Theme labeling method and system based on dynamic LDA

Abstract

The invention discloses a dynamic LDA-based theme labeling method and system, and the method comprises the steps: collecting historical power operation and maintenance work order data from a plurality of heterogeneous data sources, carrying out the preprocessing operation, and constructing a professional dictionary in the power field; dividing the timestamp information data into time slices according to the timestamp information data, inputting the power operation and maintenance work order data of each time slice as prior information into the particle filter-based dynamic LDA model to train the model and optimize model parameters, and storing the trained particle filter-based dynamic LDA model corresponding to the current time slice, and the dynamic LDA model is used for training a next time slice, and performing theme labeling on new power operation and maintenance work order data by using the trained dynamic LDA model based on particle filtering. According to the method, accurate topic labeling of the text in the power field is realized.

Description

Technical Field

[0001] The invention belongs to the field of artificial intelligence, and in particular relates to a topic tagging method and system based on dynamic LDA. Background Art

[0002] The safe and stable supply of power systems is of vital importance. The power system is a large and complex system, and its safe and stable operation is inseparable from efficient operation and maintenance management. As an important carrier for recording the operating status, fault information, and maintenance records of power equipment, power operation and maintenance work orders contain a wealth of power system operation and maintenance knowledge. How to mine effective information from massive power operation and maintenance work order text data to provide decision support for the safe and stable operation of the power system is a key issue that needs to be urgently addressed in the current power system operation and maintenance field.

[0003] Traditional power operation and maintenance work order analysis mainly relies on manual experience, which has problems such as low efficiency, large errors, and difficulty in dealing with massive data, making it difficult to meet the higher requirements for power operation and maintenance in the smart grid era. In recent years, with the rapid development of artificial intelligence technology, it is necessary to apply natural language processing (NLP) technology to power operation and maintenance work order analysis to realize intelligent applications such as fault diagnosis, trend prediction, and knowledge extraction.

[0004] In the field of text mining, a large amount of data is unstructured, and it is difficult to directly obtain relevant and expected information from the information. Topic Model is an important tool for text mining. It can identify themes in documents and mine hidden information in corpus. It is widely used in scenarios such as topic aggregation, information extraction from unstructured text, and feature selection. In recent years, it has received a lot of attention in industry and academia. Latent Dirichlet Allocation (LDA) is one of the most representative models. LDA was proposed by Blei, David M., Ng, Andrew Y., and Jordan in 2003 to infer the topic distribution of documents. It can discover hidden topic patterns in the corpus; annotate documents according to themes; and organize, sort, summarize and retrieve documents using annotations. However, the traditional LDA model is difficult to handle the dynamic and professional characteristics of power operation and maintenance work order data, which limits its application in the field of power operation and maintenance. Summary of the invention

[0005] In order to solve the deficiencies in the prior art, the present invention provides a topic tagging method and system based on dynamic LDA to solve the technical problem of accurate topic tagging of text data in the power field.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions.

[0007] The present invention first discloses a topic tagging method based on dynamic LDA, which comprises the following steps:

[0008] Step 1: Collect historical power operation and maintenance work order data from multiple heterogeneous data sources, convert the text data in the collected historical power operation and maintenance work order data into a unified format, perform preprocessing operations, and build a professional dictionary in the power field;

[0009] Step 2: According to the timestamp information of the power operation and maintenance work order data, the power operation and maintenance work order data is divided into different time slices, and the power operation and maintenance work order data of each time slice is input as prior information into the dynamic LDA model based on particle filtering to train the model, wherein the model training adopts an incremental Gibbs sampling algorithm to sample new data, update particle weights and model parameters, and realize incremental update of the model;

[0010] Step 3: Optimize the model parameters, store the trained particle filter-based dynamic LDA model corresponding to the current time slice for training in the next time slice, and use the trained particle filter-based dynamic LDA model to perform topic labeling on new power operation and maintenance work order data.

[0011] The present invention further includes the following preferred embodiments:

[0012] The data sources include internal data of the State Grid and external public data. The internal data of the State Grid includes:

[0013] Technical standards, specifications, and regulations text data, used to define the operation and maintenance specifications of power equipment;

[0014] Power equipment operation and maintenance manual, including the equipment's structure, principle, common faults and treatment methods;

[0015] Fault diagnosis case library, which contains detailed records of historical fault cases, including fault phenomena, cause analysis, and treatment measures;

[0016] The external public data includes:

[0017] Power industry news, including power system operation, new technology applications, and fault events;

[0018] Technical forum discussion thread, including analysis and discussion of fault cases by power industry experts and engineers.

[0019] The pre-processing operation further comprises:

[0020] First, data cleaning is performed to remove irrelevant information, including HTML tags, special characters, and punctuation marks, and the data format is unified; data duplication is performed, and missing fields are filled or deleted;

[0021] Then, we use the electric power professional word segmentation tool to perform word segmentation processing, and use the new word discovery algorithm to identify new words, professional terms and equipment models in the electric power field, and add them to the word segmentation dictionary;

[0022] Construct a stop word dictionary to remove words whose contribution to text semantics is below the threshold, and add stop words specific to the field of power operation and maintenance work orders;

[0023] Use stemming algorithms to convert words with different word forms into the same stem form;

[0024] The word segmentation results are tagged with part-of-speech tagging tools to identify the grammatical properties of words; named entity recognition tools are used to identify named entities in the text and professional entities in the power field.

[0025] The step of inputting the power operation and maintenance work order data of each time slice as prior information into the dynamic LDA model based on particle filtering further includes:

[0026] Take the professional dictionary in the power field as prior knowledge, integrate the dictionary information into the model, and give higher weight to the professional words in the dictionary when calculating the word frequency; or set a higher prior probability for the professional words in the dictionary in the Dirichlet prior distribution of the topic-word distribution; or use the dictionary information to construct the topic-word association matrix as the prior information of the topic-word distribution to guide the model to learn the topic-word distribution of the power field knowledge;

[0027] Collect relevant information about power equipment, and use knowledge representation and reasoning technology to construct this information into a knowledge graph of power equipment; based on the power equipment knowledge graph, extract the semantic features of entities in the power operation and maintenance work order, and use these semantic features as input features of the PF-DTM model and integrate them into the bag-of-words representation of the document;

[0028] Use the relationship information in the knowledge graph to establish association constraints between topics;

[0029] Collect and organize the experience and rules of power operation and maintenance experts in fault diagnosis and fault handling, and convert the expert experience into rules that can be understood and executed by computers; integrate the expert rules as constraints into the training process of the PF-DTM model to guide the model to learn the distribution and evolution laws of topics.

[0030] The incremental Gibbs sampling algorithm samples new data, updates particle weights and model parameters, and implements incremental updates of the model, further comprising:

[0031] Step 2.1: For the data of the first time slice, use the standard PF-DTM model for training to obtain the initial topic-word distribution , document-topic distribution θ, and particle set;

[0032] Step 2.2: For a new time slice t, the particle set and model parameters of the previous time slice t-1 are used as prior information. For each particle, other particles and model parameters are fixed unchanged. Based on the data of the current time slice and the prior information, the Gibbs sampling algorithm is used to sample and update the topic allocation of the particle.

[0033] Step 2.3: Re-estimate the topic-word distribution based on the updated particle set and the parameters of the state transfer function g, recalculate the weight of each particle according to the updated model parameters; resample according to the particle weights to generate a new particle set for the incremental update of the next time slice;

[0034] Step 2.4: Repeat steps 2.2 and 2.3 until all time slices of data are processed.

[0035] The step 2.3 further comprises:

[0036] Randomly generate N particles for each time slice, each particle represents a possible topic distribution θt; initialize the weights of all particles to 1 / N, indicating that each particle has the same probability in the initial state;

[0037] For each time slice t, according to the observation equation, the probability of the subject distribution represented by each particle generating observation data is calculated as the weight of each particle; resampling is performed according to the weight of the particle to generate new N particles;

[0038] Estimate the topic distribution θt of time slice t based on the resampled particles;

[0039] Based on the estimated topic distribution θt and observed data, update the model parameters, including the topic-word distribution and parameters of the state transition function g;

[0040] Repeat the above steps until the model converges.

[0041] The optimization of the model parameters further comprises:

[0042] Select the time slice size Δt according to the data characteristics and analysis objectives;

[0043] Use perplexity and topic consistency metrics to evaluate the effects of different number of topics K and select the best value for the number of topics;

[0044] Select the number of particles N based on model accuracy and computational efficiency;

[0045] The state transfer function g is chosen based on prior knowledge of the data.

[0046] The present invention also discloses a dynamic LDA-based topic tagging system using the above-mentioned dynamic LDA-based topic tagging method, comprising:

[0047] The data collection module is used to collect historical power operation and maintenance work order data from multiple heterogeneous data sources, convert the text data in the collected historical power operation and maintenance work order data into a unified format, perform preprocessing operations, and build a professional dictionary in the power field;

[0048] A dynamic LDA model training module is used to divide the power operation and maintenance work order data into different time slices according to the timestamp information of the power operation and maintenance work order data, and input the power operation and maintenance work order data of each time slice as prior information into a dynamic LDA model based on particle filtering to train the model, wherein the model training adopts an incremental Gibbs sampling algorithm to sample new data, update particle weights and model parameters, and realize incremental update of the model;

[0049] The topic labeling module is used to optimize the model parameters, store the trained particle filter-based dynamic LDA model corresponding to the current time slice for training the next time slice, and use the trained particle filter-based dynamic LDA model to label the new power operation and maintenance work order data.

[0050] Accordingly, the present application also discloses a terminal, including a processor and a storage medium;

[0051] The storage medium is used to store instructions;

[0052] The processor is used to operate according to the instructions to execute the steps of the aforementioned topic tagging method based on dynamic LDA.

[0053] Correspondingly, the present application also discloses a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the aforementioned topic tagging method based on dynamic LDA are implemented.

[0054] The beneficial effect of the present invention is that, compared with the prior art, the present invention provides a topic labeling method and system based on dynamic LDA, which combines the LDA model with the particle filtering algorithm and introduces an incremental learning mechanism, so that it can effectively capture the evolution trend of power operation and maintenance work order topics over time, and at the same time combines professional dictionaries in the power field, power equipment knowledge graphs, power expert experience and other field knowledge to further improve the accuracy and practicality of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1It is a flow chart of building a dynamic LDA model in the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention more clear, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0057] The embodiments described in this application are only some embodiments of the present invention, not all embodiments. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the protection scope of the present invention.

[0058] The text data of power operation and maintenance work orders is different from general text data and has the following significant characteristics:

[0059] (1) The text of the power operation and maintenance work order is filled with a large number of power professional terms, equipment models, operating parameters, etc., such as "transformer", "circuit breaker", "overload", "ground fault", etc., which require professional knowledge of the power system to accurately understand their meaning.

[0060] (2) The operating status of the power system changes over time, and the subject information reflected in the power operation and maintenance work order also shows a trend of dynamic evolution. For example, high temperature in summer may cause frequent overload failures of power lines, while in winter, low temperature operation of equipment is more likely to occur.

[0061] (3) The data sources of power operation and maintenance work orders are diverse, including user repair work orders, system alarm work orders, inspection record work orders, etc. Work orders from different sources have significant differences in text format, content structure, professional terminology, etc.

[0062] (4) The amount of data in power operation and maintenance work orders is huge, but the truly valuable information is often hidden in massive text descriptions and needs to be extracted through effective text mining technology.

[0063] In view of the shortcomings of the prior art, the present invention proposes a topic labeling method and system based on dynamic LDA. On the basis of the existing LDA topic mining, combined with the characteristics of the power industry, a dynamic LDA topic model is constructed, and the dynamic LDA model is used to achieve accurate topic labeling of text data in the power field. Targeted analysis is carried out in combination with the characteristics of power grid user operation and maintenance work orders and system operation and maintenance work orders, and multi-source heterogeneous data such as national grid standards and specifications, expert systems, knowledge base systems, social networks, etc. are integrated to construct a text corpus in the power field. Based on the LDA model, combined with the time series analysis method, a dynamic LDA topic model is constructed, and the topic distribution and vocabulary distribution of the time slice text set are dynamically estimated through the incremental Gibbs sampling algorithm. The dynamic LDA model is used to label the text data in the power field and explore the potential topic evolution trend, aiming to build an efficient and accurate power operation and maintenance work order topic labeling system to provide technical support for power system fault diagnosis, trend prediction, knowledge discovery, etc.

[0064] LDA (Latent Dirichlet Allocation) is an unsupervised probabilistic topic model used to discover potential topic information from a document collection. The basic idea is to assume that each document is a mixture of multiple topics, and each topic is composed of multiple words with a specific probability. The goal of the LDA model is to infer the potential topic structure by analyzing the co-occurrence relationship of words in the document and give the probability distribution of documents on different topics.

[0065] The mathematical representation of the LDA model is as follows:

[0066] Document collection: D = {d 1 , d 2 , ..., d M}, where M represents the total number of documents, d i Represents the i-th document.

[0067] Theme set: T = {t 1 , t 2 , ..., t K}, where K represents the total number of topics, t k represents the kth topic.

[0068] Word set: W = {w 1 , w 2 , ..., w V}, where V represents the total number of words, w v Indicates the vth word.

[0069] LDA model assumptions:

[0070] Each document i It can be expressed as a mixture distribution θ of K topicsi ={θ i1 ,θ i2 , ..., θ iK}, where θ ik Represents document d i Belong to the topic k probability.

[0071] Each theme k It can be expressed as the probability distribution of V words in Word w v Belong to the topic k probability.

[0072] The generation process of the LDA model can be described as:

[0073] For each document d i :

[0074] From the Dirichlet distribution D ir (α) to sample a topic distribution θ i .

[0075] For document d i Each word in w j :

[0076] From the topic distribution θ i Sampling a topic z j .

[0077] From topic z j Corresponding word distribution Sample a word w from j .

[0078] Repeat the above process until all documents are generated.

[0079] The parameters α and β of the LDA model control the sparsity of the document-topic distribution and topic-word distribution respectively. The larger the α, the more uniform the topic distribution of the document; the larger the β, the more uniform the word distribution of the topic.

[0080] The training of the LDA model iteratively updates the model parameters so that the model can better explain the observed document data.

[0081] The LDA model can be used for topic analysis of power operation and maintenance work orders, for example:

[0082] Fault type identification: Power operation and maintenance work orders are classified according to themes, such as "transformer failure", "line failure", "circuit breaker failure", etc., to help operation and maintenance personnel quickly identify fault types and improve fault handling efficiency.

[0083] Example 1:

[0084] Work Order 1: "10kV#3 line tripped, protection action information is current quick break, preliminary judgment is that the line is short-circuited."

[0085] Work Order 2: "The user reported that the voltage at home was too low. After investigation, it was found that the transformer in the substation was faulty."

[0086] The LDA model can classify these two work orders into the topics of "line fault" and "transformer fault", respectively, helping operation and maintenance personnel to quickly identify the fault type.

[0087] Fault cause analysis: By analyzing the words with high probability of appearing under each topic, we can dig out the common causes of different types of faults. For example, if the words with high probability of appearing under the topic of "line fault" include "tree obstacle", "foreign object", "lightning strike", etc., it indicates that the main cause of the line fault may be caused by external environmental factors.

[0088] Example 2:

[0089] Work Order 3: "The heavy rain caused the tree barrier of the 10kV#5 line to trip. The line load has been transferred and emergency repairs are currently being organized."

[0090] Work Order 4: "The thunderstorm caused the 10kV #8 line to trip due to lightning strikes. Personnel have been arranged to patrol the line and check the fault point."

[0091] The LDA model can identify keywords related to line faults, such as "tree barriers" and "lightning strikes", to help operation and maintenance personnel analyze the causes of faults.

[0092] Fault trend prediction: By analyzing the changing trends of topic distribution in different time periods, future fault development trends can be predicted, such as which types of faults are likely to increase and which areas have a higher risk of failure, thereby providing support for the operation and maintenance decisions of the power system.

[0093] Example 3: If the LDA model finds that the probability of topics related to "transformer overload" has increased significantly recently, it means that the risk of transformer overload operation is increasing, and operation and maintenance personnel need to take timely measures to avoid large-scale power outages.

[0094] The traditional LDA model assumes that the document collection is static, that is, all documents are from the same time period. However, the data of power operation and maintenance work orders is dynamic, and the traditional LDA model has difficulty capturing the evolution of topics over time. The traditional LDA model only treats words as independent symbols, ignoring the semantic relationship between words and professional knowledge in the power field, which limits the model's ability to understand the text of power operation and maintenance work orders.

[0095] In order to solve the limitation that the traditional LDA model is difficult to process dynamic text data, the present invention uses the dynamic LDA model (Dynamictopic Model, DTM) to introduce time information into the LDA model, assuming that documents in different time periods have different topic distributions, and there is a certain evolutionary relationship between topic distributions. By introducing time information, the DTM model can effectively capture the evolution of topics over time, such as the emergence of new topics, the disappearance of old topics, the fusion and splitting between topics, etc. The DTM model can better distinguish topics in different time periods by considering the evolutionary relationship of topics, thereby improving the accuracy of topic identification. By analyzing the evolutionary trajectory of the topic, potential evolutionary patterns, such as periodic changes, trend changes, etc., can be found to provide support for decision-making.

[0096] The present invention adopts a dynamic LDA model to perform topic labeling on power operation and maintenance work orders. First, data collection and preprocessing are performed to collect power operation and maintenance work order data, and preprocessing operations such as data cleaning, word segmentation, and stop word removal are performed to prepare for subsequent topic modeling. Then, according to the timestamp information of the power operation and maintenance work order, the data is divided into different time slices, such as by day, week, month, etc., so that the dynamic LDA model can capture the evolution trend of the topic. Select a suitable dynamic LDA model and use the preprocessed power operation and maintenance work order data for model training. Analyze the evolution trend of topics in different time slices, such as the emergence of new topics, the disappearance of old topics, the fusion and splitting between topics, etc., and explain the evolution law of topics in combination with the power system operation and maintenance knowledge. Use the trained dynamic LDA model to label new power operation and maintenance work orders, and perform applications such as fault diagnosis, trend prediction, and knowledge discovery based on the labeling results.

[0097] The present invention adopts a dynamic LDA model PF-DTM based on particle filtering. Particle filtering is a non-parametric Bayesian inference method that can effectively handle nonlinear and non-Gaussian dynamic systems. By representing the topic distribution of each time slice as a group of particles and iteratively updating the particles using a particle filtering algorithm, the evolution patterns of the appearance, disappearance, fusion, and splitting of the topic can be accurately captured.

[0098] The present invention introduces an incremental learning mechanism into the PF-DTM model, so that the model can continuously update the topic model according to new data on the basis of retaining existing knowledge without retraining the entire model. The present invention adopts an incremental Gibbs sampling algorithm for model training. In each new time slice, the particle set and model parameters of the previous time slice are used as prior information, and the incremental Gibbs sampling algorithm is used to sample the new data, update the particle weights and model parameters, and realize the incremental update of the model. This incremental learning mechanism can significantly improve the training efficiency of the model and enable the model to track the evolution trend of the topic in real time.

[0099] See also Figure 1 As shown, the topic tagging method based on dynamic LDA disclosed in the present invention includes the following steps:

[0100] Step 1: Collect historical power operation and maintenance work order data from multiple heterogeneous data sources, convert the text data in the collected historical power operation and maintenance work order data into a unified format, perform preprocessing operations, and build a professional dictionary in the power field.

[0101] In the data preparation stage, data collection and preprocessing of heterogeneous data sources are carried out to prepare for the subsequent PF-DTM model construction and experimental analysis. The data sources include internal data of the State Grid and external public data.

[0102] The internal data of the State Grid includes text data such as technical standards, specifications, and procedures, which are used to define the operation and maintenance specifications of power equipment. They contain rich domain knowledge and help the model understand the professional terms and fault phenomenon descriptions in the work order; the power equipment operation and maintenance manual contains information such as the structure, principle, common faults and treatment methods of the equipment, which can be used to build a knowledge graph of power equipment and enhance the model's understanding of fault information; the fault diagnosis case library contains detailed records of historical fault cases, including fault phenomena, cause analysis, treatment measures, etc., which can be used for model training and evaluation.

[0103] External public data includes power industry news, including information on power system operation, new technology applications, fault events, etc., which helps the model capture the evolution trend of the topic; technical forum discussion posts, including analysis and discussion of fault cases by power industry experts and engineers, can be used to supplement domain knowledge and expert experience.

[0104] In the preprocessing stage, the collected text data is converted into a unified format, such as txt or XML, to facilitate subsequent processing; unstructured data containing tables, pictures and other information is extracted and converted into text form or structured data for model processing;

[0105] First, clean the data and remove irrelevant information, including HTML tags, special characters, punctuation marks, etc. Unify the data format, such as date and time formats, units, etc. Remove duplicate or highly similar work order records through data deduplication. Finally, fill in or delete missing fields;

[0106] Then, we use Jieba and other electric power professional word segmentation tools to perform word segmentation processing, and combine them with professional dictionaries in the electric power field to improve the accuracy of word segmentation. We use new word discovery algorithms to identify new words, professional terms, and equipment models in the electric power field, and add them to the word segmentation dictionary to improve the model's sensitivity to new information;

[0107] Construct a stop word dictionary to remove words with a semantic contribution degree lower than the threshold in the text, such as "de", "shi", "le", etc. Then, according to the characteristics of power operation and maintenance work orders, add domain-specific stop words, such as some common device operation status description words, etc.;

[0108] Adopt a stemming algorithm to convert words with different word forms into the same stem form. For example, convert "tripping", "opening", "closing" into "trip" to reduce the vocabulary size and improve the model efficiency;

[0109] Use a part-of-speech tagging tool to perform part-of-speech tagging on the word segmentation results to identify the grammatical attributes of words, such as nouns, verbs, adjectives, etc., providing a basis for subsequent feature selection; Use a named entity recognition tool to identify named entities in the text, such as personal names, place names, organization names, and professional entities in the power field, such as device names, fault types, operation parameters, etc.

[0110] Step 2: According to the timestamp information of the power operation and maintenance work order data, divide the power operation and maintenance work order data into different time slices, and use the power operation and maintenance work order data of each time slice as prior information to input into a dynamic LDA model based on particle filter to train the model. Among them, the model training uses an incremental Gibbs sampling algorithm to sample new data, update the particle weights and model parameters, and achieve the incremental update of the model.

[0111] In order to further improve the accuracy and interpretability of the model, the present invention integrates power domain knowledge into each link of the PF-DTM model, realizing the multi-level integration of power domain knowledge.

[0112] First, collect and organize materials such as power industry standards, specifications, and equipment manuals, extract vocabulary such as professional terms, equipment models, and fault types, and construct a power domain professional dictionary. This dictionary can be used in preprocessing links such as word segmentation, part-of-speech tagging, and named entity recognition, and as prior knowledge to be integrated into the learning process of the topic-word distribution.

[0113] Use the power domain professional dictionary as prior knowledge to guide the model in learning the topic-word distribution When doing so, increase the attention to professional vocabulary. In specific embodiments, the dictionary information can be integrated into the model in various ways. For example, when calculating the word frequency of words, assign a higher weight to the professional vocabulary in the dictionary; or in the Dirichlet prior distribution of the topic-word distribution, set a higher prior probability for the professional vocabulary in the dictionary; or use the dictionary information to construct a topic-word association matrix as the prior information of the topic-word distribution to guide the model to learn a topic-word distribution more in line with power domain knowledge.

[0114] Collect relevant information about power equipment, such as equipment type, function, parameters, connection relationship, common faults, etc. Use knowledge representation and reasoning technology to build this information into a knowledge graph of power equipment. Based on the knowledge graph of power equipment, extract the semantic features of entities in the power operation and maintenance work order, such as equipment type, fault type, geographic location, etc. These semantic features can be used as input features of the PF-DTM model, for example, by integrating them into the bag-of-words representation of the document to enhance the model's understanding of the semantic information of the work order.

[0115] By using the relationship information in the knowledge graph, such as the connection relationship between devices, the fault propagation relationship, etc., we can establish association constraints between topics. For example, if two devices have a connection relationship in the knowledge graph, we can constrain the PF-DTM model to bring the topics related to these two devices closer together during the evolution process, or increase the probability of the two topics appearing at the same time during the particle sampling process.

[0116] Collect and organize the experience and rules of power operation and maintenance experts in fault diagnosis and fault handling, and convert these expert experiences into rules that can be understood and executed by computers, such as IF-THEN rules and decision trees. Incorporate expert rules as constraints into the training process of the PF-DTM model to guide the model to learn topic distribution and evolution laws that are more in line with the actual situation. For example:

[0117] According to expert rules, the probabilities of specific words under certain topics are adjusted, such as increasing the probability of words related to the cause of the fault under the topic; according to expert rules, the topic evolution path is constrained, such as limiting the possibility of merging or splitting between certain topics; or according to expert rules, the topic allocation of particles is screened during the particle sampling process, such as eliminating particles that do not meet the expert rules.

[0118] In the model training phase, the professional dictionary in the power field is integrated into the topic-word distribution as prior knowledge to guide the model to learn a topic-word distribution that is more in line with the knowledge in the power field. The entity information and relationship information in the power equipment knowledge graph are used to enhance the semantic understanding ability of the model, for example, through semantic feature extraction and topic association constraints. The experience of power experts is converted into rules or constraints and integrated into the model training process to guide the model to learn topic distribution and evolution laws that are more in line with the actual situation. By integrating power field knowledge into different links of the model, deep knowledge integration is achieved, which significantly improves the domain adaptability of the model and the accuracy of topic labeling.

[0119] The PF-DTM model introduces a particle filter algorithm to handle the dynamic changes of topic distribution over time. Particle filtering uses a set of "particles" to represent the possible states of the topic distribution θt in each time slice. Each particle represents a possible topic distribution and has a corresponding weight, which indicates how well the distribution fits the actual data. As new data arrives, particles are updated according to the observed data. Particles with high weights represent more reasonable topic distributions, while particles with low weights are eliminated, thereby tracking the evolution of topics.

[0120] Based on the PF-DTM model, a topic feature extraction method for operation and maintenance work order processing is designed. The document-topic distribution θ learned by the PF-DTM model is used to represent the topic features of the work order, that is, a K-dimensional vector, in which each element represents the probability that the work order belongs to the corresponding topic. This topic feature representation method can effectively capture the topic information of the work order and is used for subsequent similarity calculation and retrieval.

[0121] The particle filter-based dynamic LDA (PF-DTM) model integrates knowledge in the power field and introduces an incremental learning mechanism to improve the accuracy, efficiency and interpretability of the model, so that it can be effectively applied to the intelligent processing of power operation and maintenance work orders.

[0122] The present invention introduces an incremental learning mechanism into the PF-DTM model, so that the model can continuously update the topic model according to new data while retaining existing knowledge, without the need to retrain the entire model.

[0123] Specifically, the particle set and model parameters of the previous time slice are used as the prior information of the current time slice, and the incremental Gibbs sampling algorithm is used to sample new data, update the particle weights and model parameters, and realize the incremental update of the model.

[0124] The present invention adopts the incremental Gibbs sampling algorithm to realize the incremental learning of the PF-DTM model, and its main steps are as follows:

[0125] Step 2.1: For the data of the first time slice, use the standard PF-DTM model for training to obtain the initial topic-word distribution The document-topic distribution θ, and the collection of particles.

[0126] Step 2.2: For a new time slice t, the particle set and model parameters of the previous time slice t-1 are used as prior information. For each particle, other particles and model parameters are fixed unchanged. Based on the data of the current time slice and the prior information, the Gibbs sampling algorithm is used to sample and update the topic allocation of the particle.

[0127] From the topic distribution θ of the current time slice t t The topic z is sampled fromtj ,For each topic k, the probability of being selected in time slice (t) is θ t,k ;

[0128] P(z tj = k) = θ t,k ,k=1,2,...,K

[0129] Step 2.3: Re-estimate the topic-word distribution based on the updated particle set and the parameters of the state transfer function g; that is, through the subject z j The word distribution of φ zj Generate word w tj , re-estimate word w tj The generation probability of zj,v Indicates that word v in the vocabulary appears in topic z j The probability of:

[0130] P(w tj =v|z tj =z j )=φ zj,v ,v=1,2,...,V

[0131] According to the updated model parameters, recalculate the weight of each particle; update the word w tj With theme z j The joint probability of , using the Bayesian formula to combine the likelihood of the word and the prior of the topic:

[0132]

[0133] Where: P(w tj |φ zj ) is the word w tj In Topic z j The likelihood probability under tj |θ t ) is the theme z j Prior probability at time slice t;

[0134] Resample according to the topic distribution and particle weights to generate a new set of particles for incremental update of the next time slice:

[0135] where w tj,k is the weight of particle k, based on the observed word and the previous state; is the new particle obtained after resampling.

[0136] Step 2.4: Repeat steps 2.2 and 2.3 until all time slices of data are processed.

[0137] The PF-DTM model uses a state space model to describe the dynamic evolution of topic distribution. The topic distribution θ of time slice t is described by t From the previous time slice θ t-1 The state equation of the evolution process is:

[0138] θ t =g(θ t-1 ,ε t )

[0139] Where g(·) represents the state transfer function, which can be a linear or nonlinear function, such as a linear Gaussian model, a nonlinear dynamic system model, etc., which is used to capture the trend of topic evolution. ε t Represents state noise, which is usually assumed to obey Gaussian distribution and is used to simulate random fluctuations in topic evolution.

[0140] Use the description of the distribution of given topics θ t and word distribution φ zj Generate word w in time slice t under the condition tj The observation equation is:

[0141] w tj ~Mult(θ t ,φ zj )

[0142] where w tj represents the jth word in time slice t; φ zj Indicates the topic z j The corresponding word distribution, that is, the probability of occurrence of different words under the topic, is usually a vector of length V, where V is the vocabulary size.

[0143]

[0144] where w tj,v represents the frequency (0 or 1) of the jth word in word v in time slice t, and represents word w tj Is it equal to the vth word in the vocabulary? zj,v Represents the occurrence probability of word v corresponding to topic zj.

[0145] represents the product of all items with indices v from 1 to V.

[0146] Step 3: Optimize the model parameters, store the trained particle filter-based dynamic LDA model corresponding to the current time slice for training in the next time slice, and use the trained particle filter-based dynamic LDA model to perform topic labeling on new power operation and maintenance work order data.

[0147] The performance of the PF-DTM model is affected by multiple parameters and needs to be reasonably set and optimized according to the specific data set and application scenario.

[0148] The time slice size (Δt) determines the time granularity of the model's analysis of topic evolution. Choosing a smaller time slice can capture the short-term changes of the topic more finely, but it may increase the computational complexity and overfitting risk of the model. Choosing a larger time slice can reduce the computational complexity of the model, but may lose some important short-term information. It is necessary to choose an appropriate time slice size based on the data characteristics and analysis objectives, such as days, weeks, or months.

[0149] The number of topics (K) determines the number of potential topics that the model can identify. Too few topics will result in the model being unable to capture all the topic information in the data, while too many topics will make the model too complex and overfitting. Indicators such as perplexity and topic consistency can be used to evaluate the effects of different numbers of topics and select the best value.

[0150] The number of particles (N) indicates the number of samples used to approximate the posterior probability of the topic distribution. The more particles there are, the more accurate the estimation of the posterior probability is, but the computational complexity is also higher. It is necessary to balance the model accuracy and computational efficiency according to the actual situation and select the appropriate number of particles.

[0151] The state transfer function (g) describes the evolution of topic distribution over time. You can choose the appropriate function form based on the prior knowledge of the data, for example: Linear Gaussian model: suitable for scenarios where the topic evolution is relatively smooth. Nonlinear dynamic system model: suitable for scenarios where the topic evolution is relatively complex and there are nonlinear relationships. Gaussian process model: suitable for scenarios where the topic evolution law is difficult to presume.

[0152] The hyperparameters of the Dirichlet distribution include α and β. α controls the sparsity of the document-topic distribution. The larger the α value, the more uniform the topic distribution of the document, indicating that the document tends to contain more topics. β controls the sparsity of the topic-word distribution. The larger the β value, the more uniform the word distribution of the topic, indicating that the topic tends to contain more words. The α and β parameters can be optimized using methods such as cross-validation.

[0153] The training process of the PF-DTM model uses the particle filter algorithm to iteratively update the particle weights based on the observed data, and finally obtains the model parameters that accurately describe the topic evolution process.

[0154] 1. Randomly generate N particles for each time slice, each particle represents a possible topic distribution θt. Initialize the weights of all particles to 1 / N, indicating that each particle has the same probability in the initial state.

[0155] 2. For each time slice t, the weight of each particle is calculated based on the observed data (i.e., the text data of the power operation and maintenance work order within the time slice). The weight is calculated by calculating the probability of the subject distribution represented by each particle generating the observed data according to the observation equation. Resample according to the weight of the particle to generate new N particles. During the resampling process, particles with high weights are more likely to be copied multiple times, while particles with low weights may be eliminated.

[0156] 3. Estimate the topic distribution θt of time slice t based on the resampled particles. For example, the weighted average of all particles can be used as the estimated value of the topic distribution.

[0157] 4. Update model parameters, including topic-word distribution, based on the estimated topic distribution θt and observed data and the parameters of the state transfer function g. Parameter update can use methods such as maximum likelihood estimation and expectation maximization (EM) algorithm.

[0158] 5. Repeat steps 2-4 until the model converges.

[0159] Electric power operation and maintenance work orders can be divided into two categories: power grid user operation and maintenance work orders and system operation and maintenance work orders. There are certain differences between the two in text features and topic distribution, and different topic labeling strategies need to be adopted.

[0160] For the operation and maintenance work orders of power grid users, since the language expression is more colloquial, it is necessary to build a dictionary for user language for tasks such as word segmentation, part-of-speech tagging, and named entity recognition. According to the user's geographic location information, the user operation and maintenance work orders are clustered to analyze the characteristics of power problems in different regions. For example, the user operation and maintenance work orders in urban areas can be clustered together to analyze the fault characteristics of urban distribution networks; the user operation and maintenance work orders in rural areas can be clustered together to analyze the fault characteristics of rural power grids. Text classification technology is used to identify the user's electricity consumption type, and different topic models are constructed according to different types of users. For example, a topic model can be constructed for residential users and another topic model can be constructed for commercial users to more accurately identify the electricity demand and fault types of different types of users.

[0161] For system operation and maintenance work orders, it is necessary to build a professional dictionary in the power field for tasks such as word segmentation, part-of-speech tagging, and named entity recognition to improve the topic model's ability to recognize power professional terms. Use external knowledge bases, such as power equipment knowledge graphs, to enhance the semantic understanding ability of topic models. For example, the entity information and relationship information in the equipment knowledge graph can be integrated into the dynamic LDA model to improve the model's understanding of the power operation and maintenance work order text and the accuracy of topic recognition. Incorporate time series information into the dynamic LDA model to analyze the evolution of power system faults. For example, information on the changes in the operating parameters of power equipment over time can be used as input features of the dynamic LDA model to study the impact of different operating parameters on the occurrence of faults.

[0162] In the experimental verification phase, the constructed PF-DTM model was experimentally verified using real power operation and maintenance work order data, and compared with other baseline models to evaluate the effectiveness and advancement of the model.

[0163] First, the preprocessed power field text data is divided into training set, validation set and test set. Appropriate baseline models are selected for comparative experiments, such as the traditional LDA model, the dynamic LDA model based on Kalman filtering (KF-DTM), and the dynamic LDA model based on variational inference (VI-DTM).

[0164] The following evaluation metrics are selected to evaluate the performance of the model:

[0165] Perplexity: measures the model's ability to predict unseen data. The lower the perplexity, the better the model's prediction effect.

[0166] Topic Coherence: Evaluate the semantic consistency of topics to ensure that the words in each topic have good semantic relevance. Commonly used indicators include UMass, UCI, NPMI, etc.

[0167] Topic Diversity: Assess the degree of difference between different topics to avoid topics that are too similar or repeated;

[0168] Topic Stability: Evaluate the stability of the model in identifying topics at different time slices to avoid topic drift;

[0169] Use the training set data to train the PF-DTM model and the baseline model. Use the validation set data to adjust the model parameters, such as the number of topics, time window size, etc., and select the model with the best performance. Use the test set data to evaluate the final model and compare it with the baseline model. Compare the performance of the PF-DTM model and the baseline model in terms of perplexity, topic consistency, topic diversity, topic stability and other indicators.

[0170] Analyze whether the topics identified by the PF-DTM model are consistent with the knowledge in the power field, such as analyzing the semantic information of high-probability words under each topic, and whether the evolution trend of the topic over time is consistent with the operation law of the power system. Compare and analyze the differences in topics identified by different models, and the ability of different models to capture the evolution trend of topics.

[0171] In the result analysis and optimization stage, the performance of the PF-DTM model on different evaluation indicators and the impact of different model parameters on model performance are analyzed. The evolution trend of the topics identified by the PF-DTM model over time is analyzed, such as the emergence of new topics, the disappearance of old topics, the fusion and splitting of topics, etc., and the evolution law of topics is explained in combination with the knowledge of power system operation and maintenance. Typical power operation and maintenance work order cases are selected to analyze the topic labeling results of the PF-DTM model for these cases, as well as how the model captures the key information and evolution trends in the cases. According to the experimental results, the model parameters are further optimized, such as adjusting the number of topics, time window size, number of particles, etc. Try to use more complex functions to model the evolution process of topics, such as nonlinear dynamic system models, Gaussian process models, etc., to improve the model's ability to capture complex evolution trends. For example, try to use deep learning models to learn the vector representation of words in the power field, or build a more refined knowledge graph of power equipment to further improve the model's semantic understanding ability. In an optional embodiment, the PF-DTM model is combined with other models, such as a text classification model, a sequence labeling model, etc., to build a more complete power operation and maintenance work order analysis system.

[0172] The beneficial effect of the present invention is that, compared with the prior art, the present invention provides a topic labeling method and system based on dynamic LDA, which combines the LDA model with the particle filtering algorithm and introduces an incremental learning mechanism, so that it can effectively capture the evolution trend of power operation and maintenance work order topics over time, and at the same time combines professional dictionaries in the power field, power equipment knowledge graphs, power expert experience and other field knowledge to further improve the accuracy and practicality of the model.

[0173] Specifically, by introducing the particle filter algorithm, accurate modeling of the dynamic evolution of topic distribution is achieved, overcoming the limitation that the traditional LDA model is difficult to process dynamic text data. At the same time, the incremental Gibbs sampling algorithm is used for model training, so that the model can efficiently handle the characteristics of the continuous accumulation of power operation and maintenance work order data, without the need to retrain all historical data in each time slice, which significantly improves the training efficiency of the model. The power field-specific information such as the State Grid standards and specifications, expert systems, and knowledge base systems are converted into prior knowledge and integrated into the training process of the PF-DTM model, which improves the model's ability to recognize topics specific to the power field and enhances the model's domain adaptability. Specifically, the present invention explores three ways of knowledge integration: integrating professional dictionaries in the power field into the topic-word distribution to improve the model's ability to recognize professional terms. Using the entity information and relationship information in the knowledge graph of power equipment, the semantic understanding ability of the model is enhanced, for example, through semantic feature extraction and topic association constraints. The experience of power experts is converted into rules or constraints and integrated into the model training process to guide the model to learn topic distribution and evolution laws that are more in line with actual conditions. The expected experimental results show that compared with the traditional LDA model and other topic models, the PF-DTM model proposed in this paper can achieve higher accuracy, recall rate and F1 value in the power operation and maintenance work order topic labeling task, verifying the effectiveness and superiority of the model. At the same time, the PF-DTM model can identify more reasonable topic structures and evolution trends, and provide more accurate topic information for subsequent fault diagnosis, trend prediction and knowledge discovery tasks.

[0174] The present invention may be a system, a method and / or a computer program product. The present invention also discloses a dynamic LDA-based topic tagging system based on the aforementioned dynamic LDA-based topic tagging method, comprising:

[0175] The data collection module is used to collect historical power operation and maintenance work order data from multiple heterogeneous data sources, convert the text data in the collected historical power operation and maintenance work order data into a unified format, perform preprocessing operations, and build a professional dictionary in the power field;

[0176] A dynamic LDA model training module is used to divide the power operation and maintenance work order data into different time slices according to the timestamp information of the power operation and maintenance work order data, and input the power operation and maintenance work order data of each time slice as prior information into a dynamic LDA model based on particle filtering to train the model, wherein the model training adopts an incremental Gibbs sampling algorithm to sample new data, update particle weights and model parameters, and realize incremental update of the model;

[0177] The topic labeling module is used to optimize the model parameters, store the trained particle filter-based dynamic LDA model corresponding to the current time slice for training the next time slice, and use the trained particle filter-based dynamic LDA model to label the new power operation and maintenance work order data.

[0178] Based on the spirit of the present invention, those skilled in the art can easily think that a computer program product can be obtained based on the aforementioned dynamic LDA-based topic tagging method. The computer program product may include a computer-readable storage medium, which carries computer-readable program instructions for enabling a processor to implement various aspects of the present disclosure. That is, the present application also includes a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to perform the steps of the aforementioned dynamic LDA-based topic tagging method.

[0179] Computer readable storage medium can be a tangible device that can keep and store the instructions used by the instruction execution device. Computer readable storage medium can be, for example, - but not limited to - electrical storage device, magnetic storage device, optical storage device, electromagnetic storage device, semiconductor storage device or any suitable combination of the above. More specific examples (non-exhaustive list) of computer readable storage medium include: portable computer disk, hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disk read-only memory (CD-ROM), digital versatile disk (DVD), memory stick, floppy disk, mechanical encoding device, for example, punch card or groove protrusion structure with instructions stored thereon and any suitable combination of the above. Computer readable storage medium used here is not interpreted as instantaneous signal itself, such as radio wave or other free propagating electromagnetic wave, electromagnetic wave propagated by waveguide or other transmission medium (for example, light pulse by optical fiber cable) or electrical signal transmitted by wire.

[0180] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0181] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages-such as Smalltalk, C++, etc., and conventional procedural programming languages-such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network-including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA) may be personalized by utilizing the state information of a computer-readable program instruction, and the electronic circuit may execute a computer-readable program instruction, thereby realizing various aspects of the present disclosure.

[0182] Finally, 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 above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A topic tagging method based on dynamic LDA, characterized in that: The following steps are involved: Step 1: Collect historical power operation and maintenance work order data from multiple heterogeneous data sources, convert the text data in the collected historical power operation and maintenance work order data into a unified format, perform preprocessing operations, and build a professional dictionary in the power field; Step 2: According to the timestamp information of the power operation and maintenance work order data, the power operation and maintenance work order data is divided into different time slices, and the power operation and maintenance work order data of each time slice is input as prior information into the dynamic LDA model based on particle filtering to train the model, wherein the model training adopts an incremental Gibbs sampling algorithm to sample new data, update particle weights and model parameters, and realize incremental update of the model; Step 3: Optimize the model parameters, store the trained particle filter-based dynamic LDA model corresponding to the current time slice for training in the next time slice, and use the trained particle filter-based dynamic LDA model to perform topic labeling on new power operation and maintenance work order data.

2. The topic tagging method based on dynamic LDA according to claim 1, characterized in that: The data sources include internal data of the State Grid and external public data. The internal data of the State Grid includes: Technical standards, specifications, and regulations text data, used to define the operation and maintenance specifications of power equipment; Power equipment operation and maintenance manual, including the equipment's structure, principle, common faults and treatment methods; Fault diagnosis case library, which contains detailed records of historical fault cases, including fault phenomena, cause analysis, and treatment measures; The external public data includes: Power industry news, including power system operation, new technology applications, and fault events; Technical forum discussion thread, including analysis and discussion of fault cases by power industry experts and engineers.

3. The topic tagging method based on dynamic LDA according to claim 2 is characterized in that: The pre-processing operation further comprises: First, data cleaning is performed to remove irrelevant information, including HTML tags, special characters, and punctuation marks, and the data format is unified; data duplication is performed, and missing fields are filled or deleted; Then, we use the electric power professional word segmentation tool to perform word segmentation processing, and use the new word discovery algorithm to identify new words, professional terms and equipment models in the electric power field, and add them to the word segmentation dictionary; Construct a stop word dictionary to remove words whose contribution to text semantics is below the threshold, and add stop words specific to the field of power operation and maintenance work orders; Use stemming algorithms to convert words with different word forms into the same stem form; The word segmentation results are tagged with part-of-speech tagging tools to identify the grammatical properties of words; named entity recognition tools are used to identify named entities in the text and professional entities in the power field.

4. The topic tagging method based on dynamic LDA according to claim 3 is characterized in that: The step of inputting the power operation and maintenance work order data of each time slice as prior information into the dynamic LDA model based on particle filtering further includes: Take the professional dictionary in the power field as prior knowledge, integrate the dictionary information into the model, and give higher weight to the professional words in the dictionary when calculating the word frequency; or set a higher prior probability for the professional words in the dictionary in the Dirichlet prior distribution of the topic-word distribution; or use the dictionary information to construct the topic-word association matrix as the prior information of the topic-word distribution to guide the model to learn the topic-word distribution of the power field knowledge; Collect relevant information about power equipment, and use knowledge representation and reasoning technology to construct this information into a knowledge graph of power equipment; based on the power equipment knowledge graph, extract the semantic features of entities in the power operation and maintenance work order, and use these semantic features as input features of the PF-DTM model and integrate them into the bag-of-words representation of the document; Use the relationship information in the knowledge graph to establish association constraints between topics; Collect and organize the experience and rules of power operation and maintenance experts in fault diagnosis and fault handling, and convert the expert experience into rules that can be understood and executed by computers; integrate the expert rules as constraints into the training process of the PF-DTM model to guide the model to learn the distribution and evolution laws of topics.

5. The topic tagging method based on dynamic LDA according to claim 4 is characterized in that: The incremental Gibbs sampling algorithm samples new data, updates particle weights and model parameters, and implements incremental updates of the model, further comprising: Step 2.1: For the data of the first time slice, use the standard PF-DTM model for training to obtain the initial topic-word distribution φ, document-topic distribution θ, and particle set; Step 2.2: For a new time slice t, the particle set and model parameters of the previous time slice t-1 are used as prior information. For each particle, other particles and model parameters are fixed unchanged. Based on the data of the current time slice and the prior information, the Gibbs sampling algorithm is used to sample and update the topic allocation of the particle. Step 2.3: Based on the updated particle set, re-estimate the parameters of the topic-word distribution φ and the state transfer function g, and recalculate the weight of each particle based on the updated model parameters; resample the particle weights to generate a new particle set for the incremental update of the next time slice; Step 2.4: Repeat steps 2.2 and 2.3 until all time slices of data are processed.

6. The dynamic LDA-based topic tagging system according to claim 5, characterized in that: The step 2.3 further comprises: Randomly generate N particles for each time slice, each particle represents a possible topic distribution θt; initialize the weights of all particles to 1 / N, indicating that each particle has the same probability in the initial state; For each time slice t, according to the observation equation, the probability of the subject distribution represented by each particle generating observation data is calculated as the weight of each particle; resampling is performed according to the weight of the particle to generate new N particles; Estimate the topic distribution θt of time slice t based on the resampled particles; Update the model parameters, including the topic-word distribution φ and the parameters of the state transfer function g, based on the estimated topic distribution θt and the observed data; Repeat the above steps until the model converges.

7. The dynamic LDA-based topic tagging system according to claim 6, characterized in that: The optimization of the model parameters further comprises: Select the time slice size Δt according to the data characteristics and analysis objectives; Use perplexity and topic consistency metrics to evaluate the effects of different number of topics K and select the best value for the number of topics; Select the number of particles N based on model accuracy and computational efficiency; The state transfer function g is chosen based on prior knowledge of the data.

8. A topic tagging system based on dynamic LDA, characterized in that: include: The data collection module is used to collect historical power operation and maintenance work order data from multiple heterogeneous data sources, convert the text data in the collected historical power operation and maintenance work order data into a unified format, perform preprocessing operations, and build a professional dictionary in the power field; A dynamic LDA model training module is used to divide the power operation and maintenance work order data into different time slices according to the timestamp information of the power operation and maintenance work order data, and input the power operation and maintenance work order data of each time slice as prior information into a dynamic LDA model based on particle filtering to train the model, wherein the model training adopts an incremental Gibbs sampling algorithm to sample new data, update particle weights and model parameters, and realize incremental update of the model; The topic labeling module is used to optimize the model parameters, store the trained particle filter-based dynamic LDA model corresponding to the current time slice for training the next time slice, and use the trained particle filter-based dynamic LDA model to label the new power operation and maintenance work order data.

9. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to operate according to the instruction to execute the steps of the topic tagging method based on dynamic LDA according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the dynamic LDA-based topic tagging method described in any one of claims 1 to 7 are implemented.

Images