Expressway electromechanical operation and maintenance knowledge base construction method based on big data

Abstract

The invention discloses an expressway electromechanical operation and maintenance knowledge base construction method based on big data, and relates to the technical field of traffic transportation informatization, and the method comprises the steps: generating relation candidates through a co-occurrence relation, syntactic dependence and cross-modal similarity according to a fragment set, an entity set and an attribute table, performing causal direction inspection by combining a time sequence and a causal trigger word to generate a causal triple with confidence; splicing the causal triple into a causal chain set, identifying a typical evolution mode, binding traceability information for each causal chain node and edge, and generating an evidence set; and combining the semantic equivalent causal chain facts based on the evidence set to form a candidate fact set, establishing an evidence binding table, and performing comprehensive credibility calculation on the candidate fact set to generate a fusion fact set. According to the method, the electromechanical operation and maintenance knowledge base has dynamic evolution and high reliability at the same time, so that preventive maintenance and intelligent decision-making of highway electromechanical equipment are effectively supported.

Description

Technical Field

[0001] This invention relates to the field of information technology in transportation, and in particular to a method for constructing a knowledge base for the operation and maintenance of highway electromechanical systems based on big data. Background Technology

[0002] With the continuous expansion of highway traffic and the increasing variety and quantity of electromechanical equipment, highway electromechanical maintenance has gradually become a crucial link in ensuring road safety and improving service levels. In recent years, with the rapid development of technologies such as big data, artificial intelligence, and knowledge graphs, academia and industry have attempted to apply these emerging technologies to highway electromechanical maintenance scenarios. For example, by deploying sensors to collect equipment operation data, recording maintenance events through work order systems, and then extracting text information using natural language processing methods, automated management of some maintenance knowledge can be achieved. Simultaneously, with the application of multimodal technologies such as speech recognition and video analytics, the sources of electromechanical maintenance knowledge are becoming more diversified, exhibiting characteristics of large data volume, complex structure, and high real-time requirements.

[0003] However, existing technologies still have certain shortcomings. On the one hand, existing research mostly focuses on single data sources or static knowledge bases, lacking systematic integration and causal modeling of multi-source heterogeneous data for highway electromechanical maintenance. This makes it difficult to accurately reveal the dynamic relationship between equipment failures, causes, and remedial measures, resulting in lagging knowledge base updates and insufficient evolutionary capabilities. On the other hand, current knowledge updates often employ simple rule matching or statistical methods, lacking traceability mechanisms and credibility calculations. This makes it difficult to guarantee the reliability and interpretability of knowledge conclusions, leading to uncertainty in the knowledge base's ability to support decision-making and guide actual maintenance. Summary of the Invention

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

[0005] Therefore, this invention provides a method for constructing a knowledge base for highway electromechanical operation and maintenance based on big data, which solves the problems of insufficient dynamic evolution capability and insufficient knowledge credibility guarantee in existing highway electromechanical operation and maintenance knowledge construction technologies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for constructing a highway electromechanical operation and maintenance knowledge base based on big data. The method includes collecting multi-source heterogeneous data on highway electromechanical operation and maintenance and establishing a source index. The specific steps are as follows: Collect multi-source heterogeneous data on the operation and maintenance of electromechanical systems on highways; A source index is established based on the source number, collection method, collection time, and collection location of multi-source heterogeneous data.

[0007] This invention provides a method for constructing a highway electromechanical maintenance knowledge base based on big data. The method includes preprocessing multi-source heterogeneous data to obtain standardized event tables, text tables, and multimedia transcription tables. The specific steps are as follows: The multi-source heterogeneous data is uniquely encoded by the device, time-aligned, and spatially mapped to generate pre-processed multi-source heterogeneous data. The preprocessed multi-source heterogeneous data is divided into data categories and formatted uniformly to generate standardized event tables, text tables, and multimedia transcription tables.

[0008] This invention provides a method for constructing a highway electromechanical maintenance knowledge base based on big data. The method includes segmenting standardized event tables, text tables, and multimedia transcription tables into fragment sets using a sliding window to generate semantic representation vectors. The specific steps are as follows: Standardized event tables, text tables, and multimedia transcription tables are divided into time intervals using a sliding window to generate a collection of segments; Semantic encoding is performed on the fragment set to generate a preliminary semantic representation of the fragments, and the semantic representation vector is obtained through vector fusion.

[0009] This invention provides a method for constructing a highway electromechanical maintenance knowledge base based on big data, which includes extracting entity sets and attribute tables. The specific steps are as follows: Extract the entity set from the fragment set based on the semantic representation vector; The attributes of the entity set are identified, unified, and standardized to generate an attribute table.

[0010] This invention provides a method for constructing a highway electromechanical maintenance knowledge base based on big data. The method includes generating candidate relations based on a fragment set, an entity set, and an attribute table, using co-occurrence relations, syntactic dependencies, and cross-modal similarity. The specific steps are as follows: The co-occurrence of entity sets within a set of statistical fragments is analyzed to generate co-occurrence relationships. The dependency structure between entity sets is identified by syntactic dependency analysis, and syntactic dependency relationships are generated. Calculate the semantic similarity of fragments from different sources in a collection and generate cross-modal similarity relationships; Co-occurrence relations, syntactic dependency relations, and cross-modal similarity relations are merged into relation candidates.

[0011] This invention provides a method for constructing a highway electromechanical maintenance knowledge base based on big data. The method includes: combining time sequence and causal trigger words to perform causal direction verification, generating causal triples with confidence levels. The specific steps are as follows: Based on the temporal sequence information and causal trigger words in the candidate relation and fragment set, a causal direction test is performed to obtain a set of relations with causal characteristics; The causal relationships in the relation set are weighted based on source credibility, freshness, and consistency to generate causal triples with confidence.

[0012] This invention provides a method for constructing a highway electromechanical maintenance knowledge base based on big data. The method includes: concatenating causal triples into a set of causal chains, identifying typical evolutionary patterns, binding source information to each causal chain node and edge, and generating an evidence set. The specific steps are as follows: The causal triples with confidence are concatenated according to device identification and time sequence to generate a set of causal chains; The frequency of occurrence of causal chain sets is statistically analyzed and pattern mining is performed to identify typical evolutionary patterns; Based on the source index, source information is bound to nodes and edges in the causal chain set; An evidence set is generated by combining typical evolutionary patterns with the causal chain set after binding source information.

[0013] This invention provides a method for constructing a highway electromechanical maintenance knowledge base based on big data. The method includes merging semantically equivalent causal chain facts based on an evidence set to form a candidate fact set, establishing an evidence binding table, performing a comprehensive credibility calculation on the candidate fact set, and generating a fused fact set. The specific steps are as follows: Based on the evidence set, semantically equivalent causal chain facts are identified and merged to form a candidate fact set; Establish the correspondence between the candidate fact set and the evidence set, and generate an evidence binding table; Based on the evidence binding table, a weighted calculation is performed on the source credibility, freshness, and consistency to obtain a set of candidate facts with confidence, and then a fused set of facts is obtained.

[0014] As a preferred embodiment of the method for constructing a highway electromechanical maintenance knowledge base based on big data as described in this invention, the step of writing the main conclusion and the traceability index into the electromechanical maintenance knowledge base when the main conclusion in the fused fact set meets the data entry credibility threshold is as follows: Extract the main conclusions from the fused fact set and select the main conclusions that meet the credibility threshold for inclusion in the database; Based on the traceability information, the traceability index of the main conclusion that meets the credibility threshold is extracted and written into the electromechanical operation and maintenance knowledge base.

[0015] The beneficial effects of this invention are as follows: by generating causal triples, high-confidence causal relationships between multi-source heterogeneous data are extracted; by generating fused fact sets, redundancy resolution and source traceability credibility assurance are achieved; the combination of the two enables the electromechanical maintenance knowledge base to have both dynamic evolution and high reliability, thereby effectively supporting the preventive maintenance and intelligent decision-making of highway electromechanical equipment. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating the method for constructing a knowledge base for highway electromechanical maintenance based on big data.

[0018] Figure 2 This is a flowchart for preprocessing and standardizing multi-source heterogeneous data.

[0019] Figure 3 A flowchart for generating causal triples and constructing causal chains.

[0020] Figure 4 This is a flowchart for candidate fact fusion and credibility calculation. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0024] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for constructing a highway electromechanical maintenance knowledge base based on big data, including the following steps: S1: Collect multi-source heterogeneous data on the operation and maintenance of electromechanical systems on highways, establish a source index, and preprocess the multi-source heterogeneous data to obtain standardized event tables, text tables, and multimedia transcription tables. S1.1: Collect multi-source heterogeneous data on the operation and maintenance of electromechanical systems on highways; Furthermore, data from different sources is obtained through monitoring equipment, sensing devices, and management platform logs deployed along the highway. The data includes equipment operating status information, fault alarm records, inspection work order records, textual maintenance instructions, and voice monitoring transcripts, forming multi-source heterogeneous data covering structured, semi-structured, and unstructured data.

[0025] S1.2: Establish a source index based on the source number, collection method, collection time, and collection location of the multi-source heterogeneous data; Furthermore, each piece of multi-source heterogeneous data is assigned a source number to identify the unique source of the data collection, and the corresponding data collection method is recorded, including sensor collection, manual entry or automatic log generation. At the same time, the collection time and collection location information are specified. The above four types of information are combined into index items to generate a source index that can identify the source attributes of multi-source heterogeneous data.

[0026] S1.3: Perform device-specific encoding, time alignment, and spatial location mapping on multi-source heterogeneous data to generate pre-processed multi-source heterogeneous data; Furthermore, data involving different devices will be assigned unique device codes to distinguish the operating status and event information of different devices; all multi-source heterogeneous data will be aligned under a unified time base to ensure the consistency of time order between data from different sources; the collection locations of multi-source heterogeneous data will be mapped to the spatial coordinates of highways to establish a mapping relationship between multi-source heterogeneous data and geographical location, and after processing, pre-processed multi-source heterogeneous data that can guarantee device uniqueness, time consistency and spatial locationability will be generated.

[0027] S1.4: Divide the preprocessed multi-source heterogeneous data according to data categories and unify the format to generate standardized event tables, text tables and multimedia transcription tables; Furthermore, event-type data in the preprocessed multi-source heterogeneous data is classified into an event table, text description data into a text table, and text data obtained from speech transcription or video transcription into a multimedia transcription table. After the classification is completed, the format of each type of data is standardized so that data from different sources can be stored and retrieved compatiblely under the same data structure, resulting in standardized event tables, text tables, and multimedia transcription tables.

[0028] It should be noted that data category refers to the classification of preprocessed multi-source heterogeneous data based on content characteristics and presentation format. Specifically, it includes three categories: one is event data describing equipment operating status, alarm conditions, and maintenance events, which is formatted and then categorized into the event table; another is textual data such as manually entered inspection work orders, maintenance instructions, and management records, which is standardized and categorized into the text table; and the third is transcribed data from voice recordings, video surveillance subtitles, or voice interaction content, which is cleaned and formatted and then categorized into the multimedia transcription table. This method completes the data category classification of the preprocessed multi-source heterogeneous data.

[0029] S2: By using a sliding window, the standardized event table, text table, and multimedia transcription table are segmented into a set of fragments, a semantic representation vector is generated, and the entity set and attribute table are extracted. S2.1: The standardized event table, text table, and multimedia transcription table are divided into time intervals using a sliding window to generate a collection of segments; Furthermore, the standardized event table, text table, and multimedia transcription table are divided at equal intervals based on continuous time intervals, so that data from different sources can be merged together within the same time range to form temporally continuous and mutually aligned data segments. The data segments formed after being divided according to time intervals are defined as a segment set.

[0030] S2.2: Semantically encode the fragment set to generate a preliminary semantic representation of the fragments, and obtain the semantic representation vector through vector fusion; Furthermore, word segmentation, part-of-speech tagging, and semantic feature extraction are performed on the text content and transcribed content in the fragment set. The structured field information in the event table is mapped to the corresponding semantic tags and uniformly converted into semantic vectors to form a preliminary semantic representation of the fragments. Multiple preliminary semantic representations of fragments are merged and weighted through vector fusion methods to generate a semantic representation vector that can comprehensively reflect the semantic features of the fragment set.

[0031] S2.3: Extract the entity set from the fragment set based on the semantic representation vector; Furthermore, by utilizing keywords, field identifiers, and contextual information in the semantic representation vector, the equipment name, fault type, maintenance operation, time information, and location information related to electromechanical operation and maintenance are identified and summarized as entity elements in the entity set. This ensures that the core information in the fragment set is extracted in the form of the entity set, thereby completing the generation of the entity set.

[0032] S2.4: Identify, unify, and standardize the attributes of the entity set to generate an attribute table; Furthermore, the context of each entity in the entity set is analyzed to determine the corresponding attribute categories. For example, the equipment entity corresponds to the equipment number, model, and operating status; the fault entity corresponds to the fault level, impact range, and duration. The expression methods of similar attributes in different fragment sets are unified, such as unifying the time format and the vocabulary for status descriptions. Incomplete or ambiguous attribute content (e.g., attributes such as "around 9 pm last night," "TS-3# / ts3," "not very stable," "relatively serious," "near the service area," and "temperature: 30" appearing in the fragment: after normalization, they are recorded as "2025-07-15 21:00:00," equipment number "TS-003," operating status "jittering (jitter rate 0.35)," fault level "P2," location "X service area (latitude and longitude + 300m buffer)," and value "30 ℃" (with statistical caliber)). This ensures that attribute content can be expressed under the same semantic dimension, generating an attribute table that can be used for subsequent knowledge construction.

[0033] S3: Based on the fragment set, entity set, and attribute table, generate relation candidates through co-occurrence relations, syntactic dependencies, and cross-modal similarity, and combine temporal order and causal trigger words to perform causal direction verification, generating causal triples with confidence. S3.1: Analyze the co-occurrence of entity sets in the statistical fragment set and generate co-occurrence relationships; Furthermore, the number of times each pair of entities co-occurs in the same fragment set is recorded. The ratio of the number of co-occurrences to the total number of fragments in the fragment set is used as the co-occurrence ratio. The co-occurrence ratio is used as a measure of the strength of the association between entities. When the co-occurrence ratio reaches the association strength threshold, it is considered a strong association. When it is below the association strength threshold, it is considered a weak association, thus generating a co-occurrence relationship.

[0034] It should be noted that the association strength threshold is the average of the co-occurrence ratios of all entity pairs within the fragment set.

[0035] S3.2: Identify the dependency structure between entity sets using syntactic dependency analysis methods and generate syntactic dependency relationships; Furthermore, dependency parsing is used to perform dependency analysis on the text table and multimedia transcription table to identify the dependency relationships between words. These dependency relationships are then mapped to entity sets in the fragment set to form subject-predicate, verb-object, or modification relationships between entities, thus generating syntactic dependencies.

[0036] S3.3: Calculate the semantic similarity of fragments from different sources in the fragment set and generate cross-modal similarity relationships; Furthermore, semantic representation vectors are extracted from the event table, text table, and multimedia transcription table. The cosine similarity between semantic representation vectors is calculated. When the similarity is greater than the average of all semantic similarity results, it is determined to be a cross-modal similarity relationship, and a cross-modal similarity relationship is generated.

[0037] S3.4: Merge co-occurrence relations, syntactic dependency relations, and cross-modal similarity relations into relation candidates; Furthermore, co-occurrence relations, syntactic dependency relations, and cross-modal similarity relations are stored uniformly. Relationships involving the same entities are merged, retaining source information and relation type. The merged result is used as relation candidates.

[0038] S3.5: Based on the temporal sequence information and causal trigger words in the candidate relation and fragment set, perform causal direction verification to obtain a set of causal relations; Furthermore, the chronological order of events is determined based on the time sequence of entities in the fragment set in the candidate relation, and the direction of the relation is judged by combining the causal trigger words in the text table and the multimedia transcription table. When the semantic direction of the causal trigger word is consistent with the time sequence, the candidate relation is judged to be causal, thus forming a set of causal relations.

[0039] S3.6: Based on source credibility, freshness, and consistency, perform weighted calculations on causal relationships in the relation set to generate causal triples with confidence levels; Furthermore, based on each causal relationship in the relation set, the acquisition method in the source index is read and mapped to a source credibility weight according to a fixed-level rule. The fixed-level rule is that sensor acquisition corresponds to the upper level of source credibility weight, automatic log generation corresponds to the middle level of source credibility weight, and manual input corresponds to the lower level of source credibility weight. The upper, middle, and lower levels are used as ordinal labels and mapped to the range of zero to one in a fixed order to form source credibility weights that can participate in the calculation. The acquisition time and the current processing time are extracted from the source index to calculate the time difference. The time difference is input into the freshness calculation method that outputs the value according to the exponential decay law to obtain the freshness weight. The rate parameter of exponential decay is taken as the reciprocal of the average time window length covered by the fragment set to ensure that... The freshness weight is consistent with the time scale of the fragment set and its value is limited to the range of zero to one. For the same equipment identification, the same road segment location, and the same or similar fault category within the same time window, the causal relationship to be evaluated is compared with the set of causal relationships generated in the historical period one by one. The proportion of the three results of complete consistency, partial consistency, and complete inconsistency is statistically calculated and used as the consistency weight. The source credibility weight, freshness weight, and consistency weight are summed up with equal weight and mapped to the range of zero to one. The result is written into the triplet entry as the confidence of the causal relationship and a source traceability record is established with the source number to establish a one-to-one correspondence. This generates a causal triplet with confidence for subsequent causal chain splicing and evidence set generation.

[0040] S4: Concatenate causal triples into a set of causal chains, identify typical evolutionary patterns, bind source information to each node and edge of the causal chain, and generate a set of evidence. S4.1: Concatenate the causal triples with confidence levels according to device identifiers and time sequence to generate a set of causal chains; Furthermore, the causal triples with confidence scores are grouped according to device identifiers and then arranged in chronological order. When the tail entity matches the head entity of the next causal triple with confidence scores, sequential splicing is performed. When the sequential splicing is uninterrupted, a causal chain is formed. If the sequential splicing condition is not met, a new causal chain is started. For each causal chain, the device identifier, time start, time end, and the sequence of causal triples with confidence scores are recorded. All causal chains are summarized into a causal chain set. When multiple causal triples with confidence scores with the same timestamp appear, they are spliced ​​in descending order of confidence score, thus generating a causal chain set.

[0041] S4.2: Statistically analyze the frequency of occurrence of causal chain sets and perform pattern mining to identify typical evolutionary patterns; Furthermore, for each causal chain in the causal chain set, a sequence of relation types and a sequence of key entity categories are extracted as pattern descriptions. The frequency of occurrence of the pattern descriptions is statistically analyzed and sorted by frequency. Sequence pattern mining and frequent subgraph mining methods are used to mine patterns in the causal chain set. When the frequency of occurrence of a certain pattern description is not lower than the average frequency of occurrence of all pattern descriptions and the number of causal chains it covers is not lower than the median of the distribution of the number of causal chains covered, the pattern description is identified as a typical evolutionary pattern. The typical evolutionary pattern and the corresponding support sample list are output synchronously for subsequent steps. Thus, the typical evolutionary pattern is identified.

[0042] It should be noted that the mean and median are derived from the pattern frequency distribution and causal chain coverage distribution within the current statistical period.

[0043] S4.3: Based on the source index, bind source information to nodes and edges in the causal chain set; Furthermore, the source index includes source number, collection method, collection time, and collection location information. Each node in the causal chain set corresponds to an entity instance, and each edge corresponds to a causal triple with confidence. Based on the source index, a tracing entry is established for each entity instance and each causal triple with confidence. The tracing entry is written with the source number, collection method, collection time, and collection location, and is associated one-to-one with the entity instance or the causal triple with confidence. After the tracing entries of all nodes and all edges are bound, the causal chain set with bound tracing information is output, thus completing the binding of tracing information.

[0044] S4.4: Combine the typical evolutionary patterns and the causal chain set after binding the source information to generate an evidence set; Furthermore, the typical evolutionary pattern is used as a template to perform matching in the causal chain set after binding the source information. The matching condition is that the relationship type sequence and entity category sequence in the causal chain are consistent with the typical evolutionary pattern and the time order is consistent. When the matching is successful, the matched causal triple sequence with confidence and the corresponding source entry are extracted and evidence information unit is generated. The evidence information unit includes the causal chain identifier, the start and end position of the matching, the causal triple sequence with confidence and the list of source entries. All evidence information units are grouped and summarized according to the typical evolutionary pattern to form an evidence set, thus generating the evidence set.

[0045] S5: Based on the evidence set, merge semantically equivalent causal chain facts to form a candidate fact set, establish an evidence binding table, perform comprehensive credibility calculation on the candidate fact set, and generate a fused fact set; S5.1: Based on the evidence set, identify semantically equivalent causal chain facts and merge them to form a candidate fact set; Furthermore, using the evidence set as input, the causal triple sequences with confidence in the evidence set are standardized in terms of equipment name, fault category, handling measures, time expression, and geographical location expression to form comparable causal chain fact descriptions. Then, merging is performed according to the semantic equivalence judgment rules. The semantic equivalence judgment rules are: consistent equipment identification, consistent fault category, same causal path or existence of a superior inclusion relationship, adjacent or overlapping time windows and consistent geographical location or consistent range of the same road segment. Causal chain facts that meet the semantic equivalence judgment rules are merged into a semantically equivalent causal chain fact, and the evidence source and time range pointed to by the source index are recorded. All merged results constitute a candidate fact set, thus achieving the goal of forming a candidate fact set.

[0046] S5.2: Establish the correspondence between the candidate fact set and the evidence set, and generate an evidence binding table; Furthermore, using the evidence binding table as input, the source credibility, freshness, and consistency weights are calculated for each candidate fact and then weighted. Source credibility is derived from the source index in the evidence binding table that marks the collection method, mapped to a numerical value according to a fixed-level rule, and then averaged over the number of pieces of evidence to obtain the source credibility of the candidate fact. Freshness is derived from the time difference between the evidence time range in the evidence binding table and the current processing time, calculated using an exponential decay law, and then averaged over the number of pieces of evidence to obtain the freshness of the candidate fact. Consistency is derived from the candidate fact's consistency with the same equipment identifier, the same road segment location, and the same or similar fault category. The consistency of candidate facts is determined by comparing the proportion of consistent results in causal direction and causal path among candidate facts within the same time window. The three weights are summarized according to their importance ratio and mapped to a range of zero to one as the confidence level of the candidate facts, thus obtaining a set of candidate facts with confidence. At the same time, a fused set of facts is obtained according to the candidate fact confidence threshold and conflict resolution rules. The candidate fact confidence threshold is the average of the candidate fact confidence levels. The conflict resolution rule is that for the same device, the same fault category, and the same time window, only the candidate fact with the highest confidence level is retained, and the remaining candidate facts are used as alternative conclusions.

[0047] S5.3: Based on the evidence binding table, perform weighted calculations on source credibility, freshness, and consistency to obtain a set of candidate facts with confidence levels, and obtain the fused fact set; Furthermore, using the evidence binding table as input, the source credibility, freshness, and consistency weights are calculated for each candidate fact, and then weighted and calculated. Source credibility is derived from the source index in the evidence binding table indicating the collection method, mapped to a numerical value according to a fixed-level rule, and then averaged over the number of pieces of evidence to obtain the source credibility of the candidate fact. Freshness is derived from the difference between the evidence time range in the evidence binding table and the current processing time, calculated using time decay, and then averaged over the number of pieces of evidence to obtain the freshness of the candidate fact. Consistency is derived from the comparison of consistency between the candidate fact and other candidate facts on the same topic in terms of causal direction and causal path. If the proportion of candidate facts is consistent, the three weights are summarized according to their importance ratio and mapped to a range of zero to one as the confidence level of the candidate facts, thus obtaining a set of candidate facts with confidence. At the same time, a fused fact set is obtained based on the candidate fact confidence threshold and conflict resolution rules. The candidate fact confidence threshold is the average of the candidate fact confidence levels. The conflict resolution rule is that for the same device, the same fault category, and the same time window, only the candidate fact with the highest confidence level is retained, and the remaining candidate facts are used as alternative conclusions. This completes the goal of obtaining a set of candidate facts with confidence and obtaining a fused fact set.

[0048] S6: When the main conclusion in the fused fact set meets the credibility threshold for inclusion in the database, the main conclusion and the traceability index are written into the electromechanical operation and maintenance knowledge base. S6.1: Extract the main conclusions from the fused fact set and select the main conclusions that meet the credibility threshold for inclusion in the database; Furthermore, the fused fact set is grouped by device identifier, fault category, and time window. The confidence level, device identifier, fault category, time window, and geographical range information of each fused fact are read. Based on the built-in confidence ranking of each group, the fused facts with the highest confidence level are selected to form the master conclusion set. Then, the confidence level of each master conclusion in the master conclusion set is compared with the database confidence threshold one by one. The list of master conclusions that meet the database confidence threshold is output, and the master conclusions that do not meet the database confidence threshold are marked as candidate conclusions for subsequent updates. This completes the goal of extracting master conclusions from the fused fact set and selecting master conclusions that meet the database confidence threshold.

[0049] It should be noted that, taking the fused fact set as input, groups are established according to device identification, fault category, time window and geographical range. The confidence list of the fused fact set is extracted from each group, the average value of the confidence list is calculated and the average value is used as the confidence threshold for the corresponding group to be included in the database.

[0050] S6.2: Based on the traceability information, extract the traceability index of the main conclusion that meets the credibility threshold and write it into the electromechanical operation and maintenance knowledge base; Furthermore, based on the evidence binding table, a list of evidence information unit identifiers corresponding to each main conclusion is retrieved, and the source index list is read accordingly. The source number, collection method, collection time, and collection location are combined to form a traceability index, which is then organized together with the equipment identifier, fault category, time window, geographical range, and main conclusion content in the main conclusion as an entry record. The entry record is written into the electromechanical operation and maintenance knowledge base, and a one-to-one correspondence between the primary key and the traceability index is established, thus achieving the goal of writing the main conclusion and traceability index into the electromechanical operation and maintenance knowledge base.

[0051] In summary, this invention achieves the extraction of high-confidence causal relationships among multi-source heterogeneous data through the generation of causal triples, and achieves redundancy resolution and source traceability credibility assurance through the generation of fused fact sets. The combination of these two aspects enables the electromechanical maintenance knowledge base to simultaneously possess dynamic evolution and high reliability, thereby effectively supporting the preventive maintenance and intelligent decision-making of highway electromechanical equipment.

[0052] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for constructing a highway electromechanical maintenance knowledge base based on big data, characterized in that: include, Collect multi-source heterogeneous data on the operation and maintenance of electromechanical systems on highways, establish a source index, and preprocess the multi-source heterogeneous data to obtain standardized event tables, text tables, and multimedia transcription tables. The standardized event table, text table, and multimedia transcription table are segmented into a set of fragments using a sliding window, generating a semantic representation vector, and extracting the entity set and attribute table. Based on the fragment set, entity set, and attribute table, relation candidates are generated through co-occurrence relations, syntactic dependencies, and cross-modal similarity. Causal direction is then checked by combining temporal order and causal trigger words to generate causal triples with confidence. The causal triples are concatenated into a set of causal chains, and typical evolutionary patterns are identified. Source information is bound to each node and edge of the causal chain to generate a set of evidence. Based on the evidence set, semantically equivalent causal chain facts are merged to form a candidate fact set, and an evidence binding table is established. The comprehensive credibility of the candidate fact set is calculated to generate a fused fact set. When the main conclusion in the fused fact set meets the credibility threshold for inclusion in the database, the main conclusion and the traceability index are written into the electromechanical operation and maintenance knowledge base.

2. The method for constructing a highway electromechanical maintenance knowledge base based on big data as described in claim 1, characterized in that: The specific steps for collecting multi-source heterogeneous data on highway electromechanical operation and maintenance and establishing a source index are as follows. Collect multi-source heterogeneous data on the operation and maintenance of electromechanical systems on highways; A source index is established based on the source number, collection method, collection time, and collection location of multi-source heterogeneous data.

3. The method for constructing a highway electromechanical maintenance knowledge base based on big data as described in claim 2, characterized in that: After preprocessing the multi-source heterogeneous data, standardized event tables, text tables, and multimedia transcription tables are obtained. The specific steps are as follows. The multi-source heterogeneous data is uniquely encoded by the device, time-aligned, and spatially mapped to generate pre-processed multi-source heterogeneous data. The preprocessed multi-source heterogeneous data is divided into data categories and formatted uniformly to generate standardized event tables, text tables, and multimedia transcription tables.

4. The method for constructing a highway electromechanical maintenance knowledge base based on big data as described in claim 3, characterized in that: The process of segmenting the standardized event table, text table, and multimedia transcription table into a set of fragments using a sliding window to generate semantic representation vectors involves the following specific steps. Standardized event tables, text tables, and multimedia transcription tables are divided into time intervals using a sliding window to generate a collection of segments; Semantic encoding is performed on the fragment set to generate a preliminary semantic representation of the fragments, and the semantic representation vector is obtained through vector fusion.

5. The method for constructing a highway electromechanical maintenance knowledge base based on big data as described in claim 4, characterized in that: The specific steps for extracting the entity set and attribute table are as follows. Extract the entity set from the fragment set based on the semantic representation vector; The attributes of the entity set are identified, unified, and standardized to generate an attribute table.

6. The method for constructing a highway electromechanical maintenance knowledge base based on big data as described in claim 5, characterized in that: The process of generating relation candidates based on the fragment set, entity set, and attribute table using co-occurrence relations, syntactic dependencies, and cross-modal similarity involves the following specific steps. The co-occurrence of entity sets within a set of statistical fragments is analyzed to generate co-occurrence relationships. The dependency structure between entity sets is identified by syntactic dependency analysis, and syntactic dependency relationships are generated. Calculate the semantic similarity of fragments from different sources in a collection and generate cross-modal similarity relationships; Co-occurrence relations, syntactic dependency relations, and cross-modal similarity relations are merged into relation candidates.

7. The method for constructing a highway electromechanical maintenance knowledge base based on big data as described in claim 6, characterized in that: The process of combining time sequence and causal trigger words to perform causal direction testing and generate causal triples with confidence levels is described in the following steps. Based on the temporal sequence information and causal trigger words in the candidate relation and fragment set, a causal direction test is performed to obtain a set of relations with causal characteristics; The causal relationships in the relation set are weighted based on source credibility, freshness, and consistency to generate causal triples with confidence.

8. The method for constructing a highway electromechanical maintenance knowledge base based on big data as described in claim 7, characterized in that: The specific steps for assembling causal triples into a set of causal chains, identifying typical evolutionary patterns, binding source information to each node and edge of the causal chain, and generating an evidence set are as follows. The causal triples with confidence are concatenated according to device identification and time sequence to generate a set of causal chains; The frequency of occurrence of causal chain sets is statistically analyzed and pattern mining is performed to identify typical evolutionary patterns; Based on the source index, source information is bound to nodes and edges in the causal chain set; An evidence set is generated by combining typical evolutionary patterns with the causal chain set after binding source information.

9. The method for constructing a highway electromechanical maintenance knowledge base based on big data as described in claim 8, characterized in that: The process involves merging semantically equivalent causal chain facts based on an evidence set to form a candidate fact set, establishing an evidence binding table, calculating the comprehensive credibility of the candidate fact set, and generating a fused fact set. The specific steps are as follows: Based on the evidence set, semantically equivalent causal chain facts are identified and merged to form a candidate fact set; Establish the correspondence between the candidate fact set and the evidence set, and generate an evidence binding table; Based on the evidence binding table, a weighted calculation is performed on the source credibility, freshness, and consistency to obtain a set of candidate facts with confidence, and then a fused set of facts is obtained.

10. The method for constructing a highway electromechanical maintenance knowledge base based on big data as described in claim 9, characterized in that: When the main conclusion in the fused fact set meets the database credibility threshold, the main conclusion and the traceability index are written into the electromechanical operation and maintenance knowledge base. The specific steps are as follows. Extract the main conclusions from the fused fact set and select the main conclusions that meet the credibility threshold for inclusion in the database; Based on the traceability information, the traceability index of the main conclusion that meets the credibility threshold is extracted and written into the electromechanical operation and maintenance knowledge base.

Images