data of the time period is input into the LSTM model to generate a forecast value. Then, each forecast value is multiplied by its corresponding weighting coefficient, which reflects the importance of different historical moments. Next, all weighted forecast values ​​are summed and divided by the sum of all weighting coefficients to calculate a weighted average preliminary forecast value.

[0076] DRL model is based on time point Factory Operation and Maintenance Knowledge Graph and time point Factory operation and maintenance status Output predicted value.

[0077] The deep reinforcement learning model uses the plant operation and maintenance knowledge graph, combined with the complex dynamic relationship between status and behavior in historical data, to continuously improve the prediction ability of plant operation and maintenance behavior through the decision optimization process. It also captures the dynamic relationship between plant operation and maintenance status and behavior by analyzing the multi-dimensional plant operation and maintenance data at past moments, and obtains the prediction value by weighted averaging the prediction results at historical moments. The expression is: ; in, Indicates The weighted coefficient of the DRL model at each historical moment, Indicates time point The knowledge graph of factory operation and maintenance at all times, Indicates time point The factory operation and maintenance status at all times.

[0078] It should be noted that the calculation process of the above expression is as follows: First, for the past time points, each of which corresponds to a plant operation and maintenance knowledge graph and plant operation and maintenance status at a historical moment. These data are input into the deep reinforcement learning (DRL) model to generate a prediction value; then, each prediction value is multiplied by its corresponding weighting coefficient, and these weights reflect the importance of the prediction results at different historical moments; then, all weighted prediction values ​​are summed and divided by the sum of all weighting coefficients to calculate a weighted average prediction value.

[0079] The risk assessment and early warning module generates early warning notifications and formulates maintenance measures based on the failure risk prediction value.

[0080] Based on historical failure risk data, cluster analysis is used to set the upper risk threshold. and the lower risk threshold ; Based on the failure risk prediction value , passing the upper risk threshold and the lower risk threshold , assess the failure risk level of plant operation and maintenance.

[0081] It should be noted that, first, the historical fault risk data of plant operation and maintenance are collected and sorted; then, clustering algorithms (such as K-means, hierarchical clustering, etc.) are applied to divide these data into several clusters according to similarity, and each cluster represents a risk condition of a different level; then, the characteristics of each cluster are analyzed to determine the boundary points that can distinguish between low, medium and high risks, and accordingly set the lower risk threshold (indicating the upper limit of the low risk interval) and the upper risk threshold (indicating the lower limit of the high risk interval).

[0082] when When , it indicates that the failure risk of plant operation and maintenance is in the low failure risk zone; when When , it indicates that the failure risk of plant operation and maintenance is in the medium failure risk zone; when When , it indicates that the failure risk of plant operation and maintenance is in the high failure risk zone; Based on the fault risk level of plant operation and maintenance, real-time tracking and decision support are provided for equipment operating status and potential fault risks, and early warning notifications are automatically generated.

[0083] It should be noted that we continuously monitor and collect multi-dimensional plant operation and maintenance data, use advanced data analysis and machine learning algorithms to evaluate fault risks in real time, automatically determine the current risk level based on preset thresholds, and predict the development trend of potential faults. We provide optimized maintenance suggestions based on historical fault data. Once it is detected that the risk exceeds the set threshold, it will immediately trigger the automatic generation of early warning notifications, and promptly notify relevant personnel through emails, text messages and other channels to ensure rapid response and appropriate preventive or corrective measures.

[0084] Based on the fault risk prediction value and early warning notification, and combined with the equipment operating environment and equipment status, formulate corresponding maintenance measures.

[0085] It should be noted that corresponding maintenance measures are formulated as follows: Maintenance measures for low-risk areas: Regular inspection: Arrange regular inspection and maintenance according to the equipment operation cycle, such as cleaning, lubrication, replacement of wearing parts, etc., to ensure that the equipment remains in good condition; Optimize maintenance: Optimize maintenance plans based on historical operation and maintenance data, such as adjusting maintenance cycles or optimizing resource allocation to improve equipment reliability and lifespan.

[0086] Improved monitoring: In low-risk situations, enhance real-time monitoring of equipment to facilitate early identification of potential risks when they occur; Maintenance measures for medium risk areas: Strengthen monitoring: Strengthen real-time monitoring of equipment, especially fault predictive indicators (such as temperature, pressure, etc.), and use data analysis to determine the trend of potential faults; Predictive maintenance: Based on historical data and prediction models, repair or replace parts prone to problems in advance to avoid further risks; Additional testing: Perform additional testing or load testing on equipment with a higher probability of failure to verify that the equipment is within an acceptable risk range; Maintenance measures for high-risk areas: Immediate maintenance: Activate the emergency plan to shut down the equipment, perform maintenance, replace faulty parts, or take other measures to prevent further damage to the equipment or cause a larger failure; Shutdown processing: If the equipment failure has reached the point where it cannot be repaired, it may be necessary to shut down the equipment and start spare parts replacement or emergency repair procedures; Report generation: Generate detailed risk reports to relevant departments or management, analyze the root causes of problems, and develop long-term improvement measures to prevent similar failures from happening again.

[0087] In summary, the present invention achieves comprehensive monitoring of the factory operation and maintenance status and accurate fault risk prediction by integrating multi-dimensional data collection and preprocessing, intelligent feature extraction, knowledge graph construction, and an intelligent operation and maintenance prediction model that integrates deep reinforcement learning and long short-term memory networks. It not only improves the automation and intelligence level of operation and maintenance management, but also effectively prevents equipment failures and reduces unplanned downtime through early warning and optimized maintenance measures, thereby significantly improving the efficiency and reliability of factory operations, reducing maintenance costs, and enhancing the stability and safety of the production system.

[0088] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. Intelligent factory operation and maintenance management system driven by multi-dimensional data, characterized by: include, Data collection and preprocessing module, collects multi-dimensional plant operation and maintenance data, and preprocesses the multi-dimensional plant operation and maintenance data; The feature extraction module extracts features based on the pre-processed multi-dimensional plant operation and maintenance data through natural language processing technology and principal component analysis algorithm to obtain plant operation and maintenance features; The knowledge graph construction module builds the factory operation and maintenance knowledge graph based on the factory operation and maintenance characteristics in the RDF standard format; Model building module, which builds an intelligent operation and maintenance prediction model based on deep reinforcement learning model and long short-term memory network model; The fault risk value prediction module obtains the fault risk prediction value through the intelligent operation and maintenance prediction model based on the plant operation and maintenance knowledge graph; The risk assessment and early warning module generates early warning notifications and formulates maintenance measures based on the failure risk prediction value.

2. The intelligent factory operation and maintenance management system based on multi-dimensional data drive according to claim 1 is characterized in that: The multi-dimensional plant operation and maintenance data includes vibration frequency, power consumption, operation logs and fault reports; The preprocessing of the multi-dimensional plant operation and maintenance data includes preliminary cleaning, format conversion and standardization.

3. The intelligent factory operation and maintenance management system based on multi-dimensional data drive according to claim 2 is characterized by: Based on the pre-processed multi-dimensional plant operation and maintenance data, feature extraction is performed through natural language processing technology and principal component analysis algorithm to obtain plant operation and maintenance features. The specific steps are as follows: Based on the pre-processed multi-dimensional factory operation and maintenance data, factory text data and factory numerical data are extracted through data analysis methods; Through the word embedding method, the factory affairs text data is converted into word vector form; Based on the factory affairs text data in the form of word vectors, the global context feature vector of the factory affairs text data is obtained by aggregation through aggregation method; Based on the factory numerical data, the covariance matrix is ​​obtained by statistical analysis; The covariance matrix is ​​decomposed by the eigenvalue decomposition method to extract the principal component eigenvalues ​​and corresponding principal component eigenvectors in the factory numerical data; Based on the order of the extracted principal component eigenvalues, the principal component eigenvector corresponding to the largest principal component eigenvalue is used as the characteristic principal component; Map the factory affairs numerical data to the characteristic principal components to obtain the factory affairs characteristics after dimensionality reduction; Based on the global context feature vector of factory affairs text data and the factory affairs features after dimensionality reduction, the factory affairs operation and maintenance features are obtained by connecting them in vector order through the splicing method.

4. The intelligent factory operation and maintenance management system based on multi-dimensional data drive according to claim 3 is characterized by: Based on the factory operation and maintenance characteristics, the factory operation and maintenance knowledge graph is constructed through the RDF standard format. The specific steps are as follows: Based on the factory operation and maintenance characteristics, the factory operation and maintenance entity types, factory operation and maintenance entity attributes, and factory operation and maintenance entity relationships of the knowledge graph are defined through feature analysis and semantic abstraction methods to obtain the semantic structure of the factory operation and maintenance knowledge graph; Based on the factory affairs text data and factory affairs numerical data in the multi-dimensional factory affairs operation and maintenance data, they are mapped into attribute values ​​of factory affairs operation and maintenance entities through natural language processing methods and data analysis methods; Based on the semantic structure of the plant operation and maintenance knowledge graph and the attribute values ​​of the plant operation and maintenance entities, the plant operation and maintenance knowledge graph triples are constructed using the RDF standard format; Based on the plant operation and maintenance knowledge graph triples, it is organized into an RDF graph structure through GraphDB to obtain the plant operation and maintenance knowledge graph.

5. The intelligent factory operation and maintenance management system based on multi-dimensional data drive according to claim 4 is characterized in that: The intelligent operation and maintenance prediction model is constructed based on the deep reinforcement learning model and the long short-term memory network model. The specific steps are as follows: Through graph embedding method and time series annotation method, the historical plant operation and maintenance knowledge graph is converted into time series data format; The model is based on the long short-term memory network model; The input layer of the LSTM model receives the historical plant operation and maintenance knowledge graph converted into a time series data format; LSTM layer uses memory cells and gate mechanisms. The LSTM model selectively remembers and forgets information, capturing long-term dependencies in time series. The fully connected layer extracts features and reduces the dimension based on the output of the LSTM layer and maps it to the output layer; The output layer outputs preliminary prediction values; The deep reinforcement learning layer optimizes the decision-making process by using the reward function through interactive learning with the environment, adjusts the parameters of the deep reinforcement learning model, and optimizes the output results of the LSTM model; The deep reinforcement learning output layer outputs the final prediction value; Construct a preliminary intelligent operation and maintenance prediction model; Through the back propagation algorithm and reinforcement learning algorithm, the preliminary intelligent operation and maintenance prediction model is trained, evaluated and optimized, and finally an intelligent operation and maintenance prediction model is constructed.

6. The intelligent factory operation and maintenance management system based on multi-dimensional data drive according to claim 5 is characterized by: Through the back propagation algorithm and reinforcement learning algorithm, the preliminary intelligent operation and maintenance prediction model is trained, evaluated and optimized, and finally the intelligent operation and maintenance prediction model is constructed. The specific steps are as follows: The historical plant operation and maintenance knowledge graph converted into time series data format is divided into a training set and a validation set through random partitioning. The prediction performance of the LSTM model is evaluated by minimizing the loss function based on the training set, and the weights and parameters of the LSTM model are optimized through the back-propagation algorithm; Through the prediction value output by the LSTM model, the deep reinforcement learning model uses the policy gradient method to adjust the decision process according to the reward signal, adjust the output of the LSTM model, and optimize the fault risk prediction accuracy; Use the validation set to evaluate the prediction performance of the preliminary intelligent operation and maintenance prediction model by analyzing the accuracy, recall, and F1-score, and fine-tune the reward function, LSTM layer parameters, and reinforcement learning strategy based on the evaluation results; The k-fold cross-validation method is used to verify the initially trained intelligent operation and maintenance prediction model, and finally the intelligent operation and maintenance prediction model is output.

7. The intelligent factory operation and maintenance management system based on multi-dimensional data drive according to claim 6 is characterized by: Based on the plant operation and maintenance knowledge graph, the fault risk prediction value is obtained through the intelligent operation and maintenance prediction model. The specific steps are as follows: Based on the plant operation and maintenance knowledge graph converted into time series data format, the time series data is processed through the long short-term memory network model, and the deep reinforcement learning model is combined with the plant operation and maintenance knowledge graph to capture the complex dynamic relationships in plant operation and maintenance; Based on complex dynamic relationships, a multidimensional data-driven approach is used to obtain the failure risk prediction value at each time point. The expression is: ； in, Indicates time point The failure risk prediction value of represents the weight of the deep reinforcement learning model, Indicates that the DRL model is based on time points Factory Operation and Maintenance Knowledge Graph and time point Factory operation and maintenance status The predicted value of the output, represents the weight of the long short-term memory network model, Indicates that the LSTM model is based on time points To the time point The initial prediction value of the plant operation and maintenance knowledge graph output converted into time series data format, Indicates the length of the time period, Indicates time point Operation and maintenance feedback data, An index variable representing a time point.

8. The intelligent factory operation and maintenance management system based on multi-dimensional data drive according to claim 7 is characterized in that: LSTM model is based on time points To the time point The initial prediction value of the plant operation and maintenance knowledge graph output in the time series data format is converted into the following specific steps: The LSTM model analyzes the plant operation and maintenance knowledge graph converted into a time series data format to predict the plant operation and maintenance status at the current moment, and obtains the preliminary prediction value by weighted averaging the prediction results at historical moments. The expression is: ； in, Represents the number of historical moments, An index variable representing a historical moment, Indicates The weight coefficient of the LSTM model at each historical moment, express arrive The time interval is converted into a plant operation and maintenance knowledge graph in the time series data format. Indicates time point Move forward time plus time period , Indicates time point Move forward a moment.

9. The intelligent factory operation and maintenance management system based on multi-dimensional data drive according to claim 7 is characterized in that: DRL model is based on time point Factory Operation and Maintenance Knowledge Graph and time point Factory operation and maintenance status The output prediction value, the specific steps are as follows, The deep reinforcement learning model uses the plant operation and maintenance knowledge graph, combined with the complex dynamic relationship between status and behavior in historical data, to continuously improve the prediction ability of plant operation and maintenance behavior through the decision optimization process. It also captures the dynamic relationship between plant operation and maintenance status and behavior by analyzing the multi-dimensional plant operation and maintenance data at past moments, and obtains the prediction value by weighted averaging the prediction results at historical moments. The expression is: ； in, Indicates The weighted coefficient of the DRL model at each historical moment, Indicates time point The knowledge graph of factory operation and maintenance at all times, Indicates time point The factory operation and maintenance status at all times.

10. The intelligent factory operation and maintenance management system based on multi-dimensional data drive according to claim 7, characterized in that: The specific steps of generating early warning notification and formulating maintenance measures based on the fault risk prediction value are as follows: Based on historical failure risk data, cluster analysis is used to set the upper risk threshold. and the lower risk threshold ; Based on the failure risk prediction value , passing the upper risk threshold and the lower risk threshold , assess the failure risk level of plant operation and maintenance; when When , it indicates that the failure risk of plant operation and maintenance is in the low failure risk zone; when When , it indicates that the failure risk of plant operation and maintenance is in the medium failure risk zone; when When , it indicates that the failure risk of plant operation and maintenance is in the high failure risk zone; Based on the fault risk level of plant operation and maintenance, real-time tracking and decision support of equipment operation status and potential fault risks are carried out, and early warning notifications are automatically generated; Based on the fault risk prediction value and early warning notification, and combined with the equipment operating environment and equipment status, formulate corresponding maintenance measures.

Images