A method and system for analyzing and statistically processing electrical fire information based on a knowledge graph

Abstract

The present invention relates to the technical field of electrical fire information analysis and statistical scheme design, and particularly relates to a method and system for electrical fire information analysis and statistics based on a knowledge graph. Through the knowledge credibility evaluation model and optimization algorithm, the present invention effectively improves the accuracy of knowledge fusion, making the constructed knowledge graph more accurate and complete, and enabling better support for electrical fire information analysis and statistical work. The introduction of the intelligent reasoning engine can deeply mine the potential causal relationships and influencing factors of electrical fires, generate detailed and accurate analysis and statistical results, and provide more comprehensive information support for users. The establishment of the dynamic update mechanism ensures that the knowledge graph can reflect newly emerging electrical fire information and knowledge in real time, always maintaining the timeliness and effectiveness of the information, and greatly improving the practicability and reliability of electrical fire information analysis and statistics.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical fire information analysis and statistics scheme design, and particularly relates to a method and system for electrical fire information analysis and statistics based on a knowledge graph. Background Art

[0002] In the prior art, since the knowledge related to electrical fires comes from a wide range of sources, including equipment manuals, accident reports, industry standards, etc., when integrating these multi-source heterogeneous knowledge into a knowledge graph, it is difficult for the prior art to accurately identify and eliminate the conflicts and redundant information therein, resulting in the accuracy and integrity of the knowledge graph being affected. In the face of complex electrical fire scenarios, the reasoning ability of the existing solutions based on the knowledge graph is limited, and it is impossible to deeply mine the potential causal relationships and influencing factors of electrical fires, making it difficult to provide accurate and comprehensive analysis and statistics results. Moreover, due to the continuous update of technologies and standards in the electrical field, electrical fire cases are also constantly occurring and changing. Most of the existing knowledge graph construction schemes lack a dynamic update mechanism for the knowledge graph and cannot timely reflect the newly emerging electrical fire information and knowledge, thus affecting the effectiveness and timeliness of information analysis and statistics.

[0003] Therefore, the prior art still needs to be further developed. Summary of the Invention

[0004] The purpose of the present invention is to overcome the above technical deficiencies and provide a method and system for electrical fire information analysis and statistics based on a knowledge graph to solve the problems existing in the prior art.

[0005] To achieve the above technical objectives, according to the first aspect of the present invention, the present invention provides a method for electrical fire information analysis and statistics based on a knowledge graph, and the method includes:

[0006] S100. Data collection and preprocessing: Collect multi-source heterogeneous data related to electrical fires, including equipment specifications, accident reports, and industry standards; perform cleaning, deduplication, and normalization processing on the collected data to obtain a standardized data set;

[0007] S200. Knowledge extraction: Use natural language processing technology and machine learning algorithms to extract entities, attributes, and relationships related to electrical fires from the standardized data set to construct a preliminary knowledge graph;

[0008] S300. Knowledge fusion and evaluation: Introduce a knowledge credibility evaluation model to evaluate the credibility of the knowledge in the preliminary knowledge graph, calculate the credibility score of each piece of knowledge; fuse and optimize the conflicting and redundant knowledge according to the credibility score to generate an optimized knowledge graph;

[0009] S400. Construction of Intelligent Inference Engine: Construct an intelligent inference engine, which includes an inference algorithm and a rule base; use the intelligent inference engine to deeply infer the optimized knowledge graph, mine the potential causal relationships and influencing factors of electrical fires, and generate analysis and statistical results;

[0010] S500. Establishment of Dynamic Update Mechanism: Set up a real-time monitoring module to monitor the changes of knowledge sources and the emergence of new electrical fire cases in real time; when it is detected that the knowledge source is updated or a new case appears, use the knowledge update algorithm to integrate new knowledge and cases into the knowledge graph to achieve the dynamic update of the knowledge graph.

[0011] Specifically, in the data collection and preprocessing step, for the multi-source heterogeneous data of electrical fires collected, a cleaning algorithm based on regular expressions is used to remove noise information, data deduplication is performed through a hash algorithm, and the data is normalized according to the standard terms and formats in the electrical field. The standardized data set is classified and stored according to different application scenarios related to electrical fires.

[0012] Specifically, in the knowledge extraction step, for the entity extraction related to electrical fires, a conditional random field algorithm combined with an electrical field knowledge dictionary is used for identification. For attribute extraction, a support vector machine algorithm is used to classify and extract attribute values based on a large amount of training data. For relationship extraction, a relationship classification model based on deep learning is used. The models of natural language processing technology and machine learning algorithms are pre-trained through labeled data in the electrical fire field before extraction to improve the accuracy of extraction.

[0013] Specifically, in the knowledge fusion and evaluation step, the knowledge credibility evaluation model calculates the credibility score of each piece of knowledge by integrating the dimensions of the source credibility of the knowledge, the consistency with existing knowledge, and the frequency of citation. For the knowledge with a credibility score at the critical value, an artificial auxiliary review mechanism is introduced to further screen and evaluate the knowledge.

[0014] Specifically, in the step of constructing the intelligent inference engine, the inference algorithms of the intelligent inference engine include rule-based inference algorithms, Bayesian inference algorithms, and graph-based inference algorithms. The rules in the rule base are classified and stored according to the causes of electrical fires, including electrical equipment failure rules, environmental factor rules, and human operation rules. The inference process is carried out in a multi-round iterative and depth search manner.

[0015] Specifically, in the step of constructing the intelligent inference engine, the inference engine automatically annotates and supplements the knowledge graph with the inference results. For the newly mined potential causal relationships and influencing factors of electrical fires, corresponding entities, attributes, and relationships are automatically generated and added to the knowledge graph.

[0016] Specifically, in the step of establishing the dynamic update mechanism, the real-time monitoring module determines whether the knowledge source has changed by monitoring the web page structure change of the knowledge source and identifying the data version change. For the generation of new electrical fire cases, information collection is triggered by monitoring the new records in the electrical fire database. The newly acquired knowledge enters the knowledge update process after preliminary screening and format conversion.

[0017] Specifically, in the knowledge update step, the knowledge update algorithm includes an update method based on incremental learning and an update method based on fusion reconstruction. When the sum of the amount of new knowledge and the number of electrical fire cases is less than the first preset threshold, the incremental learning method is used to quickly update the knowledge graph; when the sum of the amount of new knowledge and the number of electrical fire cases is greater than or equal to the first preset threshold, the fusion reconstruction method is used to comprehensively update the knowledge graph.

[0018] Specifically, in the step of generating the analysis and statistical results, the generated analysis and statistical results are classified and displayed according to the electrical fire risk level, and the potential causal relationships and influencing factors of electrical fires are presented graphically through visualization tools. The graphical display includes node diagrams, flowcharts, and causal relationship diagrams.

[0019] According to the second aspect of the present invention, there is provided an electrical fire information analysis and statistical system based on a knowledge graph, including:

[0020] A data processing module: used to execute the data collection and preprocessing step, clean, deduplicate, and normalize the multi-source heterogeneous data of electrical fires to generate a standardized data set;

[0021] A knowledge extraction module: uses natural language processing techniques and machine learning algorithms to extract electrical fire-related entities, attributes, and relationships from the standardized data set to construct a preliminary knowledge graph;

[0022] A knowledge fusion and optimization module: evaluates the credibility of the knowledge in the preliminary knowledge graph through a knowledge credibility evaluation model, fuses and optimizes the conflicting and redundant knowledge, and generates an optimized knowledge graph;

[0023] An inference analysis module: constructs an intelligent inference engine, deeply infers the optimized knowledge graph, mines the potential causal relationships and influencing factors of electrical fires, and generates analysis and statistical results;

[0024] A dynamic update module: sets up a real-time monitoring module to monitor the changes of the knowledge source and the generation of new electrical fire cases in real time, and realizes the dynamic update of the knowledge graph through the knowledge update algorithm.

[0025] Beneficial effects:

[0026] Through the knowledge credibility evaluation model and optimization algorithm, the present invention effectively improves the accuracy of knowledge fusion, making the constructed knowledge graph more accurate and complete, and better supporting the analysis and statistics of electrical fire information. The introduction of the intelligent reasoning engine can deeply explore the potential causal relationships and influencing factors of electrical fires, generate detailed and accurate analysis and statistics results, and provide more comprehensive information support for users. The establishment of the dynamic update mechanism ensures that the knowledge graph can reflect newly emerging electrical fire information and knowledge in real time, always maintaining the timeliness and effectiveness of information, and greatly improving the practicality and reliability of electrical fire information analysis and statistics. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flowchart of the method for analyzing and statistics of electrical fire information based on a knowledge graph provided in a specific embodiment of the present invention;

[0028] Figure 2 is a schematic diagram of the system composition of the system for analyzing and statistics of electrical fire information based on a knowledge graph provided in a specific embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Based on the embodiments in this application, other similar embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application. In addition, the directional terms mentioned in the following embodiments, such as "upper", "lower", "left", "right", etc., are only references to the directions in the drawings. Therefore, the directional terms used are for illustration rather than limitation of the present invention.

[0030] The present invention will be further described below in conjunction with the drawings and preferred embodiments.

[0031] Please refer to Figure 1 , the present invention provides a method for analyzing and statistics of electrical fire information based on a knowledge graph, including:

[0032] S100. Data collection and preprocessing: Collect multi-source heterogeneous data related to electrical fires, including equipment manuals, accident reports, and industry standards; clean, deduplicate, and normalize the collected data to obtain a standardized data set.

[0033] It should be noted here that before step S100, it includes:

[0034] Preset a first preset threshold in the dynamic update module.

[0035] It is understandable that the first preset threshold can be set by the personnel of the present invention according to needs, and the present invention does not specifically limit the value of the first preset threshold here, as long as it is applicable to the method for analyzing and counting electrical fire information based on the knowledge graph proposed by the present invention.

[0036] Preferably, the present invention preferably sets the first preset threshold to 100. The above setting is obtained by those skilled in the art through a large number of tests, further improving the reliability of the knowledge dynamic update scheme of the present invention.

[0037] It is understandable that "the first preset threshold is preferably 100" is obtained through a large number of tests. The specific basis and acquisition method are as follows:

[0038] 1. Test data basis

[0039] During the R & D process, a large amount of electrical fire-related data was collected, and multiple experiments were conducted for different application scenarios and processing requirements. These experiments covered various types of data processing tasks, including but not limited to data cleaning, deduplication, fusion, etc., to comprehensively evaluate the impact of different thresholds on system performance and result accuracy.

[0040] 2. Performance index evaluation

[0041] In each experiment, multiple different thresholds were set, and a series of performance indexes were evaluated for each threshold. These performance indexes include but not limited to the accuracy, integrity, update efficiency of the knowledge graph, and the reliability of the analysis and statistical results, etc. By analyzing and comparing a large amount of experimental data, the change trends of various performance indexes under different thresholds were observed.

[0042] 3. Determine the preferred threshold

[0043] After comprehensively analyzing the experimental results, it is found that when the threshold is set to 100, the system shows a better balance in multiple performance indexes. For example, in terms of the accuracy of the knowledge graph, it can effectively identify and process most knowledge conflicts and redundancy problems, while maintaining a high knowledge fusion efficiency; in terms of the reliability of the analysis and statistical results, it can provide stable and accurate data support for subsequent reasoning and analysis. Therefore, considering various factors comprehensively, it is determined that the first preset threshold is preferably 100.

[0044] In summary, the technical solution of the present application has clear basis and detailed implementation methods in both the calculation steps of the knowledge credibility evaluation model and the determination of the first preset threshold, and there is no problem of insufficient disclosure.

[0045] Specifically, in the data collection and preprocessing step, for the collected multi-source heterogeneous electrical fire data, a cleaning algorithm based on regular expressions is used to remove noise information, data deduplication is performed through a hash algorithm, and the data is normalized according to the standard terms and formats in the electrical field. The standardized data set is classified and stored according to different application scenarios related to electrical fires.

[0046] Specifically, in the data collection and preprocessing step, for the collected multi-source heterogeneous electrical fire data, the following specific methods are used for preprocessing:

[0047] Data cleaning: The cleaning algorithm based on regular expressions is specifically designed to create a regular expression template for the characteristics of data in the electrical field. For example, for the specific character combinations that an electrical equipment model may contain (such as specific arrangements of letters and numbers), through regular expression matching and replacement operations, noise information such as comments, extra spaces, and line breaks in the data is removed. At the same time, for the irregular characters (such as garbled characters and special symbols) generated due to acquisition or input errors in the data, they are also identified and corrected through corresponding regular expression rules.

[0048] Data deduplication: The hash algorithm is used for data deduplication. For each piece of data related to electrical fires, its hash value is calculated, and the hash value is used as the unique identifier of the data. When storing the data, first check whether there is already data with the same hash value in the database. If it exists, it is regarded as duplicate data and deleted.

[0049] Data normalization: The data is normalized according to the standard terms and formats in the electrical field. For aspects such as electrical equipment names, fault types, and technical parameters, conversions are made with reference to the glossaries and format specifications formulated in relevant international and domestic standards (such as IEEE standards, national standards, etc.). For example, different names for the same type of electrical equipment provided by different equipment manufacturers are unified into standard terms, and the recording format of the fault occurrence time is unified into a specific date-time format (such as "YYYY-MM-DD HH:MM:SS"). The standardized data set is classified and stored according to different application scenarios related to electrical fires (such as electrical fires in industrial factories, electrical fires in residential areas, etc.). Independent data tables are established in the database for each application scenario to facilitate subsequent data processing and analysis.

[0050] Furthermore, in the preferred embodiment, when analyzing and statistically processing electrical fire information, the first problem faced is the processing of multi-source heterogeneous data. This embodiment collects electrical fire-related data from multiple aspects, and these data sources include but are not limited to detailed equipment manuals provided by various electrical equipment manufacturers, accident reports sorted and recorded by the fire department, professional industry standard documents issued by the electrical industry association, etc.

[0051] For the collected data, start a comprehensive data cleaning process. Take the product manual of a common wire and cable as an example. There may be some explanatory content in it, such as "This type of wire and cable is an experimental product and is not recommended for use under extreme conditions". Through a pre-designed regular expression template, such explanatory and annotative content can be accurately identified and removed, making the data cleaner.

[0052] In the data deduplication step, the hash algorithm is used to calculate the hash value of each piece of data. Suppose there are two technical parameter documents of a certain brand of miniature circuit breaker collected from different channels. Although the expressions are slightly different, the core information is the same. The hash algorithm will process these two documents separately and generate the same hash value, thus determining that these two documents are duplicate data and only one of them is retained.

[0053] After deduplication, in order to unify the form and standard of the data, data normalization is carried out. For example, for the description of the fault types of electrical equipment, there may be various expression ways in different data sources, such as "short circuit", "short connection", "circuit break", etc. By referring to the standard terms and formats in the electrical field (such as the term definitions of electrical faults in the IEEE standard and national standards), they are uniformly converted into the standard term "short circuit". At the same time, for the record of the fault occurrence time, it is standardized according to the format of "YYYY-MM-DD HH:MM:SS". The standardized data set is classified and stored according to the application scenarios, such as being divided into different data sets for industrial scenarios, residential life scenarios, etc., which is convenient for subsequent management and analysis.

[0054] S200, Knowledge Extraction: Using natural language processing technology and machine learning algorithms, extract entities, attributes, and relationships related to electrical fires from the standardized data set to construct a preliminary knowledge graph.

[0055] Specifically, in the knowledge extraction step, for the entity extraction related to electrical fires, the conditional random field algorithm is combined with the electrical field knowledge dictionary for identification. For attribute extraction, the support vector machine algorithm is used to classify and extract attribute values based on a large amount of training data. For relationship extraction, a relationship classification model based on deep learning is used. The models of the natural language processing technology and machine learning algorithms are pre-trained with the labeled data in the electrical fire field before extraction to improve the accuracy of extraction.

[0056] Specifically, in the knowledge extraction step, the specific algorithms and processes are as follows:

[0057] Entity extraction: For entity extraction related to electrical fires, the Conditional Random Field (CRF) algorithm is combined with an electrical domain knowledge dictionary for recognition. First, the electrical domain knowledge dictionary is constructed and preprocessed, and converted into a feature vector form recognizable by the CRF algorithm. The knowledge dictionary contains common electrical equipment names, fault types, electrical parameters, etc., and assigns specific labels (such as "electrical equipment", "fault type", "electrical parameter", etc.) to each word. Then, the CRF algorithm is used to extract features from the input text data, such as part-of-speech, context information, etc., and trained in combination with the feature vectors of the knowledge dictionary to finally identify the electrical fire-related entities in the text.

[0058] Attribute extraction: Attribute extraction uses the Support Vector Machine (SVM) algorithm to perform attribute value classification extraction based on large-scale training data. First, a large amount of labeled data is prepared, and the attributes and their values corresponding to each entity are labeled. For example, for the "wire and cable" entity, its attributes such as "conductivity" and "voltage withstand level" and their corresponding values are labeled. Then, the SVM classification model is trained using this labeled data. By learning the feature patterns of different attribute values, the model can classify and extract the attributes of entities in new texts.

[0059] Relationship extraction: Relationship extraction uses a deep learning-based relationship classification model, such as using a Recurrent Neural Network (RNN) or its variant Long Short-Term Memory Network (LSTM) to construct a relationship judgment model. During the training process of the model, text fragments containing entity pairs are input, and the relationship type between entities is judged by learning the context semantic information of the entities. At the same time, to avoid overfitting, an attention mechanism is introduced to enable the model to pay more attention to the key information related to relationship judgment. The models of the natural language processing technology and machine learning algorithms are pre-trained through the labeled data in the electrical fire field before extraction. During the pre-training process, the hyperparameters of the model are optimized through methods such as cross-validation to improve the accuracy of extraction.

[0060] Furthermore, in the preferred embodiment, knowledge extraction is a key step in constructing an accurate knowledge graph. In the entity extraction process, for the text data related to electrical fires, the Conditional Random Field (CRF) algorithm is combined with the electrical domain knowledge dictionary. First, the knowledge dictionary is preprocessed, and common electrical equipment names (such as transformers, distribution boxes, etc.), fault types (overcurrent fault, leakage fault, etc.), electrical parameters (rated power, insulation resistance, etc.) and other words are entered, and at the same time, corresponding labels are assigned to each word. When processing a description text of an electrical fire accident, "The transformer in a certain factory caused a short circuit and caught fire due to an overcurrent fault", the CRF algorithm will accurately identify entities such as "transformer", "overcurrent fault", and "short circuit" according to the feature vectors of the knowledge dictionary and the semantic features in the text, and assign corresponding labels to them.

[0061] Attribute extraction is done with the help of the support vector machine (SVM) algorithm. Prepare a large amount of data with labeled attributes and attribute values ​​as training samples. For example, for the "wire and cable" entity, there are samples with attributes and values ​​labeled with "conductivity is 40% IACS" and "voltage rating is 1000V". The SVM algorithm forms a classification model by learning the characteristic patterns of these samples. When faced with new text data, the model can classify and extract the attributes of the "wire and cable" entity.

[0062] Relation extraction relies on a relation classification model built using a deep learning-based recurrent neural network (RNN) and its variant, the long short-term memory network (LSTM). During the training process, a text segment containing entity pairs is input, such as "aging of electrical switches and electrical lines, which in turn cause electrical fires". The model learns contextual semantic information and determines that there is a causal relationship between "electrical switches", "aging of electrical lines" and "electrical fires". At the same time, the attention mechanism enables the model to focus on key information closely related to relationship judgment, such as the specific vocabulary "which in turn causes" in the text that describes the causal relationship, so as to extract relationships more accurately.

[0063] S300, knowledge fusion evaluation: Introduce a knowledge credibility evaluation model, conduct credibility evaluation on the knowledge in the preliminary knowledge graph, and calculate the credibility score of each piece of knowledge; fuse and optimize conflicting and redundant knowledge based on the credibility score to generate an optimized knowledge graph.

[0064] Specifically, in the knowledge fusion evaluation step, the knowledge credibility evaluation model calculates the credibility score of each piece of knowledge by comprehensively considering the source credibility of the knowledge, the consistency with the existing knowledge, and the frequency of citation. For knowledge whose credibility score is at a critical value, a manual assisted review mechanism is introduced to further screen and evaluate the knowledge.

[0065] Specifically, in the knowledge fusion evaluation step, the construction and calculation process of the knowledge credibility evaluation model is as follows:

[0066] Credibility score calculation: The knowledge credibility assessment model calculates the credibility score of each piece of knowledge based on multiple dimensions, such as the credibility of the source of the knowledge, the consistency with existing knowledge, and the frequency of citation. In terms of source credibility, different weights are assigned according to the source of knowledge (such as standards issued by authoritative organizations, works by well-known experts in professional fields, and strictly reviewed industry reports). For example, information released by authoritative organizations has a higher weight, while information in personal blogs has a relatively lower weight. In terms of consistency with existing knowledge, semantic similarity is calculated by comparing new knowledge with related knowledge already in the knowledge graph. Word vector models (such as Word2Vec, BERT, etc.) can be used to convert knowledge into vector representations, and then the similarity between vectors is calculated. In terms of citation frequency, the number of citations of knowledge in relevant academic literature, industry reports, etc. is counted, and the number of citations is used as an evaluation indicator. The credibility score of each piece of knowledge is calculated by weighted combination of the scores of these indicators.

[0067] Human-assisted review mechanism: For knowledge with a critical credibility score, a human-assisted review mechanism is introduced to further screen and evaluate the knowledge. An audit team consisting of electrical experts and professional information processing personnel is formed to assign suspicious knowledge to them. Experts judge the credibility of knowledge based on their domain expertise, while information processing personnel provide support from the perspective of data processing and text analysis. The audit team conducts in-depth analysis of the background information and relevant context involved in the suspicious knowledge, and ultimately determines the credibility score of the knowledge.

[0068] Furthermore, in a preferred embodiment, the knowledge fusion evaluation link is intended to ensure the accuracy and consistency of the knowledge in the knowledge graph. For each piece of knowledge, the credibility evaluation model calculates the credibility score from multiple dimensions.

[0069] Take the knowledge that "new smart meters may lose data in complex electromagnetic environments, thereby increasing the risk of electrical fires" as an example. First, evaluate the credibility of its source. If the knowledge comes from a professional and authoritative electrical research institution, the credibility score will be higher; if it comes from a personal blog and is not verified, the credibility score will be lower.

[0070] In terms of consistency with existing knowledge, the new knowledge is compared with similar knowledge already in the knowledge graph. For example, if there is already relevant knowledge in the knowledge graph that "smart meters have high data stability within the normal working range, and only when subjected to extremely strong electromagnetic interference may short-term data anomalies occur", the difference between the new knowledge and the existing knowledge needs to be further analyzed. If the "complex electromagnetic environment" in the new knowledge is clearly defined and clearly different from extremely strong electromagnetic interference, the similarity between the two needs to be re-evaluated; if the difference is large, it needs to be handled with caution.

[0071] The citation frequency is also one of the important indicators. Count the number of times this knowledge is cited in relevant professional academic literature and industry reports. If this knowledge is frequently cited in multiple authoritative documents, it indicates a relatively high credibility.

[0072] When the credibility score is in a critical situation, an artificial auxiliary review mechanism is introduced. A team composed of experts in the electrical field, combined with their professional knowledge, judges the new knowledge from aspects such as electrical principles and industry standards. At the same time, information processors provide support from aspects such as data processing rules and text processing logic, and finally determine the credibility score of this knowledge.

[0073] It should be noted here that the specific calculation steps and optimization algorithms of the knowledge credibility evaluation model are as follows:

[0074] The knowledge credibility evaluation model calculates the credibility score of each piece of knowledge by comprehensively considering multiple dimensions such as the source credibility of the knowledge, the consistency with existing knowledge, and the citation frequency. The specific calculation steps are as follows:

[0075] 1. Calculation of source credibility

[0076] Determine the category and weight of the knowledge source: According to the authority and reliability of the knowledge source, classify the knowledge source into different categories, such as standards issued by authoritative institutions, works of well-known experts in the professional field, strictly reviewed industry reports, etc., and assign different weights to each category. For example, the weight of information issued by authoritative institutions is set as W 1 (a relatively high value), and the weight of information in personal blogs is set as W 2 (a relatively low value).

[0077] In the preferred embodiment, define the category and weight of the knowledge source: Divide the knowledge source into different categories, such as standards issued by authoritative institutions (weight set as W 权威 = 0.6), works of well-known experts in the professional field (weight set as W 专家 = 0.4). The setting of the weight is based on a comprehensive evaluation of the reliability of various knowledge sources and can be determined through expert discussions, industry research, etc.

[0078] Evaluate the source credibility score of the knowledge: For each piece of knowledge, determine the category to which its knowledge source belongs, and calculate the source credibility score S1 according to the weight corresponding to this category. The calculation formula is: S1 = W 类别 where W 类别 is the weight corresponding to this knowledge source category.

[0079] 2. Calculation of consistency with existing knowledge

[0080] Convert new knowledge and existing knowledge into vector representations: Use word vector models (such as Word2Vec, BERT, etc.) to convert new knowledge and relevant existing knowledge in the knowledge graph into vector representations. Assume the vector representation of new knowledge is and the vector representation of relevant existing knowledge is .

[0081] Calculate semantic similarity: Measure the consistency between new knowledge and existing knowledge by calculating the similarity between vectors. Similarity calculation methods include cosine similarity, etc. The calculation formula is: S2 = ( * ) / ( * ), where S2 represents the semantic similarity score.

[0082] 3. Calculation of citation frequency

[0083] Count the number of times knowledge is cited: Collect information sources such as relevant academic literature and industry reports, and count the number of times N that the knowledge is cited in these sources.

[0084] Normalize the citation times: To make the citation frequency comparable under different scales of data, normalize the citation times. For example, set a maximum citation times Nmax (which can be determined according to the actual situation), then the calculation formula for the citation frequency score S3 is: S3 = N / Nmax.

[0085] 4. Comprehensive calculation of credibility score

[0086] Determine the weights of each dimension: Assign weights W 1 , W 2 and W 3 to the three dimensions of source credibility, consistency with existing knowledge, and citation frequency, respectively, and satisfy W 1 + W 2 + W 3 = 1. The determination of weights can be obtained through expert experience, experimental verification, etc.

[0087] Calculate the credibility score: Combine the scores of the above three dimensions according to the weights to obtain the credibility score S of each piece of knowledge. The calculation formula is: S = W 1 × S1 + W 2 × S2 + W 3 × S3.

[0088] It can be understood that the implementation steps of the specific optimization algorithm for fusing and optimizing knowledge with conflicts and redundancies and generating an optimized knowledge graph according to the credibility score include:

[0089] 1. Rule-based optimization algorithm

[0090] Rule Definition: A series of electrical fire inference rules are predefined, which formally describe various conditions and logical relationships for the occurrence of electrical fires. For example, "If the insulation layer of a wire or cable is damaged and the current it carries exceeds its rated current, an electrical fire may be triggered."

[0091] Rule Matching and Application: Traverse the nodes and edges in the knowledge graph to find situations that meet the rule conditions, and draw conclusions based on the rules. For example, when it is found that the current status node of a large motor in the knowledge graph shows that it has been running at high load continuously for more than 24 hours, and the insulation layer status node shows that the insulation layer is damaged, according to the above rules, it can be inferred that the risk of electrical fire at the location of the motor increases, and corresponding markings are made in the knowledge graph.

[0092] 2. Optimization Algorithm Based on Bayesian Inference

[0093] Data Collection and Model Training: Based on historical data and expert experience in the field of electrical fires, collect data on various factors (such as types of electrical equipment failures, environmental humidity, temperature, etc.) under different electrical fire occurrence situations, and label these data. Use these labeled data to train a Bayesian network model to learn the probabilities of various electrical fire occurrences and the probability distributions of related factors.

[0094] Probability Calculation and Update: During inference, according to the known evidence (the known information in the knowledge graph), calculate the probability of the occurrence of an unknown event through Bayes' formula. For example, given that the current environmental humidity in a certain area is relatively high and the temperature is relatively high, combined with the aging situation of electrical equipment, the probability of an electrical fire occurring in this area can be calculated through the Bayesian inference algorithm, and the result is fed back to the knowledge graph to update the probability information of relevant nodes.

[0095] 3. Optimization Algorithm Based on Graph

[0096] Graph Structure Analysis and Path Search: Utilize the graph structure characteristics of the knowledge graph to perform inference through path search and propagation algorithms between nodes. For example, by calculating the shortest path or associated path between nodes, potential causal relationships of electrical fires can be discovered.

[0097] Information Propagation and Update: Define a propagation weight model. According to the connection relationship between nodes and the attributes of edges (such as the weight of the edge represents the tightness of the relationship), propagate inference information from one node to adjacent nodes. For example, if it is found that there is a close association between multiple cases of unauthorized wiring of wires in an old community and electrical fires, through the information propagation algorithm, corresponding relationships can be established between the electrical equipment nodes and fault type nodes related to this old community in the knowledge graph, and the knowledge graph is updated.

[0098] Specifically, the optimization algorithm based on Bayesian inference is as follows:

[0099] The basic formula of Bayes' theorem is:

[0100] P(A|B) = P(B|A)P(A) / P(B);

[0101] Among them: P(A|B) represents the probability that event A occurs under the condition that event B occurs, that is, the posterior probability;

[0102] P(B|A) represents the probability that event B occurs under the condition that event A occurs;

[0103] P(A) represents the prior probability that event A occurs;

[0104] P(B) represents the probability that event B occurs.

[0105] It can be understood that in the electrical fire risk assessment scenario of the present invention, we assume that event A represents "an electrical fire occurs", and event Bi represents various factors affecting the occurrence of an electrical fire, such as B1 represents "electrical equipment failure", B2 represents "high environmental humidity", B3 represents "high temperature", etc. The specific calculation steps are as follows:

[0106] 1. Calculate the prior probability P(A)

[0107] The prior probability P(A) can be obtained by statistical analysis of historical electrical fire data. Assuming that among the N pieces of electrical fire-related data collected, the number of cases of electrical fires is nA, the calculation formula for the prior probability P(A) is:

[0108] P(A) = nA / N.

[0109] 2. Calculate the conditional probabilities P(Bi|A) and P(Bi)

[0110] Calculate P(Bi|A): P(Bi|A) represents the probability that a certain factor Bi appears under the condition of an electrical fire. By analyzing the cases that meet event A in the historical electrical fire data and counting the number of cases n that also meet factor Bi ABi , then the calculation formula for P(Bi|A) is:

[0111] P(Bi|A) = n ABi / n A

[0112] For example, among 100 electrical fire cases, 60 cases have electrical equipment failures, then P(Bi|A) = 60 / 100 = 0.6.

[0113] Calculate P(Bi): P(Bi) represents the probability of the occurrence of factor Bi. By analyzing a large amount of electrical system operation data (including data with and without fires), the number of cases n that meet factor Bi is counted. Bi , and the total number of cases is N. total , then the calculation formula for P(Bi) is:

[0114] P(Bi) = n Bi / N total。

[0115] For example, among 1000 electrical system operation data, 200 data show electrical equipment failures, then P(Bi) = 200 / 1000 = 0.2.

[0116] 3. Calculate the posterior probability P(A|B1, B2, ⋯, Bn)

[0117] In practical applications, we usually need to calculate the probability of an electrical fire based on the simultaneous occurrence of multiple factors, that is, P(A|B1, B2, ⋯, Bn). According to the total probability formula and the chain rule of Bayes' theorem, it can be expanded as:

[0118] P(A|B1, B2, ⋯, Bn) = P(B1, B2, ⋯, Bn|A)P(A) / P(B1, B2, ⋯, Bn).

[0119] Assume that each factor is independent of each other (in practical applications, the correlation between factors can be considered according to the specific situation), then P(B1, B2, ⋯, Bn|A) can be approximately expressed as P(B1|A)P(B2|A)⋯P(Bn|A), and P(B1, B2, ⋯, Bn) can be approximately expressed as P(B1)P(B2)⋯P(Bn).

[0120] Therefore, the calculation formula for P(A|B1, B2, ⋯, Bn) can be simplified to:

[0121] P(A|B1, B2, ⋯, Bn) = P(B1|A)P(B2|A)⋯P(Bn|A)P(A) / P(B1)P(B2)⋯P(Bn).

[0122] In a preferred embodiment of the present invention, it is assumed that P(A)=0.1 is known (i.e., the probability of an electrical fire occurring historically is 10%), P(B1∣A)=0.6 (the probability of an electrical equipment failure when an electrical fire occurs is 60%), P(B1)=0.2 (the overall probability of an electrical equipment failure is 20%), P(B2∣A)=0.7 (the probability of high environmental humidity when an electrical fire occurs is 70%), P(B2)=0.3 (the overall probability of high environmental humidity is 30%), P(B3∣A)=0.5 (the probability of high temperature when an electrical fire occurs is 50%), and P(B3)=0.4 (the overall probability of high temperature is 40%).

[0123] Then, when the three factors of electrical equipment failure, high environmental humidity, and high temperature occur simultaneously, the probability P(A∣B1,B2,B3) of an electrical fire occurring is as follows:

[0124] P(A∣B1,B2,B3)=P(B1∣A)P(B2∣A)P(B3∣A)P(A) / P(B1)P(B2)P(B3)=0.6×0.7×0.5×0.1 / 0.2×0.3×0.4 = 0.9167.

[0125] Through the above formula and calculation process, we can dynamically evaluate the risk probability of an electrical fire under different conditions based on the known probability information of electrical fire-related factors, thereby providing decision support for the prevention and control of electrical fires. In this technical solution, this probability information can be used to update the attributes of relevant nodes and edges in the knowledge graph to reflect the real-time changes in electrical fire risks.

[0126] Through the above detailed calculation steps, fusion optimization methods, and implementation steps of specific optimization algorithms, the technical solution of this application can effectively evaluate the knowledge credibility, fuse and optimize the conflicting and redundant knowledge, and generate an optimized knowledge graph, thereby providing more accurate and comprehensive support for the analysis and statistics of electrical fire information.

[0127] S400, Construction of an intelligent reasoning engine: Construct an intelligent reasoning engine, which includes an inference algorithm and a rule base; use the intelligent reasoning engine to perform in-depth reasoning on the optimized knowledge graph, mine the potential causal relationships and influencing factors of electrical fires, and generate analysis and statistical results.

[0128] Specifically, in the step of constructing the intelligent reasoning engine, the inference algorithm of the intelligent reasoning engine includes a rule-based inference algorithm, a Bayesian inference algorithm, and a graph-based inference algorithm. The rules in the rule base are stored classified according to the causes of electrical fires, including electrical equipment failure rules, environmental factor rules, and human operation rules. The reasoning process is carried out in a multi-round iterative and depth-searching manner.

[0129] Specifically, in the step of constructing the intelligent inference engine, the inference engine uses the inference results to automatically annotate and supplement the knowledge graph. For the newly discovered potential causal relationships and influencing factors of electrical fires, corresponding entities, attributes, and relationships are automatically generated and added to the knowledge graph.

[0130] Specifically, in the step of constructing the intelligent inference engine, its specific implementation method and the way of organizing the rule base are as follows:

[0131] Inference algorithms:

[0132] Rule-based inference algorithm: A large number of electrical fire inference rules are predefined in the rule base. These rules formally describe various conditions and logical relationships for the occurrence of electrical fires. For example, "If the insulation layer of the wire and cable is damaged and the current carried exceeds its rated current, then an electrical fire may be triggered." During the inference process, the engine traverses the nodes and edges in the knowledge graph, searches for situations that meet the rule conditions, and draws conclusions based on the rules.

[0133] Bayesian inference algorithm: An inference algorithm used to process uncertainty and probabilistic information. Based on historical data and expert experience in the field of electrical fires, it models the probabilities of various electrical fires occurring and the probability distributions of related factors. For example, it statistically analyzes the probabilities of electrical fires occurring in different regions and different types of electrical equipment, as well as the influence of various environmental factors (such as humidity, temperature, etc.) on the probability of electrical fires occurring. During inference, according to the known evidence (the known information in the knowledge graph), it calculates the probability of an unknown event occurring through the Bayesian formula.

[0134] Graph-based inference algorithm: Utilizes the graph structure characteristics of the knowledge graph and performs inference through path search and propagation algorithms between nodes. For example, by calculating the shortest path or associated path between nodes, potential causal relationships of electrical fires are discovered. A propagation weight model can be defined, and according to the connection relationships between nodes and the attributes of edges (such as the weight of the edge representing the tightness of the relationship), the inference information is propagated from one node to adjacent nodes.

[0135] Rule base organization: The rules in the rule base are stored classified according to the causes of electrical fires, including electrical equipment failure rules, environmental factor rules, and human operation rules, etc. For each type of rule, it is further subdivided. For example, under the electrical equipment failure rules, it can be divided into short-circuit rules, overload rules, insulation failure rules, etc. This classified storage method facilitates quickly locating and applying relevant rules during inference.

[0136] Specifically, in the step of constructing the intelligent inference engine, the process of the inference engine using the inference results to automatically annotate and supplement the knowledge graph is as follows:

[0137] Automatic annotation: Based on the potential causal relationships and influencing factors of electrical fires inferred by the inference engine, automatic annotation is performed in the knowledge graph. For example, if it is inferred that a certain electrical device is prone to failure and cause a fire under specific environmental conditions, corresponding labels are added to the corresponding entities (electrical devices) and relationships (the relationship between failure and environmental conditions) in the knowledge graph to mark their fire risk characteristics in this scenario.

[0138] Relationship supplementation: For the newly discovered potential causal relationships and influencing factors of electrical fires, corresponding entities, attributes, and relationships are automatically generated and added to the knowledge graph. For example, if it is inferred that a new electrical connection method may increase the risk of electrical fires, a new "electrical connection method" entity is created in the knowledge graph, an attribute of "fire risk factor" is added to it, and a new relationship is established between this entity and relevant electrical device and fault type entities. To ensure the consistency and integrity of the knowledge graph, consistency checks and validations are performed when adding new content, such as checking whether the new entities and relationships conform to the logical constraints and semantic rules of the knowledge graph. At the same time, the newly added content is regularly sorted and archived for subsequent maintenance and use.

[0139] Furthermore, in the preferred embodiment, the intelligent inference engine includes a variety of inference algorithms and a rich rule base for deeply mining electrical fire information.

[0140] In terms of the rule-based inference algorithm, numerous electrical fire-related rules are defined in the rule base. For example, "If an electrical device operates at a high load for a long time (the current exceeds 80% of the rated current and lasts for more than 24 hours), the risk of electrical fires increases significantly." When the inference engine analyzes that the current status node of a large motor node in the knowledge graph shows that it has been operating at a high load continuously for more than 24 hours, it can be inferred based on this rule that the risk of electrical fires at the location of this motor increases.

[0141] The Bayesian inference algorithm plays an important role in dealing with uncertain information. Taking the relationship between electrical fire risk and factors such as humidity and temperature as an example, by collecting a large amount of historical humidity, temperature data and corresponding electrical system status when electrical fires occurred, a Bayesian network model is established. The model learns that under the conditions of high humidity (such as relative humidity greater than 80%) and high temperature (such as ambient temperature higher than 40°C), combined with the aging situation of electrical equipment (such as the equipment has been used for more than 15 years), the probability of an electrical fire occurring will increase significantly. When receiving real-time monitored humidity, temperature data and information such as the service life of local electrical equipment in a certain area, the Bayesian inference algorithm can calculate the possibility of an electrical fire occurring in this area.

[0142] Graph-based reasoning algorithms utilize the graph structure characteristics of the knowledge graph. For example, by searching for paths between nodes and calculating associated paths, it is found that "there is a close association between the complex electrical wire connections (random wiring of electrical wires in multiple places) in an old community and electrical fire cases", thus it can be inferred that the risk of electrical fires in this community increases due to random wiring of electrical wires.

[0143] During the reasoning process, if new potential causal relationships and influencing factors of electrical fires are obtained through these reasoning algorithms, the engine will automatically annotate and supplement the knowledge graph using the reasoning results. For example, between the node representing "random wiring of electrical wires" and the nodes of electrical equipment connected to it in the knowledge graph, a relationship label of "increasing fire risk" is added. At the same time, "random wiring of electrical wires" is added as a new entity of electrical connection method to the knowledge graph, and relevant attributes and relationships are assigned to it.

[0144] S500. Establishment of dynamic update mechanism: Set up a real-time monitoring module to monitor the changes of knowledge sources and the emergence of new electrical fire cases in real time; when knowledge source updates or new cases occur, use knowledge update algorithms to integrate new knowledge and cases into the knowledge graph to achieve dynamic update of the knowledge graph.

[0145] Specifically, in the step of establishing the dynamic update mechanism, the real-time monitoring module determines whether the knowledge source has changed by monitoring the changes in the web page structure of the knowledge source and identifying data version changes. For the emergence of new electrical fire cases, information collection is triggered by monitoring new records in the electrical fire database. The new knowledge collected enters the knowledge update process after preliminary screening and format conversion.

[0146] Specifically, in the step of establishing the dynamic update mechanism, the specific implementation methods of real-time monitoring and information collection are as follows:

[0147] Monitoring of changes in the web page structure of the knowledge source: The real-time monitoring module regularly (for example, every certain time interval, such as 1 hour) crawls the web pages of knowledge sources (such as electrical industry websites, academic literature databases, etc.) to obtain the source code of the web pages. The newly crawled web page source code is compared line by line with the previously stored web page source code, and the change in the web page structure is judged by analyzing changes in aspects such as HTML tag structure and text position. For the parts with structural changes, new data and information are extracted.

[0148] Identification of data version changes: For knowledge sources with clear data version identifiers, such as databases and document libraries, the data version number is parsed to identify whether the data has changed. The real-time monitoring module connects to the version management system of the knowledge source to obtain the latest version number of each data object (such as documents, records, etc.), and compares it with the version number stored locally. If the version number has changed, the latest data content is obtained.

[0149] Generation Monitoring of New Electrical Fire Cases: By establishing an interface connection with an electrical fire database (such as the fire case database of the fire department, the case library of relevant research institutions, etc.), the latest added records can be obtained in real time. To improve the monitoring efficiency, asynchronous pulling or event-driven methods can be used for data acquisition. When a new electrical fire case is detected, the newly acquired knowledge is initially screened (such as filtering out content that is obviously not related to the theme of electrical fires) and its format is converted (converted into a standard format compatible with the knowledge graph), and then enters the knowledge update process.

[0150] Specifically, in the knowledge update step, the knowledge update algorithm includes an update method based on incremental learning and an update method based on fusion reconstruction. When the sum of the quantity of new knowledge and the quantity of electrical fire cases is less than the first preset threshold, the incremental learning method is used to quickly update the knowledge graph; when the sum of the quantity of new knowledge and the quantity of electrical fire cases is greater than or equal to the first preset threshold, the fusion reconstruction method is used to comprehensively update the knowledge graph.

[0151] Specifically, in the knowledge update step, the knowledge update algorithm is specifically implemented as follows:

[0152] Incremental Learning Method: When the quantity of new knowledge and cases is small, the incremental learning method is used to quickly update the knowledge graph. First, feature extraction and representation learning are performed on the new knowledge and cases and the existing data in the knowledge graph, and they are transformed into the form of feature vectors. Then, using predefined learning rules and strategies, these new feature vectors are incorporated into the existing knowledge graph model. For example, for a knowledge graph model based on deep learning, new hidden layer nodes can be added or the weights of existing nodes can be adjusted on the basis of the original model to adapt to the changes in new knowledge. This method can quickly adapt to the changes in small-scale data without having a great impact on the structure of the entire knowledge graph.

[0153] Fusion Reconstruction Method: When the quantity of new knowledge and cases is large, that is, it has a great impact on the structure of the existing knowledge graph, the fusion reconstruction method is used to comprehensively update the knowledge graph. First, the new knowledge and cases are deeply integrated with the data in the original knowledge graph to construct a temporary knowledge graph. In this process, by comparing and integrating the same or similar knowledge from different sources, conflicts and redundancies are eliminated. Then, the temporary knowledge graph is optimized and verified, including but not limited to checking the integrity of nodes and relationships, logical consistency, etc. Next, according to the optimized temporary knowledge graph, the model structure of the original knowledge graph is redesigned and adjusted, such as redefining entity types, relationship types, and inference rules, etc. Finally, the optimized temporary knowledge graph replaces the original knowledge graph to complete the update process.

[0154] Furthermore, in a preferred embodiment, in order for the knowledge graph to reflect new information and knowledge in real time, a dynamic update mechanism is crucial.

[0155] The real-time monitoring module ensures the monitoring of knowledge sources through multiple methods. For the monitoring of the web page structure changes of knowledge sources, taking an authoritative website in the electrical industry as an example, the update module grabs the web page source code of this website every 1 hour. The newly grabbed source code is compared line by line with the previously stored version. If web page structure change information such as the addition or deletion of HTML tags and the change of text position is found, the corresponding new data and information related to electrical fires are extracted.

[0156] For the identification of data version changes, in the connection with the electrical fire database, the database system itself marks the versions of operations such as data modification and addition. The real-time monitoring module obtains the latest version number of each data object in real time through the interface connection with the version management system and compares it with the version number stored locally. Once the version number changes, the updated content is obtained.

[0157] In terms of the monitoring of new electrical fire cases, a real-time interface connection is established with the electrical fire database. When a new case record is monitored and added to the database, the collected case data is first preliminarily screened to remove obviously irrelevant content, such as records containing non-electrical fire keywords (such as "geological disasters", etc.). Then the data is subjected to format conversion to adjust it to a standard format compatible with the knowledge graph. For example, the time format is uniformly converted to "YYYY-MM-DD HH:MM:SS", and the text encoding is unified to UTF-8, etc., and then it enters the knowledge update process.

[0158] The knowledge update process adopts different methods according to the scale of the newly added knowledge and cases. When the amount of newly added knowledge and cases is small, an incremental learning method is adopted. Taking the addition of 10 - 20 pieces of information on the association between electrical equipment failures and electrical fires each time as an example, the features of these new information are extracted as vectors in the knowledge graph, and through predefined rules and strategies, these vectors are incorporated into the existing model. For example, for a knowledge graph model based on deep learning, by adding a small number of hidden layer nodes or fine-tuning the weights of existing nodes, the model can adapt to new knowledge and quickly update the model parameters.

[0159] When the amount of newly added knowledge and cases is large and has a significant impact on the knowledge graph structure, such as adding more than 100 pieces at a time and involving brand-new electrical fire-related knowledge, the fusion and reconstruction method is adopted. First, construct a temporary knowledge graph, fuse the newly added knowledge and cases with the original knowledge, and eliminate conflicts and redundancies. For example, if the newly added knowledge mentions the risk mechanism of a certain new electrical connection method leading to electrical fires, and there is no relevant content in the original knowledge graph, add the entity of this connection method to the temporary knowledge graph first, and establish relationships with related electrical equipment and fault type entities. Then, optimize and verify the temporary knowledge graph, check the integrity and logical consistency of nodes and relationships, and redesign the model structure of the knowledge graph, such as adding entity types, relationship types, and inference rules related to the new connection method. Finally, replace the original knowledge graph with the optimized temporary knowledge graph to complete the comprehensive update.

[0160] Specifically, in the step of generating the analysis and statistical results, the generated analysis and statistical results are classified and displayed according to the electrical fire risk level, and the potential causal relationships and influencing factors of electrical fires are presented graphically through a visualization tool. The graphical display includes node diagrams, flowcharts, and causal relationship diagrams.

[0161] Specifically, in the step of generating the analysis and statistical results, the specific implementation methods of classified display and visualization display are as follows:

[0162] Classified display: The generated analysis and statistical results are classified and displayed according to the electrical fire risk level. The division of the risk level is based on multiple preset indicators (such as the possibility of fire, the degree of potential harm, etc.). For example, the risk level is divided into three levels: high, medium, and low. For the results of each risk level, independent statistical reports and descriptive information are provided, including the quantity, distribution, and main characteristics of this level. For example, in the high-risk level part, display the number of high-risk electrical fire cases involved, the main types of electrical equipment involved, and the common fault types, etc.

[0163] Visualization display: The potential causal relationships and influencing factors of electrical fires are presented graphically through a visualization tool. The graphical display includes various forms such as node diagrams, flowcharts, and causal relationship diagrams:

[0164] Node diagram: Represent the entities related to electrical fires (such as electrical equipment, electrical parameters, environmental factors, etc.) as nodes, and the relationships between entities as edges. The size and color of the nodes can be distinguished according to the importance or risk level of the entities, and the thickness of the edges can represent the tightness of the relationships. For example, through the node diagram, the relationship between a certain electrical fire event and various factors involved can be intuitively displayed.

[0165] Flowchart: Used to show the occurrence process of electrical fires and the action mechanisms of influencing factors. Taking time as the sequence, each stage of the fire occurrence (such as hidden danger generation, trigger condition satisfaction, fire development and spread, etc.) is used as a node of the process, and the intermediate connections represent the development sequence and causal relationship of events.

[0166] Causality diagram: Focuses on showing the causal connections among various influencing factors of electrical fires. Through the connection of nodes and edges, it clearly presents how different causes (such as electrical equipment failures, environmental conditions, etc.) jointly act to cause the occurrence of electrical fires.

[0167] Customization and interaction functions: To facilitate users to display and operate according to specific needs, the visualization tool also provides customization functions. Users can select the content to be displayed and adjust the display methods (such as the size and color of nodes, the style of edges, etc.). At the same time, interaction functions are provided, such as clicking on a node to view detailed entity information, hovering over an edge to display a detailed description of the relationship, etc.

[0168] Furthermore, in the preferred embodiment, the generated analysis and statistical results are presented in a hierarchical display and multiple visualization methods, which is convenient for users to understand and operate.

[0169] In terms of hierarchical display, the results are classified and displayed according to the electrical fire risk levels (high, medium, low). For example, in the high-risk level part, the specific information such as the number of high-risk electrical fire cases involved, the typical types of electrical equipment involved (such as large industrial motors, power supply modules of data center servers, etc.), and the common types of faults (such as overcurrent short circuits, insulation damage, etc.) is displayed in detail. Through statistical data, inform users of the proportion of high-risk cases in the total cases, the distribution of high-risk electrical fires in different industries, etc.

[0170] The visualization display adopts various forms:

[0171] Node diagram: Regarding the entities related to electrical fires (electrical equipment, electrical parameters, environmental factors, etc.) as nodes and the relationships between entities as edges. Taking the electrical system of an industrial factory as an example, various equipment in the factory such as transformers, distribution cabinets, and motors are used as nodes, and the connection lines and power supply relationships between the equipment are used as edges. The importance of electrical equipment is represented by the color of the nodes (such as red for key equipment and green for secondary equipment), and the tightness of the relationship between equipment is represented by the thickness of the edges (such as thick edges for closely connected equipment combinations). Users can intuitively see the relationships among various parts of the factory's electrical system and the key nodes.

[0172] Flowchart: With time as the axis, it shows the process of electrical fire. For example, from the aging of electrical equipment (node ​​one), to the degradation of insulation performance (node ​​two), to the occurrence of local overheating (node ​​three), and finally to the occurrence of electrical fire (node ​​four), the connection between each node represents the sequence and cause-and-effect relationship of the development of events, so that users can clearly understand the formation process of electrical fire.

[0173] Cause and effect diagram: It highlights the cause and effect relationship between various factors affecting electrical fires. For example, it shows the relationship between the humidity of the operating environment of electrical equipment (factor A), the degree of insulation aging of the equipment (factor B), and electrical fires (result C). Through the connection of nodes and edges, it intuitively shows how increased humidity accelerates insulation aging, thereby increasing the risk of electrical fires.

[0174] In addition, in order to meet the personalized needs of users, the visualization tool provides customization functions. Users can choose the content to be displayed and adjust the display method according to specific needs. For example, users can choose to only view electrical fire information related to a specific type of electrical equipment, or adjust the size, color, and edge style of the node. At the same time, interactive functions are set up, and users can obtain detailed entity information by clicking on the node, such as the specific parameters of the electrical equipment, fault records, etc.; hovering on the edge can view a detailed description of the relationship, such as the specific mechanism of a certain electrical connection method causing a fire.

[0175] It can be understood that the present invention effectively improves the accuracy of knowledge fusion through the knowledge credibility evaluation model and optimization algorithm, making the constructed knowledge graph more accurate and complete, and better supporting the analysis and statistics of electrical fire information. The introduction of the intelligent reasoning engine can deeply explore the potential causal relationships and influencing factors of electrical fires, generate detailed and accurate analysis and statistical results, and provide users with more comprehensive information support. The establishment of a dynamic update mechanism ensures that the knowledge graph can reflect the emerging electrical fire information and knowledge in real time, always maintain the timeliness and effectiveness of the information, and greatly improve the practicality and reliability of electrical fire information analysis and statistics.

[0176] See also Figure 2 The present invention provides another embodiment, which provides an electrical fire information analysis and statistics system based on a knowledge graph, and the electrical fire information analysis and statistics system based on a knowledge graph includes:

[0177] Data processing module 100: used to perform data collection and preprocessing steps, clean, remove duplicates and normalize multi-source heterogeneous data of electrical fires, and generate a standardized data set;

[0178] Knowledge extraction module 200: Using natural language processing techniques and machine learning algorithms to extract entities, attributes, and relationships related to electrical fires from the standardized data set, and constructing a preliminary knowledge graph;

[0179] Knowledge fusion and optimization module 300: Evaluating the credibility of the knowledge in the preliminary knowledge graph through a knowledge credibility evaluation model, fusing and optimizing the conflicting and redundant knowledge, and generating an optimized knowledge graph;

[0180] Inference and analysis module 400: Constructing an intelligent inference engine to deeply infer the optimized knowledge graph, mining the potential causal relationships and influencing factors of electrical fires, and generating analysis and statistical results;

[0181] Dynamic update module 500: Setting up a real-time monitoring module to monitor the changes in knowledge sources and the emergence of new electrical fire cases in real time, and realizing the dynamic update of the knowledge graph through a knowledge update algorithm.

[0182] It should be noted here that through the knowledge credibility evaluation model and optimization algorithm, the present invention effectively improves the accuracy of knowledge fusion, makes the constructed knowledge graph more accurate and complete, and can better support the analysis and statistical work of electrical fire information. The introduction of the intelligent inference engine can deeply mine the potential causal relationships and influencing factors of electrical fires, generate detailed and accurate analysis and statistical results, and provide more comprehensive information support for users. The establishment of the dynamic update mechanism ensures that the knowledge graph can reflect the newly emerging electrical fire information and knowledge in real time, always maintain the timeliness and effectiveness of the information, and greatly improve the practicality and reliability of the analysis and statistical work of electrical fire information.

[0183] Furthermore, in the preferred embodiment, the present application also provides an electronic device, and the electronic device includes:

[0184] A memory; and a processor, wherein computer-readable instructions are stored on the memory, and when the computer-readable instructions are executed by the processor, the method for analyzing and statistically processing electrical fire information based on a knowledge graph is implemented. The computer device can be broadly a server, a terminal, or any other electronic device having necessary computing and / or processing capabilities. In one embodiment, the computer device may include a processor, a memory, a network interface, a communication interface, etc. connected through a system bus. The processor of the computer device can be used to provide necessary computing, processing, and / or control capabilities. The memory of the computer device may include a non-volatile storage medium and an internal memory. An operating system, a computer program, etc. may be stored in or on the non-volatile storage medium. The internal memory can provide an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface and the communication interface of the computer device can be used to connect and communicate with external devices through a network. The computer program, when executed by the processor, executes the steps of the method of the present invention.

[0185] The present invention can be implemented as a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method of the embodiments of the present invention are caused to be executed. In one embodiment, the computer program is distributed on a plurality of network-coupled computer devices or processors, so that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, can be executed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations can be executed by one or more computer devices or processors, and one or more other method steps / operations can be executed by one or more other computer devices or processors. One or more computer devices or processors can execute a single method step / operation, or execute two or more method steps / operations.

[0186] Those of ordinary skill in the art can understand that the method steps of the present invention can be completed by a computer program instructing relevant hardware such as a computer device or a processor, and the computer program can be stored in a non-transitory computer-readable storage medium, and when the computer program is executed, the steps of the present invention are caused to be executed. Depending on the situation, any reference to a memory, storage, database, or other medium herein may include non-volatile and / or volatile memories. Examples of non-volatile memories include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk, etc. Examples of volatile memories include random access memory (RAM), external cache memory, etc.

[0187] It is understandable that through the knowledge credibility evaluation model and the optimization algorithm, the present invention effectively improves the accuracy of knowledge fusion, making the constructed knowledge graph more accurate and complete, and enabling better support for the analysis and statistics of electrical fire information. The introduction of the intelligent reasoning engine can deeply explore the potential causal relationships and influencing factors of electrical fires, generate detailed and accurate analysis and statistics results, and provide more comprehensive information support for users. The establishment of the dynamic update mechanism ensures that the knowledge graph can reflect newly emerging electrical fire information and knowledge in real time, always maintaining the timeliness and effectiveness of the information, and greatly improving the practicality and reliability of the analysis and statistics of electrical fire information.

[0188] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, as long as such a combination does not exist in contradiction.

[0189] The specific embodiments of the present invention described above do not constitute a limitation on the protection scope of the present invention. Any other corresponding changes and deformations made according to the technical concept of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A statistical method for analyzing electrical fire information based on knowledge graph, characterized in that: The method comprises: S100, Data collection and preprocessing: Collect multi-source heterogeneous data related to electrical fires, including equipment manuals, accident reports, and industry standards; clean, remove duplicates, and normalize the collected data to obtain a standardized data set; S200, Knowledge Extraction: Using natural language processing technology and machine learning algorithms, extract entities, attributes and relationships related to electrical fires from standardized data sets to build a preliminary knowledge graph; S300, knowledge fusion evaluation: introduce the knowledge credibility evaluation model, conduct credibility evaluation on the knowledge in the preliminary knowledge graph, and calculate the credibility score of each piece of knowledge; integrate and optimize the conflicting and redundant knowledge according to the credibility score, and generate an optimized knowledge graph; S400, intelligent reasoning engine construction: Build an intelligent reasoning engine, which includes a reasoning algorithm and a rule base; use the intelligent reasoning engine to perform deep reasoning on the optimized knowledge graph, explore the potential causal relationships and influencing factors of electrical fires, and generate analytical statistical results; S500, dynamic update mechanism establishment: set up a real-time monitoring module to monitor the changes in knowledge sources and the generation of new electrical fire cases in real time; when the knowledge source is updated or a new case appears, use the knowledge update algorithm to integrate the new knowledge and cases into the knowledge graph to achieve dynamic update of the knowledge graph; In the step of generating analysis and statistical results, the generated analysis and statistical results are displayed in grades according to the electrical fire risk levels, and the potential cause-effect relationships and influencing factors of electrical fires are presented in a graphical manner through visualization tools. The graphical presentation includes node diagrams, flow charts and cause-effect diagrams.

2. The electrical fire information analysis and statistical method based on knowledge graph according to claim 1 is characterized in that: In the data collection and preprocessing steps, for the collected multi-source heterogeneous data of electrical fires, a regular expression-based cleaning algorithm is used to remove noise information, a hash algorithm is used to deduplicate data, and the data is normalized according to standard terminology and format in the electrical field. The standardized data set is classified and stored according to different application scenarios related to electrical fires.

3. The electrical fire information analysis and statistical method based on knowledge graph according to claim 1 is characterized in that: In the knowledge extraction step, for the extraction of entities related to electrical fires, a conditional random field algorithm is used in combination with an electrical field knowledge dictionary for identification. For attribute extraction, a support vector machine algorithm is used to perform attribute value classification extraction based on large-scale training data. Relationship extraction adopts a relationship classification model based on deep learning. The models of the natural language processing technology and machine learning algorithm are pre-trained with labeled data in the field of electrical fires before extraction to improve the accuracy of extraction.

4. The electrical fire information analysis and statistical method based on knowledge graph according to claim 1 is characterized in that: In the knowledge fusion evaluation step, the knowledge credibility evaluation model calculates the credibility score of each piece of knowledge by comprehensively considering the credibility of the knowledge source, the consistency with the existing knowledge and the frequency of citation. For knowledge whose credibility score is at a critical value, a manual assisted review mechanism is introduced to further screen and evaluate the knowledge.

5. The electrical fire information analysis and statistical method based on knowledge graph according to claim 1 is characterized in that: In the intelligent reasoning engine construction step, the reasoning algorithms of the intelligent reasoning engine include rule-based reasoning algorithms, Bayesian reasoning algorithms and graph-based reasoning algorithms. The rules in the rule base are stored according to the causes of electrical fires, including electrical equipment failure rules, environmental factor rules and human operation rules. The reasoning process is carried out in the form of multiple rounds of iterations and deep search.

6. The electrical fire information analysis and statistical method based on knowledge graph according to claim 1 is characterized in that: In the step of building an intelligent reasoning engine, the reasoning engine automatically annotates and supplements the knowledge graph using the reasoning results, and automatically generates corresponding entities, attributes, and relationships for newly discovered potential causal relationships and influencing factors of electrical fires and adds them to the knowledge graph.

7. The electrical fire information analysis and statistical method based on knowledge graph according to claim 1 is characterized in that: In the step of establishing the dynamic update mechanism, the real-time monitoring module determines whether the knowledge source has changed by monitoring the changes in the web page structure of the knowledge source and identifying the changes in the data version. For the generation of new electrical fire cases, information collection is triggered by monitoring the new records in the electrical fire database. The collected new knowledge enters the knowledge update process after preliminary screening and format conversion.

8. The electrical fire information analysis and statistical method based on knowledge graph according to claim 1 is characterized in that: In the knowledge updating step, the knowledge updating algorithm includes an updating method based on incremental learning and an updating method based on fusion reconstruction. When the sum of the amount of newly added knowledge and the amount of electrical fire cases is less than a first preset threshold, the incremental learning method is used to quickly update the knowledge graph; when the sum of the amount of newly added knowledge and the amount of electrical fire cases is greater than or equal to the first preset threshold, the fusion reconstruction method is used to comprehensively update the knowledge graph.

9. An electrical fire information analysis and statistics system based on knowledge graph, characterized in that: The method for analyzing and statistically analyzing electrical fire information based on a knowledge graph according to any one of claims 1 to 8 comprises: Data processing module: used to perform data collection and preprocessing steps, clean, remove duplicates and normalize multi-source heterogeneous data of electrical fires to generate standardized data sets; Knowledge extraction module: uses natural language processing technology and machine learning algorithms to extract entities, attributes and relationships related to electrical fires from standardized data sets and construct a preliminary knowledge graph; Knowledge fusion and optimization module: Use the knowledge credibility evaluation model to evaluate the credibility of the knowledge in the preliminary knowledge graph, integrate and optimize the conflicting and redundant knowledge, and generate an optimized knowledge graph; Reasoning and analysis module: build an intelligent reasoning engine, perform deep reasoning on the optimized knowledge graph, explore the potential causal relationships and influencing factors of electrical fires, and generate analytical statistical results; Dynamic update module: Set up a real-time monitoring module to monitor the changes in knowledge sources and the generation of new electrical fire cases in real time, and realize the dynamic update of the knowledge graph through the knowledge update algorithm.

Images