A line loss intelligent analysis method and system based on NLP

Abstract

This invention provides a line loss intelligent analysis method and system based on NLP, relating to the field of power system operation and maintenance technology. The method includes the following steps: acquiring monitoring information of the target object and constructing an information sequence; parsing the feature items in the information sequence and calculating the risk increment value of the target object under different evaluation dimensions; cumulatively updating the state scores of each evaluation dimension of the target object based on the risk increment value to obtain the dynamic state of the target object; monitoring the evolution sequence of the state scores of each evaluation dimension over time in the dynamic state, matching it with a preset progressive fault mode, and outputting risk warning information when the match is successful. The method of this invention aims to solve the problems of difficulty in accurately assessing risk, lack of state continuity, and inability to deeply identify systemic defects in existing technologies, improving the accuracy and timeliness of line loss analysis, realizing preventive maintenance, reducing warning lag, and deeply identifying systemic risks.

Description

Technical Field

[0001] This invention relates to the field of power system operation and maintenance technology, and more specifically, to an intelligent line loss analysis method and system based on NLP. Background Technology

[0002] In the field of power system operation and maintenance, line loss analysis is a core component in ensuring the economical operation of the power grid. Traditional methods mainly rely on manual experience to analyze structured data, which is insufficient to effectively handle the ever-increasing volume of unstructured text data such as equipment operation logs and inspection reports. Although existing technologies have attempted to incorporate natural language processing techniques, three key shortcomings remain: First, when faced with colloquial and non-standardized text descriptions from frontline staff, the system struggles to accurately identify equipment status and risk levels. For example, it cannot accurately associate "the porcelain insulator is cracked" with the standard term "the insulator is damaged." Secondly, the existing system lacks the ability to track changes in the physical identity of power equipment. When lines are modified or equipment is replaced, it is impossible to establish continuity of status data between the old and new equipment. Furthermore, existing solutions can only handle isolated events and cannot identify progressive degradation patterns across time and levels. For example, they cannot link seemingly independent phenomena such as "water accumulation in cable trenches" and "voltage fluctuations" that are scattered at different times into the same fault development chain.

[0003] These problems lead to delayed system warnings, making true preventative maintenance difficult. More seriously, the complexity of the hierarchical topology of power equipment makes it difficult to accurately transmit the impact of local state changes. Comprehensive state assessments of higher-level equipment lack support from fine-grained data from lower-level equipment, while maintenance decisions for lower-level equipment cannot obtain a global perspective from the higher-level system. This data fragmentation keeps line loss analysis at the level of superficial phenomenon correlations, failing to deeply identify potential systemic risks.

[0004] There is currently no effective technical solution to the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent line loss analysis method and system based on NLP, which aims to solve the problems of difficulty in accurately assessing risks, lack of state continuity and inability to deeply identify systemic defects in the existing technology. It has the advantages of improving the accuracy and timeliness of line loss analysis, realizing preventive maintenance, reducing early warning lag and deeply identifying systemic risks.

[0006] In a first aspect, the present invention provides an NLP-based intelligent line loss analysis method, comprising the following steps: S1. Obtain monitoring information for the target object; S2. Standardize the monitoring information and associate the processed monitoring information with the target object to construct an information sequence of the target object in the time dimension; S3. Analyze the feature items in the information sequence, and calculate the risk increment value of the target object under different evaluation dimensions based on the contextual attributes corresponding to the feature items and the preset degradation knowledge rules; S4. Based on the risk increment value, the state scores of each evaluation dimension of the target object are cumulatively updated to obtain the dynamic state of the target object; S5. Monitor the evolution sequence of the state scores of each evaluation dimension in the dynamic state over time; S6. Match the evolution sequence with a preset progressive failure mode, and output risk warning information for the target object when the match is successful.

[0007] The NLP-based intelligent line loss analysis method provided by this invention can effectively process unstructured text data, construct a continuous sequence of equipment status information, and identify progressive degradation patterns, thereby realizing intelligent analysis and risk warning of power equipment line losses and overcoming the limitations of traditional methods in data processing and fault identification.

[0008] Secondly, the present invention provides an NLP-based intelligent line loss analysis system, comprising: The acquisition module is used to acquire monitoring information for the target object; A construction module is used to standardize the monitoring information and associate the processed monitoring information with the target object to construct an information sequence of the target object in the time dimension. The calculation module is used to parse the feature items in the information sequence and calculate the risk increment value of the target object under different evaluation dimensions based on the contextual attributes corresponding to the feature items and the preset degradation knowledge rules. The update module is used to cumulatively update the state scores of each evaluation dimension of the target object based on the risk increment value, so as to obtain the dynamic state of the target object; The monitoring module is used to monitor the evolution sequence of the state scores of each evaluation dimension in the dynamic state over time. The matching module is used to match the evolution sequence with a preset progressive failure mode, and output risk warning information for the target object when the match is successful.

[0009] As can be seen from the above, the NLP-based intelligent line loss analysis method provided by this invention effectively solves the shortcomings of the prior art through the following steps: First, monitoring information containing unstructured text data is obtained through step S1, and NLP technology is used to standardize this unstructured text data in step S2, thereby overcoming the problem that traditional methods have difficulty in handling colloquial and non-standardized descriptions, and achieving accurate identification of equipment status.

[0010] Secondly, in step S2, by associating the processed monitoring information with the target object, a continuous information sequence of the target object in the time dimension is constructed, which effectively solves the data fragmentation problem caused by the change of the entity identity of power equipment and ensures the integrity of the equipment life cycle information.

[0011] Furthermore, by parsing the feature items in the information sequence in step S3 and calculating the risk increment value in combination with the degradation knowledge rules, and by accumulating and updating the state score in step S4, the dynamic state of the target object is obtained, laying the foundation for subsequent fault identification.

[0012] Finally, by monitoring the evolution sequence of the dynamic state in step S5 and matching it with a preset progressive fault mode in step S6, the progressive degradation process of the equipment from initial anomaly to final failure is identified, effectively solving the shortcomings of existing systems such as delayed early warning and difficulty in implementing preventive maintenance. In summary, the method of this application can significantly improve the accuracy of line loss analysis and the timeliness of early warning, providing strong technical support for the economic operation and safety of the power grid.

[0013] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Attached Figure Description

[0014] Figure 1 This is a flowchart of an NLP-based intelligent line loss analysis method provided in an embodiment of the present invention.

[0015] Figure 2 This is a schematic diagram of a line loss intelligent analysis system based on NLP provided in an embodiment of the present invention.

[0016] Label Explanation: 100. Acquisition Module; 200. Construction Module; 300. Calculation Module; 400. Update Module; 500. Monitoring Module; 600. Matching Module. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0018] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0019] In traditional intelligent line loss analysis of power systems, existing technologies cannot effectively correlate unstructured text information from different time points and sources to identify progressive fault modes. The essence of this problem lies in the system's lack of ability to construct the evolution history of the target object's state, making it impossible to integrate fragmented, time-spanning text descriptions into a coherent fault development chain. Consequently, key performance indicators of the system, including the timeliness of fault warnings and the completeness of the analysis process, are significantly affected, enabling the system to handle only isolated events and failing to achieve early identification of progressive degradation processes.

[0020] For example, during the maintenance of a 10kV distribution line, on Monday, patrol personnel reported "slight water accumulation in the cable trench," which was independently processed by the system as a low-risk event. On Wednesday, a community electrician reported "unstable voltage and flickering lights for nearby users," which was also classified as a low-priority event. By Friday, the line loss had abnormally increased, but the system failed to correlate the aforementioned information, analyzing only the data from that day and failing to identify a clear cause of the fault. In reality, the water accumulation in the cable trench gradually eroded the weak points in the cable sheath, leading to a decline in insulation performance and ultimately causing abnormal line loss. In this scenario, the system could not connect the scattered reports of "water accumulation," "unstable voltage," and "increased line loss" to form a complete fault evolution sequence, thus missing the opportunity to issue an early warning. Furthermore, this problem prevents the system from constructing the dynamic state evolution process of the target object, limiting the analysis results to the processing of single events and failing to reflect the gradual changes in equipment status.

[0021] If the above problems are not addressed, the system will be unable to identify line loss anomalies caused by progressive degradation, rendering the fault early warning mechanism ineffective. Consequently, maintenance personnel can only perform post-incident troubleshooting, increasing the complexity of fault handling. Furthermore, the accumulation of progressive faults may evolve into severe equipment damage or power outages, affecting the safe and stable operation of the power grid. In particular, the system's lack of time-series correlation capabilities for multi-source information prevents potential risks from being identified in their early stages, thus reducing the overall reliability of line loss analysis.

[0022] For reference, see the appendix. Figure 1 This invention provides an NLP-based intelligent line loss analysis method, comprising the following steps: S1. Obtain monitoring information for the target object; the monitoring information includes unstructured text data describing the operating status of the target object; S2. Standardize the monitoring information and associate the processed monitoring information with the target object to construct the information sequence of the target object in the time dimension; S3. Analyze the feature items in the information sequence, and calculate the risk increment value of the target object under different evaluation dimensions based on the contextual attributes corresponding to the feature items and the preset degradation knowledge rules; S4. Based on the risk increment value, the state scores of each evaluation dimension of the target object are cumulatively updated to obtain the dynamic state of the target object; S5. Monitor the evolution sequence of the state scores of each evaluation dimension in the dynamic state over time; S6. Match the evolution sequence with the preset progressive failure mode, and output risk warning information for the target object when the match is successful.

[0023] For ease of understanding, the following explains some key terms in this embodiment: Monitoring information refers to various data concerning the operational status of a target object (such as power equipment or lines). Its main form is unstructured text data, such as inspection reports from maintenance personnel, equipment operation logs, and fault descriptions. This information serves as the raw input for intelligent line loss analysis.

[0024] Standardization processing refers to the preprocessing of raw unstructured text data, including text cleaning, typo correction, and standardization of industry terminology, with the aim of eliminating noise and inconsistencies in the text to make it meet the requirements of subsequent analysis.

[0025] Information sequence: refers to the continuous information flow of a target object along a time axis, constructed by associating standardized monitoring information with a specific target object according to the time dimension. This sequence reflects the changes in the state of the target object over time.

[0026] Feature terms: These are keywords, phrases, or concepts that are extracted from the information sequence and can reflect the operational status or potential risks of the target object.

[0027] Contextual attributes: These refer to the contextual information associated with a feature, used to clarify the meaning, intensity, or tendency of the feature, such as describing the urgency of the equipment status or the reliability of the report source.

[0028] Deterioration knowledge rules: These refer to a set of pre-defined rules based on the experience of power industry experts and historical data. They are used to guide the system on how to assess the risk of a target object under different evaluation dimensions based on feature items and their contextual attributes.

[0029] Risk increment value: This refers to the quantitative value calculated based on the contextual attributes of feature items and degradation knowledge rules, which affects the state scores of each evaluation dimension of the target object. A positive value indicates increased risk, and a negative value indicates decreased risk.

[0030] Evaluation dimensions: These refer to multiple aspects used to assess the operational status of a target object, such as insulation status, mechanical structure stability, environmental exposure, operational efficiency, and power supply quality.

[0031] State score: This refers to a numerical value that quantifies the current state of the target object under each evaluation dimension. Generally, a lower score indicates a higher risk or a worse state.

[0032] Dynamic status: refers to the set of status scores of the target object across all evaluation dimensions. These scores are cumulatively updated based on new risk increments, thereby reflecting the overall health status of the target object in real time.

[0033] Evolution sequence: refers to the trajectory or trend of the state scores of each evaluation dimension changing over time in a dynamic state. By monitoring the evolution sequence, potential degradation processes can be identified.

[0034] Progressive failure modes: These are pre-defined models that describe the gradual deterioration process of equipment or lines from normal operation to failure. These models are based on historical failure data and expert experience and are used to identify early signs of failure.

[0035] Risk warning information: This refers to the prompt information output by the system when the monitored evolution sequence successfully matches the preset progressive failure mode. It aims to remind operation and maintenance personnel to pay attention to potential risks and take preventive measures.

[0036] This application proposes an NLP-based intelligent line loss analysis method, aiming to address the problem in existing technologies that cannot effectively correlate unstructured text information from different times and sources to identify progressive failure modes. This method achieves intelligent analysis and risk warning of power system line losses through a series of steps.

[0037] When obtaining monitoring information for a target object, the system will set up a dedicated receiving module, such as a data entry based on a message queue, to receive text information from different sources in real time. These sources can include free text reported by on-site operation and maintenance personnel through mobile terminals, automatically generated work order system records, and scanned copies or electronic documents of historical inspection reports. For example, during an inspection, an operation and maintenance personnel enters "The insulator of Tower No. 45 on 10kV Line A has slight cracks" through a mobile terminal, or the system automatically generates a work order regarding "Abnormal oil level of a certain transformer". These all belong to monitoring information. This information exists in the form of unstructured text data, describing the operating status of the target object (such as the insulator of Tower No. 45 on 10kV Line A, a certain transformer).

[0038] Standardize the monitoring information and associate the processed monitoring information with the target object to construct an information sequence of the target object in the time dimension. After receiving the text, the system will start a text preprocessing unit. This unit will first perform basic text cleaning, such as using regular expressions to remove special symbols and extra spaces. Then, it will use a spelling corrector based on rules and dictionaries. For example, a mapping table containing common spelling mistakes in the power industry and their correct spellings is preset to automatically correct the common misspelling "jué yuán zǐ" of "insulator" to "insulator". At the same time, the system will maintain a dictionary of power industry terms to standardize colloquial abbreviations (such as mapping "porcelain insulator" to "insulator") or common names. Each processed piece of information will be assigned a unique identifier, an accurate timestamp (accurate to the second), the original source (such as "mobile terminal - inspector Zhang San"), and the reporter information. For the preliminarily processed text, the system will start an entity recognition module. This module will use named entity recognition technology to identify physical entities mentioned in the text. For example, through a pre-trained entity recognition model, specific equipment or line names such as "10kV a certain line", "Tower No. 45", and "a certain transformer" are identified. For colloquial or ambiguous descriptions, such as "the insulator on that line", "the equipment next to the substation", the system will start an entity disambiguation and association logic. This logic will make judgments by comprehensively using various information: a. The geographical location of the reporter: If the GPS location of the reporter is near a certain line segment or equipment, the system will preferentially associate the description with the equipment in that area. For example, if the reporter is near Tower No. 45 on 10kV Line A and the text mentions "the insulator on that line", the system will tend to associate it with the insulator of Tower No. 45 on 10kV Line A.

[0039] b. Area names mentioned in the report: If the text contains area information such as "a certain community" or "a certain village", the system will query the list of power equipment in that area in the geographic information system to narrow down the scope of entity recognition.

[0040] c. Historical records: The system will query the devices or lines that the reporting person has frequently reported in the past, as well as devices that have recently experienced frequent problems in the area, as an auxiliary basis for judgment.

[0041] d. Semantic similarity: For ambiguous descriptions, the system will calculate the semantic similarity between the description text and known device names and device component names, and select the entity with the highest similarity for association.

[0042] Once a physical entity is identified, the information is accurately "anchored" to the corresponding physical entity and stored in the entity's dedicated information timeline.

[0043] The system analyzes the feature terms in the information sequence and calculates the risk increment value of the target object under different evaluation dimensions based on the contextual attributes corresponding to the feature terms and the preset degradation knowledge rules. For text information that has been associated with a specific physical entity, the system will activate a health dimension analysis module. The core of this module is a rule engine and semantic analysis based on power system degradation knowledge. a. Keyword and phrase matching: The system maintains a mapping table between the health dimensions of power equipment and related keywords / phrases. For example, "humid environment" and "heavy rain" are mapped to the "insulation status" dimension; "nearby dust" and "dirt" are mapped to the "flashover risk" dimension; "unstable voltage" and "light flickering" are mapped to the "operating efficiency" or "power supply quality" dimension.

[0044] b. Context and Intensity Analysis: The system analyzes the context and sentiment of the description. For example, in the sentence "The insulator on pole #45 seems a bit off," the word "seems" is no longer simply considered as uncertainty. Instead, it is amplified in terms of risk level by combining the experience and intuition of frontline personnel through a pre-set "intuitive judgment weighting table." For instance, if the description of "a bit off" is usually associated with subsequent malfunctions, its weight will be higher than that of a general description.

[0045] c. Risk Increment Calculation: Based on the matched keywords, contextual analysis results, and pre-defined power system degradation knowledge, the system calculates a "risk increment value" for each affected health dimension. For example, if "humid environment" is identified, and the insulator type in that area is sensitive to humidity, the system will calculate a positive insulation degradation risk increment value, such as +0.5 points. This calculation formula can be expressed as: Increment Value = Base Risk Value × Contextual Weight × Equipment Sensitivity Coefficient. Wherein, the base risk value is the inherent risk value of the keyword, the contextual weight reflects the intensity and certainty of the description, and the equipment sensitivity coefficient is adjusted according to the specific type of equipment and operating environment.

[0046] The system accumulates and updates the status scores of each evaluation dimension of the target object based on the risk increment value, thus obtaining the dynamic status of the target object. Each physical device or line maintains a "digital portrait" data structure in the system. This portrait contains multiple health dimensions (e.g., insulation status, mechanical structure stability, environmental exposure, operating efficiency, power supply quality, etc.), each with a current status score (e.g., 0-100 points, with lower scores indicating higher risk). When the health dimension parsing step outputs a new risk increment value, the system immediately updates the status score of the corresponding dimension. The update logic is cumulative: new score = old score + increment value. To reflect the dynamic changes in risk, the system also introduces a "risk decay mechanism." If a dimension has no new risk increments for a period of time, its score will slowly recover (e.g., automatically increasing by 0.1 points per day), simulating the situation after the risk naturally subsides or is addressed. For example, if the "insulation status" dimension decreases due to "humid environment," but the weather subsequently clears up and no new humidity reports, the score of that dimension will gradually recover. This mechanism ensures that the digital portrait can reflect the overall health status and evolution trend of the equipment or line in real time and dynamically.

[0047] The system monitors the evolution of status scores for each evaluation dimension over time in dynamic status monitoring. It maintains a historical status record database for each device or line's digital profile. This database records the score changes for each health dimension at different time points. A trend analysis module runs periodically (e.g., hourly or daily) to analyze these historical records and identify trends in the scores of each health dimension over time. For example, it detects whether the "insulation status" score shows a downward trend for several consecutive days or weeks, or whether the "pollution flashover risk" score continuously increases during a specific season (e.g., the rainy season). Trend analysis can employ simple linear regression, moving averages, or more complex time series analysis methods. For example, for the insulation status score S(t), the system calculates its average rate of change over the past N time points: Average rate of change = (S(t) - S(tN)) / N. If the average rate of change is consistently negative and exceeds a certain threshold, it indicates a trend of continuous deterioration.

[0048] The system matches the evolution sequence with preset progressive failure modes and outputs risk warning information for the target object when a match is successful. The system continuously monitors the state evolution sequence and compares it with preset progressive failure modes. These failure modes are summarized based on the knowledge of power industry experts and historical failure data, such as the "insulation degradation mode" (manifested as a continuous decline in insulation condition score while the environmental exposure score increases) and the "pollution flashover risk accumulation mode" (manifested as a continuous increase in pollution flashover risk score while reports of environmental humidity or dust increase). When the cumulative change in a certain health dimension reaches a preset risk threshold (e.g., insulation condition score below 60 points), or when multiple related dimensions show a deteriorating trend simultaneously (e.g., insulation condition score decreases and operating efficiency score also decreases), the system identifies a potential progressive failure mode. A predictive warning module immediately generates a predictive warning. The warning information clearly indicates the affected equipment or line, the potential fault type (e.g., "insulation degradation leads to increased line loss risk"), the risk level (e.g., "moderate risk, investigation recommended soon"), and the suggested investigation direction. For example, the warning message might indicate: "The insulation condition score of tower No. 45 on the 10kV A line continues to decline and has fallen below the warning threshold. Combined with the recent increase in dampness reports, there is a risk of increased line loss due to insulation deterioration. It is recommended that inspection personnel be immediately arranged to conduct a key inspection of the insulators of this tower." This achieves a shift from "post-event tracing" to "pre-event prediction."

[0049] The following example will provide a more detailed explanation of the above technical solution: Suppose that in a certain power line area, the target object is "a section of cable on a 10kV B line". On Monday, the system receives an inspection report from user A stating that "there is slight water accumulation in the cable trench of a certain section of this line". After this unstructured text data is acquired, it is first standardized, for example, removing modifiers such as "slight", and identifying "water accumulation" as a feature related to "dampness". The system associates this processed information with the target object "a section of cable on a 10kV B line" and records its timestamp as Monday, constructing the initial part of the cable information sequence. According to the degradation knowledge rules, the system identifies that "water accumulation" is related to the "insulation condition" evaluation dimension and calculates a risk increment value (e.g., +0.5 points), causing the cable's "insulation condition" score to drop from the initial 90 points to 89.5 points.

[0050] On Wednesday, the system received another report from user B, stating that "users in the nearby area are complaining of unstable voltage and occasional flickering lights." This information, also standardized, was linked to "a section of cable on the 10kV B line." The system parsed out features such as "unstable voltage" and "flickering lights," and based on degradation knowledge rules, correlated them with the "power supply quality" evaluation dimension, calculating another risk increment (e.g., +0.8 points), causing the cable's "power supply quality" score to drop from 95 to 94.2. At this point, the cable's dynamic status included updated scores for both "insulation status" and "power supply quality."

[0051] The system continuously monitors the evolution sequence of each evaluation dimension value in the dynamic status of the cable. On Friday, the system detected that the "insulation status" score had been continuously decreasing from Monday to Friday, and the "power supply quality" score also showed a downward trend. The system matched this evolution sequence of "continuously declining insulation status" and "declining power supply quality" with the preset progressive fault mode of "cable insulation deterioration leading to increased line loss". Since the status scores of both dimensions continued to deteriorate and the decline reached the preset threshold, the system successfully matched the fault mode.

[0052] Upon successful matching, the system immediately outputs a risk warning for a section of cable on the 10kV Line B, such as: "Warning: The insulation condition of a section of cable on the 10kV Line B is continuously deteriorating, resulting in a decline in power supply quality and a risk of increased line loss. It is recommended to immediately arrange for personnel to inspect the cable trench." This example demonstrates that this method can link scattered information from different times and sources (Monday's "water accumulation" report and Wednesday's "voltage instability" report) to construct a time-series information sequence of the target object, and based on this, identify progressive fault modes, thereby achieving early warning.

[0053] The NLP-based intelligent line loss analysis method proposed in this application demonstrates its significant advantages in addressing the challenges faced in power system operation and maintenance through the above examples. Traditional existing analysis systems, when faced with reports of "water accumulation" on Mondays and "voltage instability" on Wednesdays, often treat them as independent, low-risk events, failing to correlate them. This causes the system to miss opportunities to issue early warnings at the nascent stage of a fault, keeping analysis work at a reactive level rather than truly enabling proactive prediction.

[0054] In contrast, this application achieves breakthroughs through the following key technical points: First, by standardizing and associating unstructured text data with entities, this method can accurately "anchor" scattered monitoring information to specific target objects and construct a continuous information sequence of the target object in the time dimension. This solves the problem that traditional systems cannot effectively integrate multi-source heterogeneous information. Second, this method introduces contextual attributes and degradation knowledge rules to deeply analyze the feature items in the information sequence and calculate the risk increment value. This enables the system to more accurately assess risks, avoid misjudgments caused by isolated event analysis, and quantify the impact of different factors on equipment status. Third, by accumulating and updating the status scores of each evaluation dimension of the target object and introducing a risk decay mechanism, this method can reflect the comprehensive health status and its evolution trend of equipment or lines in real time and dynamically, overcoming the shortcomings of untimely status updates in traditional systems. Finally, by monitoring the evolution sequence of status scores and matching it with preset progressive failure modes, this method achieves early identification and predictive warning of progressive failures. This enables maintenance personnel to take preventative measures before a failure occurs, thus transforming "post-event tracing" into "pre-event prediction," significantly improving the reliability and economy of power system operation.

[0055] In some embodiments, step S2, the step of standardizing the monitoring information, includes: S21A1. Perform text cleaning, typo correction, and industry terminology standardization on monitoring information; among which, when standardizing industry terminology, identify intensity indicators in the monitoring information that describe the status of equipment or the urgency of risks; S21A2. Based on the preset intensity level mapping rules, the intensity indicator words are quantified into intensity values; S21A3. The strength value is attached as an attribute to the standardized industry terminology to obtain the standardized industry terminology for the attached strength value, which is then used as the processed monitoring information.

[0056] Specifically, text cleaning, typo correction, and industry terminology standardization are crucial steps in ensuring the quality and consistency of input data. Text cleaning aims to remove noise from the original text, such as using regular expressions to remove special characters, extra spaces, HTML tags, or non-text characters, or using natural language processing libraries (such as NLTK and SpaCy) for word segmentation and lexical reconstruction. Typo correction is used to correct spelling or input errors in the text. This can be achieved through mapping and replacement using preset rules and dictionaries. For example, maintaining a mapping table containing common typos in the power industry and their correct spellings can automatically correct common typos of "insulator" to "insulator"; or automatic correction can be performed using statistical models (such as N-gram models) or deep learning models (such as Seq2Seq models). Industry terminology standardization involves converting colloquial and non-standardized descriptions into unified professional terms. For example, maintaining a power industry terminology dictionary maps "porcelain insulator" to "insulator," or "the tree is too tall" to "tree obstacle." This can also be achieved by constructing a domain ontology or utilizing contextual semantic matching techniques. In standardizing industry terminology, this application particularly emphasizes identifying intensity indicators in monitoring information that describe equipment status or the urgency of risks, such as "slight," "serious," "urgent," and "something seems wrong." This can be achieved through predefined keyword lists, rule-based pattern matching, or by using sentiment analysis models and pre-trained sequence labeling models (such as Bi-LSTM-CRF) to identify words or phrases expressing intensity or urgency in the text.

[0057] Based on predefined intensity level mapping rules, intensity indicators are quantified into intensity values, aiming to transform these subjective and vague intensity descriptions into objective and calculable numerical values. The predefined intensity level mapping rules can be defined by domain experts based on experience, for example, mapping "slight" to 1, "moderate" to 3, and "severe" to 5; or a more refined quantification model can be established by statistical analysis of historical data combined with expert knowledge; or a fuzzy logic method can be used to map vague linguistic descriptions to continuous numerical ranges. The quantification process can be a direct lookup table mapping or a numerical transformation based on functions or machine learning models.

[0058] Attaching intensity values ​​as attributes to standardized industry terms, resulting in standardized industry terms with attached intensity values, and using these as processed monitoring information, aims to tightly bind quantified intensity information with corresponding standardized industry terms, forming a richer and more informative structured data unit. This attachment can be achieved in various ways, such as storing standardized terms and intensity values ​​as key-value pairs in JSON objects, using attribute tags in XML documents, as independent fields associated with standardized terminology records in relational databases, or as node attributes in graph databases. Ultimately, these standardized industry terms with attached intensity values ​​constitute the processed monitoring information, providing more refined and accurate input for subsequent risk assessments and status updates.

[0059] The solution in this application systematically processes raw, unstructured monitoring information, particularly the implicit information on equipment status or the urgency of risks, through the aforementioned steps. First, text cleaning, typo correction, and industry terminology standardization ensure the standardization and consistency of the text content, eliminating ambiguity caused by non-standard language. In this process, the system specifically identifies indicators describing intensity or urgency. Subsequently, based on preset intensity level mapping rules, these identified intensity indicators are converted into specific numerical values, achieving objective quantification of subjective descriptions. Finally, these quantified intensity values ​​are used as attributes and attached to corresponding standardized industry terms, thus forming a standardized monitoring information that also includes quantified intensity information. This processing method enables subsequent analysis not only to understand the reported event content but also to perceive the degree or urgency of the event, providing more accurate input for parsing feature items in the information sequence and calculating risk increment values ​​in subsequent step S3, thereby improving the accuracy and reliability of the entire intelligent line loss analysis method. In this way, this application effectively solves the problem that intensity information is easily lost or difficult to quantify when processing unstructured text data, enabling the system to understand the true situation conveyed by the operation and maintenance text more comprehensively and deeply.

[0060] The following is a concrete example to illustrate this. Suppose the system receives a free-text report from a frontline maintenance worker: "The insulator on pole number 45 seems a bit off." In step S21A1, the text preprocessing unit first cleans the text, removing any special characters. Next, a typo corrector ensures the text is error-free. Then, using an industry terminology dictionary, the colloquial "insulator" is standardized to "insulator." Simultaneously, the system identifies the phrase "seems a bit off" as a strength indicator describing the equipment status, indicating a minor anomaly. In step S21A2, the system, according to a preset strength level mapping rule, quantifies "seems a bit off" as a strength value of 2 (assuming the strength value range is 1-5, with 1 being the lowest and 5 the highest). In step S21A3, the system attaches this strength value 2 as an attribute to the standardized industry term "insulator," forming standardized information with a strength attribute, which can be represented as {"equipment component": "insulator," "status description": "anomaly," "strength value": 2}. For example, if the report is "The tree under the power line is too tall and about to hit it," the system will standardize "The tree is too tall and about to hit it" as "tree obstacle," and identify "about to hit" as an intensity indicator, quantifying it as an intensity value of 4. This will ultimately form {"Risk Type": "Tree Obstacle," "Urgency Level": "High," "Intensity Value": 4}. These standardized industry terms with attached intensity values ​​are then used as processed monitoring information and enter into subsequent analysis processes.

[0061] Through the above technical solution, this application can effectively extract and quantify the intensity information of equipment status or risk urgency from unstructured text data, avoiding the loss of this key information during standardization processing. This enables subsequent risk increment calculations and dynamic status updates to be based on more comprehensive and refined data, thereby significantly improving the accuracy and reliability of target object status assessment. The system not only identifies "what happened," but also accurately perceives "the extent of the occurrence," which is of great significance for identifying progressive failure modes and achieving early prediction. It can detect potential line loss risks earlier and more accurately, providing a more reliable basis for power system operation and maintenance decisions.

[0062] In some embodiments, step S2, which involves associating the processed monitoring information with the target object to construct an information sequence of the target object over time, includes: S21B1. Identify power system topology change event information; topology change event information includes equipment replacement records, line modification records, or temporary bypass operation records; S21B2. When a topology change event is detected, a new entity identity version is created for the affected target object; the new entity identity version is associated with the old entity identity version, and the change time is recorded; S21B3. Based on the timestamp and entity identity version record of the processed monitoring information, determine whether the processed monitoring information should be associated with the current entity identity version or its historical entity identity version, and associate the processed monitoring information with the target object corresponding to the determined entity identity version; S21B4. If the target object is changed, the information sequence of the old entity identity version and the information sequence of the new entity identity version are logically integrated to construct a continuous information sequence of the target object in the time dimension.

[0063] Specifically, when identifying power system topology change events, this step aims to ensure the system can proactively detect changes in the physical structure or configuration of power equipment or lines, providing triggering conditions for subsequent entity identity version management. One implementation method is to integrate with the data interface of the Power Asset Management System (EAM) or Geographic Information System (GIS) to receive real-time information such as equipment ledger updates and work order completion status. For example, when the EAM records that a transformer has been replaced with a new model, or the GIS updates the route of a certain line segment, the system can identify the corresponding topology change event. Another implementation method is to identify statements describing equipment replacement, line modification, or temporary bypass operations by performing keyword and pattern matching on text reports submitted by maintenance personnel. For example, natural language processing technology can be used to identify key phrases such as "replaced XX model circuit breaker," "XX line underwent capacity expansion," and "temporarily connected bypass."

[0064] When a topology change event is detected, a new entity identity version is created for the affected target object. The new entity identity version is associated with the old one, and the change time is recorded. This step aims to ensure that after a device change, an independent identity can be established for the new physical entity or configuration, while preserving the association with historical entities to maintain a complete view of the device lifecycle. One implementation is to maintain a version control table for each target object in the database. When a topology change event is detected, the system generates a new entity ID or version number and establishes a parent-child relationship or chain association with the old entity ID. Simultaneously, the timestamp of the change is accurately recorded, for example, "2023-10-26 10:30:00". Another implementation is to use blockchain technology to generate a unique hash value for each entity identity version and record the association between the old and new versions and the change time as a transaction record on the blockchain, ensuring the immutability and traceability of identity version changes.

[0065] Based on the timestamp and entity identity version record of the processed monitoring information, the system determines whether the processed monitoring information should be associated with the current entity identity version or its historical entity identity version, and then associates the processed monitoring information with the target object corresponding to the determined entity identity version. This step aims to ensure that each piece of monitoring information can be accurately attributed to the physical entity version it describes that existed at a specific point in time, avoiding state analysis errors caused by information mismatch. One implementation is that when the system receives the processed monitoring information, it first extracts its timestamp. Then, it queries the entity identity version record of the target object and compares it with the change time of each version record. If the timestamp of the monitoring information is earlier than the creation time of a certain version, it is associated with a historical version before that version; if it is later than the creation time of a certain version, it is associated with that version or its subsequent version. Another implementation is to construct a timeline index and mark the valid interval of each entity identity version on the timeline. When monitoring information arrives, the system directly looks up the corresponding valid entity identity version in the timeline index based on its timestamp and associates it.

[0066] If the target object is changed, the information sequence of the old entity identity version is logically integrated with the information sequence of the new entity identity version to construct a continuous information sequence of the target object in the time dimension. This step aims to ensure that even if the physical equipment is replaced, its historical operating state and the operating state of the new equipment can be analyzed as a continuous whole, thereby supporting the complete tracking of the gradual degradation process throughout the equipment's lifecycle. One implementation is to stitch the information sequences of the old and new entity identity versions in the time dimension at the database level by creating a view or using a union query. Another implementation is to automatically load and merge the information sequences of all relevant versions at the data processing layer when performing state analysis, based on the relationships between entity identity versions; for example, by querying the data of all relevant versions using a unified logical identifier and sorting them by timestamp, a continuous information sequence reflecting the long-term state evolution of the insulator at that location can be formed.

[0067] This method proactively identifies power system topology change events, such as equipment replacement records, line modification records, or temporary bypass operation records, enabling the system to actively detect changes in the physical configuration or identity of power equipment or lines. Once such a change is identified, the system creates a new entity identity version for the affected target object, establishes a clear association with the old entity identity version, and accurately records the time of the change. This version management mechanism ensures that even if the physical entity changes, its logical identity within the system can be continued and traced. Subsequently, when processing subsequent monitoring information, the system intelligently determines whether the processed monitoring information should be associated with the current entity identity version or its historical entity identity version, based on the timestamp of the monitoring information itself and the recorded entity identity version information. This determination mechanism is crucial for ensuring the accuracy of information association; it avoids incorrectly attributing information belonging to old equipment to new equipment, and vice versa, thus guaranteeing the purity and accuracy of the information sequence for each entity identity version. Furthermore, when a target object is replaced, this method logically integrates the information sequence of the old entity identity version with the information sequence of the new entity identity version. This integration is not a simple data patchwork, but rather a logical approach that treats information sequences from different physical lifecycles as a continuous whole. In this way, the system can construct a continuous information sequence of the target object over time, seamlessly tracking its complete historical state evolution even after multiple equipment replacements. This series of steps works synergistically to solve the problems of information sequence breaks and discontinuous state analysis encountered by traditional methods during power system topology changes. By introducing entity identity version management and logical integration of information sequences, this method can maintain a continuous, accurate, and traceable information sequence for dynamically changing power equipment. This enables NLP-based intelligent line loss analysis methods to more accurately identify progressive failure modes, as early signs of failure may span equipment replacement cycles, and a continuous information sequence provides the necessary data foundation for capturing these long-term evolution trends. For example, the insulation degradation process of a device may begin before replacement and continue to develop on the new equipment. Through a continuous information sequence, the system can connect these scattered signs of degradation, thus issuing early warnings at the nascent stage of a failure, achieving a shift from "post-event tracing" to "pre-event prediction."

[0068] In some embodiments, step S2, which involves associating the processed monitoring information with the target object to construct an information sequence of the target object over time, includes: S21C1. Establish a hierarchical topology for power equipment; the hierarchical topology clarifies the inclusion relationships between equipment at each level; S21C3. Identify the hierarchical granularity to which the device entity described by the processed monitoring information belongs; S21C4. Based on the hierarchical granularity of the equipment entity, cross-level information influence transmission is performed to associate the processed monitoring information with the target object and construct the information sequence of the target object in the time dimension; the specific steps of cross-level information influence transmission include: S21C41. If the processed monitoring information describes a low-level device, the impact of the processed monitoring information on the status of the low-level device is accumulated to the status information sequence of the direct superior device of the low-level device, and further accumulated upwards to the status information sequence of higher-level devices. S21C42. If the processed monitoring information describes a high-level device, then the impact of the processed monitoring information on the status of the high-level device is decomposed and transmitted to the status information sequence of the direct subordinate devices of the high-level device, and further transmitted down to the status information sequence of the lower-level devices. S21C5. Maintain an independent but interconnected information sequence for each level of device entity; the information sequence reflects its own state changes and synchronously reflects the information influence from other levels.

[0069] First, establish a hierarchical topology for power equipment. A hierarchical topology for power equipment refers to organizing various devices in a power system, such as power plants, substations, transmission lines, distribution lines, poles, transformers, switches, and insulators, into a tree-like or network structure with clear hierarchical relationships, based on their function, geographical location, or electrical connections. This structure clearly defines the inclusion, composition, or connection relationships between different equipment entities. For example, a substation may contain multiple main transformers and outgoing line bays, and an outgoing line bay may contain circuit breakers, disconnect switches, and other equipment, while a transmission line consists of multiple poles and conductors. This hierarchical topology can be constructed based on Geographic Information System (GIS) data and the primary wiring diagram of the power system, through manual configuration or automated parsing tools, storing information such as equipment ID, equipment type, parent equipment ID, and child equipment ID, forming a hierarchical database. Alternatively, ontology modeling techniques can be used to define the concepts, attributes, and hierarchical relationships of power equipment, thereby constructing a semantic knowledge graph that dynamically reflects the hierarchical topology of the equipment.

[0070] Based on this, the hierarchical granularity of the equipment entities described by the processed monitoring information is identified. Identifying hierarchical granularity refers to determining the specific location or abstract level of the equipment entity described by the standardized monitoring information within a pre-established hierarchical topology. For example, a monitoring message might describe a specific insulator (lowest level), a power line (middle level), or even a substation (high level). Accurately identifying its granularity is fundamental for subsequent cross-level information transmission. This can be achieved by utilizing Named Entity Recognition (NER) technology, combined with a power equipment dictionary and hierarchical topology, to extract equipment names from the monitoring information and determine their level based on the definition of the equipment name within the topology. For example, identifying "10kV Line A, Tower 45 Insulator" would classify its granularity as the "insulator" level. Alternatively, rule matching and context analysis can be used. Pre-defined rules can be used to initially determine the granularity based on keywords appearing in the monitoring information (such as "power line," "substation," "tower," and "insulator"), and then combined with equipment IDs or geographical location information for precise matching within the topology to determine its level.

[0071] Subsequently, based on the hierarchical granularity of the device entity, cross-level information impact propagation is performed to associate the processed monitoring information with the target object and construct the information sequence of the target object in the time dimension. Cross-level information impact propagation refers to, after identifying the hierarchical granularity of the device entity described by the monitoring information, appropriately aggregating or decomposing the impact of this information on the device state within the hierarchical topology of the power equipment according to preset rules and logic. This ensures that the information can be correctly associated with all affected target objects and that their respective information sequences in the time dimension are updated. This process is crucial for achieving multi-granularity information integration. This propagation can employ an event-based propagation mechanism. When the state of a lower-level device changes, the system generates an event that propagates up the topology to higher-level devices and updates the state of the higher-level devices according to preset aggregation rules. Conversely, when the state of a higher-level device changes, the event propagates down and updates the state of the lower-level devices according to decomposition rules. Alternatively, an influence matrix or weight model can be used to define influence coefficients or weights for information transmission between different levels of devices. When information is generated at a certain device level, its influence will be weighted and transmitted between the upper and lower levels according to these coefficients to reflect its contribution or degree of influence on the status of different levels of devices.

[0072] Specifically, if the processed monitoring information describes lower-level equipment, the impact of the processed monitoring information on the state of the lower-level equipment is accumulated into the state information sequence of the equipment directly above the lower-level equipment, and further accumulated upwards to the state information sequence of higher-level equipment. For example, when the monitoring information describes a specific component or sub-equipment in the power system (e.g., an insulator on a tower, a winding inside a transformer), its state changes not only affect itself but also affect the higher-level equipment containing it. This step aims to aggregate the state information of these lower-level equipment according to the hierarchical topology, reflecting it in the overall state information sequence of its directly above-level equipment and even higher-level equipment (such as lines and substations). This can be achieved through weighted summation or logical aggregation. For example, if insulators on multiple towers show slight deterioration, this deterioration information can be accumulated to the line level to which it belongs, and reflected by calculating the overall insulation risk index of the line. Specifically, rules can be set, such as when the risk increment value of a lower-level equipment reaches a certain threshold, the risk increment value of its higher-level equipment also increases accordingly.

[0073] Conversely, if the processed monitoring information describes high-level equipment, the impact of the processed monitoring information on the state of high-level equipment is decomposed and transmitted to the state information sequence of the directly subordinate equipment of the high-level equipment, and further transmitted down to the state information sequence of lower-level equipment. For example, when the monitoring information describes a relatively macroscopic equipment or area in the power system (e.g., the overall operating status of a transmission line, or an environmental anomaly in a substation), its impact often affects all the subordinate equipment it contains. This step aims to decompose the state information of these high-level equipment according to the hierarchical topology, level by level, and transmit it to the respective state information sequences of their directly subordinate equipment and even lower-level equipment (such as poles and towers on the line, switches in the substation). This can be achieved through influence factor allocation. For example, if the monitoring information indicates that "there is a risk of flashover on a certain line as a whole," then this risk can be decomposed to all poles and insulators on that line according to a preset allocation ratio or rule, updating their respective flashover risk state information sequences.

[0074] Ultimately, an independent but interconnected information sequence is maintained for each device entity at each level. These information sequences reflect its own state changes and synchronously reflect the impact of information from other levels. This step aims to establish and maintain an independent time-series data structure for each device entity in the power system, regardless of its level. This information sequence not only records the device's own state changes (e.g., monitoring reports directly targeting the device) but also synchronously includes the impact of state changes from its superior or subordinate devices through a cross-level information impact transmission mechanism. This design ensures that each device's information sequence is both independent, reflecting its specific condition, and interconnected, reflecting its interaction and overall environmental impact within the power system. This can be achieved using a distributed database or graph database, where each device entity has an independent record containing a time-series field to store data such as its state score and risk increment that change over time. Hierarchical relationships between devices are represented by parent-child node relationships or edges, and information transmission updates the information sequences of related devices by querying these relationships.

[0075] This solution, combined with the aforementioned steps of acquiring monitoring information and standardization, forms an organic whole. After acquiring unstructured text data and standardizing it, this solution can accurately associate this processed information with specific equipment entities in the power system topology, and intelligently aggregate or decompose information impacts across different levels according to their hierarchical granularity. This ensures that the information sequence of each target object includes relevant information about itself and its superior and subordinate devices, thus constructing a more comprehensive, continuous, and dynamic information sequence. This multi-granular, hierarchical information sequence construction method greatly enhances the system's ability to perceive the progressive degradation process of power equipment. It enables the system to identify potential risks that accumulate over long periods across different levels from scattered monitoring information of varying granularities, providing a solid data foundation for subsequent risk increment calculations, state score updates, and progressive fault mode matching, thereby effectively addressing the limitations of traditional methods in processing complex hierarchical information.

[0076] The following is a concrete example to illustrate this. Assume a power system includes a substation with multiple 10kV outgoing lines, each consisting of several towers, and each tower is equipped with insulators. The system first establishes the following hierarchical topology: the substation is the highest level, the 10kV outgoing lines are the middle level, the towers are the next lowest level, and the insulators are the lowest level. When the system receives a processed monitoring message, such as "the surface of the insulator on tower 45 of 10kV Line A is contaminated," the system first identifies the device entity described in the message as an "insulator," belonging to the lowest level. At this point, according to the cross-level information impact transmission mechanism, the impact of the insulator contamination on its state (e.g., the insulation risk increment) not only updates the insulator's own information sequence but also accumulates upwards. Specifically, this impact accumulates in the state information sequence of "tower 45 of 10kV Line A," reflecting the overall insulation status of that tower. Furthermore, this impact accumulates upwards to the status information sequence of the "10kV Line A," reflecting the overall insulation risk of that line, and may even affect the overall operational risk assessment at the "substation" level. Conversely, if the system receives a monitoring message, such as "the air humidity in a certain substation area has been consistently high recently," the system will identify the equipment entity described in the message as a "substation," belonging to the highest level of hierarchy. At this point, the impact of this information on the overall environmental status of the substation (e.g., the incremental value of environmental exposure risk) will be decomposed and propagated downwards. Specifically, this impact will propagate to the information sequence of every 10kV outgoing line under that substation, indicating that these lines may face higher environmental risks. Further, this impact will also propagate to all towers and insulators on each outgoing line, updating their respective environmental exposure status information sequences, thereby reminding maintenance personnel to pay attention to moisture-proof and pollution-proof measures for all relevant equipment. In this way, whether it's a microscopic problem targeting a single insulator or a macroscopic environmental change affecting the entire substation, the impact can be reasonably propagated and reflected throughout the hierarchical topology. Each insulator, each tower, each line, and each substation maintains an independent but interconnected sequence of information. These sequences not only record events directly related to themselves but also include influences from higher and lower levels, thus constructing a comprehensive, dynamic, and hierarchically related view of the equipment status.

[0077] Through the above technical solution, this application effectively addresses the problem in intelligent analysis of power system line losses: when facing the complex hierarchical structure of power system equipment, and when monitoring information describes a phenomenon at a certain level or a specific sub-component, the system cannot effectively aggregate or decompose information of different granularities when constructing the information sequence of the target object in the time dimension. This results in a lack of detail in high-level information sequences, while low-level information sequences fail to fully reflect the overall environmental impact, thus hindering a comprehensive understanding of the multi-level progressive degradation process. Specifically, by establishing a hierarchical topology of power equipment, a clear structured foundation is provided for information transmission, avoiding the isolation of information in complex hierarchies. Identifying the hierarchical granularity of the monitoring information ensures that the information can be accurately located. More importantly, by designing a cross-level information influence transmission mechanism, bidirectional information flow between levels is achieved: the state influence of lower-level equipment can accumulate upwards, allowing small local changes to be perceived by the higher-level overall system; the state influence of higher-level equipment can be decomposed downwards, allowing macro-environmental changes to be refined to specific components. This mechanism ensures that each level of equipment entity maintains an independent and interconnected information sequence that reflects both its own state changes and the influence of information from other levels. This hierarchical information sequence construction method, combined with the aforementioned steps of acquiring and standardizing unstructured monitoring information, enables the system to organically integrate previously fragmented, multi-granular textual information into the hierarchical relationships of power system equipment. It not only captures signs of degradation in individual devices but also correlates these signs with environmental factors and operating conditions at higher and lower levels, constructing a complete, continuous, and dynamic health profile of the equipment from the component level to the system level. This significantly enhances the system's ability to identify multi-level progressive failure modes, allowing potential line loss risks to be detected in their early stages, thus achieving a shift from passive response to proactive prediction and significantly improving the reliability and economy of power system operation.

[0078] In some embodiments, the specific steps in step S3 include: S31. Analyze the feature items in the information sequence and identify the contextual attributes of the feature items; S32. If multiple contextual attributes are identified for the same feature item, and there are contradictions or ambiguities among the contextual attributes, then the following steps are performed: S321. Obtain the report sources corresponding to multiple contextual attributes, and determine the authority level of each report source; S322. Select the contextual attribute based on the level of authority; S323. If multiple contextual attributes have the same level of authority, the contextual attribute shall be selected based on the timestamp of the report corresponding to the contextual attribute. S324. If multiple contextual attributes still exist with the same highest authority level and latest timestamp, then select the contextual attribute based on the risk level indicated by the contextual attribute; S325. The selected contextual attributes are used as the final contextual attributes of the feature terms; S33. Based on the final contextual attributes of the feature items and the preset degradation knowledge rules, calculate the risk increment value of the target object under different evaluation dimensions.

[0079] This method involves parsing feature items in an information sequence and identifying the contextual attributes of these feature items. The information sequence is a set of monitoring information about the target object over time. Feature items are keywords, phrases, or entities extracted from this unstructured text data that reflect the state or potential risks of the target object. For example, in a power system, feature items might include equipment names such as "insulator," "cable," and "transformer," as well as words describing equipment status or environmental conditions such as "crack," "water accumulation," "abnormal noise," and "abnormal temperature." The parsing process can employ natural language processing techniques, such as named entity recognition models, keyword extraction algorithms, or rule-based pattern matching, to automatically identify and extract these feature items from the text. Contextual attributes refer to the contextual information surrounding the feature item, which further defines its meaning, intensity, urgency, or scope of impact. For example, for the feature item "crack," its contextual attributes might include modifiers describing the degree or location of the crack, such as "minor," "serious," "surface," or "internal," or report source information such as "inspection report" or "user complaint." Identifying contextual attributes can be achieved through methods such as dependency parsing, semantic role labeling, sentiment analysis, or rule-based context window analysis to capture the semantic relationship between feature items and their surrounding words.

[0080] If multiple contextual attributes are identified for the same feature item and there are contradictions or ambiguities among these contextual attributes, that is, for the same feature item, conflicting or unclear contextual descriptions are identified in different monitoring information or reports, such as one report describing "minor cracks in the insulator" while another report describes "severe damage to the insulator", the system will perform a series of conflict resolution steps.

[0081] First, the system acquires the report sources corresponding to the multiple contextual attributes and determines the authority level of each report source. A report source refers to the specific channel or entity providing monitoring information, such as automated monitoring systems, inspection reports from professional inspectors, and user complaint records. The authority level is a quantitative assessment of the credibility or importance of different report sources. For example, data from automated monitoring systems is typically assigned a higher authority level, while user complaints may have a lower authority level. The determination of the authority level can be based on preset expert experience rules, historical data verification, or by evaluating the accuracy of reports from different sources using machine learning models.

[0082] Subsequently, the system selects the contextual attribute based on the authority level. When contradictory or ambiguous contextual attributes exist, the system prioritizes the contextual attribute provided by the reporting source with the higher authority level. For example, if the automated monitoring system reports "device temperature is normal," while a regular user reports "device is overheating," the system will select "device temperature is normal" as the contextual attribute for that feature, based on the higher authority level of the automated monitoring system.

[0083] If multiple contextual attributes have the same authority level, the system selects the appropriate contextual attribute based on the timestamp of the corresponding report. A timestamp refers to the precise time when the monitoring information or report was generated or recorded. When multiple contextual attributes have the same highest authority level, the system selects the contextual attribute with the most recent timestamp, as the most recent information usually better reflects the current true state of the target object.

[0084] If multiple contextual attributes still possess the same highest authority level and latest timestamp, the system selects the appropriate attribute based on the risk level indicated by that attribute. Risk level refers to the degree of impact a contextual attribute implies on the state or operational security of the target object. In this case, the system adopts a "strict and high" principle, selecting the contextual attribute indicating a higher risk level to ensure sufficient attention to potential risks and timely response. Risk level assessment can be based on a pre-defined risk level dictionary, an expert knowledge base, or by classifying text for risk using a machine learning model.

[0085] Ultimately, the system selects the contextual attribute as the final contextual attribute of the feature item. This final contextual attribute is the unique and most reliable contextual description determined for a specific feature item after the above conflict resolution and priority judgment.

[0086] Finally, the system calculates the risk increment value of the target object under different evaluation dimensions based on the final contextual attributes of the aforementioned feature items and the preset degradation knowledge rules. The degradation knowledge rules are a predefined series of logical rules or models used to describe how specific feature items and their contextual attributes affect the degradation trend or risk changes of the target object under different evaluation dimensions. The risk increment value is a numerical value that quantifies the degree of influence of a specific event or information on the state score of the target object in each evaluation dimension. It is a dynamic and cumulative indicator used to reflect the deterioration or improvement of the target object's state.

[0087] This application's solution, after parsing feature items in an information sequence and identifying their contextual attributes, introduces an intelligent conflict resolution mechanism. This effectively addresses the technical problem in power system line loss analysis where the contextual attributes of the same feature item appearing in different reports or at different times can be contradictory or ambiguous, leading to the system's inability to accurately determine the true risk meaning when parsing the feature item's contextual attributes, thus affecting the accuracy of risk increment calculation. Specifically, by acquiring multiple report sources corresponding to contextual attributes and determining their authority levels, the system prioritizes more reliable information sources, avoiding interference from low-reliability information in risk assessment. When authority levels are the same, the latest information is selected based on the report's timestamp, ensuring the real-time nature and effectiveness of risk assessment. When both authority levels and timestamps are indistinguishable, the system adopts a "strict and high" principle, selecting the contextual attribute indicating a higher risk level, thus avoiding underestimation of potential risks and improving the timeliness and accuracy of risk warnings. This multi-layered, intelligent contextual attribute conflict resolution mechanism ensures that the final contextual attribute determined for a feature item is unique, accurate, and reliable. Based on this, the calculation of risk increment values, combined with pre-defined degradation knowledge rules, significantly improves the accuracy of risk increment value calculation. This makes the subsequent cumulative updates of the state scores for each evaluation dimension of the target object more accurate, thus more realistically reflecting the dynamic state of the target object. Combined with the steps of acquiring monitoring information for the target object, standardizing the monitoring information, and constructing an information sequence, this solution enables the system to not only identify key degradation characteristics from massive amounts of unstructured monitoring information, but also intelligently handle the contradictions and ambiguities of these characteristics in complex contexts, thereby providing a more solid and reliable data foundation for the dynamic state assessment of the target object. This is of great significance for accurately identifying the gradual degradation trend of power equipment, achieving pre-emptive prediction and proactive maintenance, effectively reducing line loss risk, and improving the operational reliability and economy of the power system.

[0088] The following is a concrete example to illustrate this. Suppose the target of a power line is "the insulator of tower No. 45 on the 10kV A line". Over a period of time, the system receives multiple monitoring messages about this insulator, which have been standardized and processed to construct an information sequence. When parsing the feature items in the information sequence, the system identifies the feature item as "insulator crack". However, for this feature item, the system identifies multiple contextual attributes, and there are contradictions or ambiguities among these contextual attributes. For example: Report A: Inspector Zhang San reported via mobile terminal on Monday morning: "There is a slight crack in the insulator of pole number 45 on the 10kV A line." Report B: The automated monitoring system recorded at noon on Monday: "No abnormalities on the surface of the insulator of pole number 45 on the 10kV A line." Report C: Community electrician Li Si complained by phone on Monday morning: "A user near pole number 45 on the 10kV A line reported unstable voltage and suspected a problem with the insulator." At this point, the system identifies three contradictory or ambiguous contextual attributes for the feature "insulator crack": "slight crack" (from Report A), "no abnormalities" (from Report B), and "problem" (from Report C). The system will perform the following steps to resolve the contradictions: First, the system obtains the report source corresponding to these contextual attributes and determines its authority level. For example, the automated monitoring system (Report B) is assigned the highest authority level (e.g., level 5), the professional inspector (Report A) is assigned the second highest authority level (e.g., level 4), and the community electrician (Report C) is assigned a lower authority level (e.g., level 2). Based on the level of authority, the system will prioritize selecting the contextual attribute "no anomaly" for report B, because the automated monitoring system has the highest level of authority.

[0089] However, if the situation is as follows: Report A: Inspector Zhang San reports on Monday morning: "The insulator on tower No. 45 of the 10kV A line has a slight crack." (Authority level 4) Report D: Another professional inspector, Wang Wu, reports on Monday morning: "The insulator on tower No. 45 of the 10kV A line has obvious wear." (Authority level 4) In this case, Report A and Report D have the same authority level. The system will further select based on the timestamp of the report corresponding to the context attribute. Assuming that the timestamps of both reports are Monday morning at 10:00, then the timestamps are also the same. In this case, the system will select based on the risk level indicated by the context attribute. For example, "slight crack" may be assessed as medium risk, and "obvious wear" may be assessed as high risk. The system will select "obvious wear," which indicates a higher risk level, as the final context attribute. Through the above-mentioned layers of screening, the system finally determines "obvious wear" as the final context attribute for the feature item "insulator crack." Subsequently, based on this final contextual attribute and preset degradation knowledge rules, such as "significant wear of the insulator will lead to insulation degradation," the system will calculate the risk increment value of "the insulator of tower No. 45 of the 10kV A line" under the "insulation condition" evaluation dimension, for example, +0.7 points. This risk increment value will be used to cumulatively update the dynamic state of the insulator.

[0090] This application's solution, after parsing feature items in an information sequence and identifying their contextual attributes, introduces an intelligent conflict resolution mechanism. This effectively addresses the technical problem in power system line loss analysis where the contextual attributes of the same feature item appearing in different reports or at different times can be contradictory or ambiguous, leading to the system's inability to accurately determine the true risk meaning when parsing the feature item's contextual attributes, thus affecting the accuracy of risk increment calculation. Specifically, by acquiring multiple report sources corresponding to contextual attributes and determining their authority levels, the system prioritizes more reliable information sources, avoiding interference from low-reliability information in risk assessment. When authority levels are the same, the latest information is selected based on the report's timestamp, ensuring the real-time nature and effectiveness of risk assessment. When both authority levels and timestamps are indistinguishable, the system adopts a "strict and high" principle, selecting the contextual attribute indicating a higher risk level, thus avoiding underestimation of potential risks and improving the timeliness and accuracy of risk warnings. This multi-layered, intelligent contextual attribute conflict resolution mechanism ensures that the final contextual attribute determined for a feature item is unique, accurate, and reliable. Based on this, the calculation of risk increment values, combined with pre-defined degradation knowledge rules, greatly improves the accuracy of risk increment value calculation. This makes the subsequent cumulative updates of the status scores of each evaluation dimension of the target object more accurate, thus reflecting the dynamic status of the target object more realistically. This is of great significance for accurately identifying the gradual degradation trend of power equipment, realizing pre-prediction and proactive maintenance, effectively reducing line loss risk, and improving the operational reliability and economy of the power system.

[0091] In some embodiments, the specific steps in step S4 include: S41. Determine the impact weight corresponding to the risk increment value; the impact weight reflects the degree of influence of the risk increment value on the state scores of each evaluation dimension of the target object; S42. Determine the time decay factor corresponding to the risk increment value; the time decay factor reflects the degree to which the impact of the risk increment value on the state scores of each evaluation dimension of the target object weakens over time; S43. Based on the influence weight and time decay factor, the risk increment value is weighted and decayed to obtain the adjusted risk increment value; S44. The adjusted risk increment value is added to the state score of each evaluation dimension of the target object to achieve cumulative update of the state score and obtain the dynamic state of the target object.

[0092] This involves determining the impact weight corresponding to the risk increment value. This impact weight is a coefficient used to quantify the degree of influence of a specific risk event on the status scores of different evaluation dimensions of the target object. Its function is to distinguish the severity of different risk events and their differentiated impacts on the health status of equipment. Methods for determining the impact weight can include: based on expert experience, having domain experts assess various risk events and assign them corresponding weight values; or based on historical data analysis, statistically analyzing historical fault data and equipment status change data, and using machine learning models (such as regression analysis, decision trees, etc.) to learn the correlation between risk events and status score changes, thereby automatically generating or optimizing the weights.

[0093] Simultaneously, a time decay factor corresponding to the risk increment value is determined. This time decay factor is a coefficient used to describe how the impact of a risk event on the target object's state score naturally weakens over time. Its purpose is to ensure that older risk events, which no longer have an immediate impact, do not continue to excessively affect the current equipment state assessment, thus reflecting the timeliness of the risk. Various models can be used to determine the time decay factor, such as: an exponential decay model, where the risk impact decreases exponentially over time; a linear decay model, where the risk impact decreases linearly over time; or a piecewise function model, where different decay rates are used in different time periods.

[0094] Subsequently, based on the determined impact weights and time decay factors, the original risk increment value is weighted and decayed to obtain a refined risk increment value. This step aims to comprehensively consider the inherent impact and timeliness of the risk event, revising the original risk increment value to obtain a risk assessment value that better reflects the actual situation. Its function is to synergistically apply the impact weights and time decay factors to the risk increment value, ensuring that the final risk value used to update the state score reflects both the severity of the event and its dynamic changes over time. Specifically, this can be achieved through mathematical calculations, such as multiplying the original risk increment value by the impact weights and time decay factors: Adjusted risk increment value = Original risk increment value × Impact weight × Time decay factor.

[0095] Finally, the adjusted risk increment value is added to the status scores of each evaluation dimension of the target object to achieve cumulative updates of the status scores, thereby obtaining a more accurate and reliable dynamic status of the target object. The purpose of this step is to integrate the refined risk increment value into the health status assessment system of the target object, achieving dynamic and cumulative updates of the status scores. Through this accumulation mechanism, the status scores of each evaluation dimension of the target object can reflect the comprehensive impact of various risk events in real time, thus forming a continuously evolving dynamic status. Specifically, this is achieved through direct numerical accumulation: new status score = old status score + adjusted risk increment value.

[0096] This scheme refines the risk increment value by introducing influence weights and time decay factors, enabling the original risk increment value calculated from the feature items in the parsed information sequence to more accurately reflect its impact on the target object's state score. This processing mechanism is closely integrated with the process of extracting risk information from unstructured text, ensuring that the risk information extracted from the text, when converted into a quantified state score, considers not only the occurrence of the risk but also the depth and duration of its impact. In this way, the dynamic state of the target object can more realistically reflect the equipment's health status, providing a more solid and reliable data foundation for subsequent monitoring of the state score evolution sequence and matching progressive failure modes, thereby significantly improving the accuracy and early warning capability of the entire intelligent line loss analysis method.

[0097] In some embodiments, the specific steps in step S5 include: S51. Establish a dynamic reference baseline for the state scores of each evaluation dimension; the dynamic reference baseline is obtained by smoothing the state scores within a preset time window. S52. Monitor the deviation between the status score and the dynamic reference baseline; S53. When the duration of the deviation between the state score and the dynamic reference baseline reaches a preset duration and the deviation exceeds a preset threshold, it is judged as a deterioration trend; S54. If the duration of the deviation does not reach the preset duration, or the deviation magnitude does not exceed the preset threshold, it is judged as a non-fault fluctuation.

[0098] The establishment of a dynamic reference baseline for the status scores of each evaluation dimension aims to provide a benchmark that adjusts dynamically over time to reflect the fluctuation range of the status scores of the target object under normal operating conditions. This dynamic reference baseline can be obtained through various smoothing methods. For example, a moving average method can be used, which calculates the average of the status scores over the past N time points as the dynamic reference baseline for the current moment; or an exponentially weighted moving average method can be used, which assigns higher weight to recent status scores to make them more sensitive to the latest trends.

[0099] Monitoring the deviation between the state score and the dynamic reference baseline refers to continuously tracking the difference between the actual state score and the established dynamic reference baseline. This monitoring can be achieved by calculating the absolute or relative difference between the state score and the dynamic reference baseline, for example, by calculating the percentage deviation between the current state score and the baseline. Alternatively, statistical methods, such as calculating the Z-score, can be used to quantify the standard deviation of the state score from the baseline, thereby more accurately assessing the significance of the deviation.

[0100] A degradation trend is identified when the duration of the deviation between the state score and the dynamic reference baseline reaches a preset duration and the deviation magnitude exceeds a preset threshold. This judgment mechanism comprehensively considers both the persistence and significance of the deviation. The duration of the deviation can be determined by maintaining a counter that records the duration or number of consecutive deviations of the state score from the baseline, triggering a check when a preset value is reached. The magnitude of the deviation can be determined by setting an absolute or relative threshold, for example, a state score below the baseline by X points or Y%. Only when both conditions are met simultaneously is a genuine degradation trend identified, effectively avoiding false alarms caused by short-term fluctuations.

[0101] If the duration of the deviation does not reach the preset duration, or the magnitude of the deviation does not exceed the preset threshold, it is judged as a non-fault-related fluctuation. This step clarifies the logic for distinguishing between normal fluctuations and fault-related degradation. This means that even if the state score deviates to a certain extent, if this deviation is brief, unsustainable, or its magnitude is insufficient to reach the preset risk threshold, it is classified as a non-fault-related fluctuation, such as normal fluctuations caused by environmental factors or measurement errors, rather than degradation of the equipment itself.

[0102] This application's solution introduces a dynamic reference baseline and a judgment mechanism based on deviation duration and magnitude to perform refined monitoring and analysis of the evolution sequence of the state scores of each evaluation dimension of the target object over time. First, the establishment of the dynamic reference baseline effectively filters short-term noise and random fluctuations by smoothing the state scores within a preset time window, thus creating a baseline that reflects the normal operating state of the target object. Based on this, the system continuously monitors the deviation between the actual state score and this dynamic reference baseline, achieving real-time tracking of state changes. Only when the deviation between the state score and the dynamic reference baseline not only lasts for a certain preset duration but also exceeds a preset threshold is the system judged as a genuine deterioration trend. Conversely, if the deviation duration is insufficient or the deviation magnitude is not significant enough, it is identified as non-faulty fluctuations. This mechanism ensures that only those continuous and significant abnormal changes are identified as potential deterioration risks, thereby avoiding misjudging normal fluctuations as deterioration trends and significantly improving the accuracy and reliability of risk identification. This approach, combined with the aforementioned NLP-based intelligent line loss analysis method, makes the monitoring and interpretation of the dynamic state evolution sequence in step S5 more accurate. It provides high-quality input for matching the evolution sequence with the preset progressive failure mode in the subsequent step S6, thereby enabling more effective identification of progressive failure modes and outputting more accurate risk warning information.

[0103] Reference Appendix Figure 2 This invention provides an NLP-based intelligent line loss analysis system (this NLP-based intelligent line loss analysis system adopts the NLP-based intelligent line loss analysis method of the above embodiments, and the specific process is described in the corresponding steps above), including: The acquisition module 100 is used to acquire monitoring information for the target object; Module 200 is used to standardize the monitoring information and associate the processed monitoring information with the target object to construct the information sequence of the target object in the time dimension. The calculation module 300 is used to parse the feature items in the information sequence and calculate the risk increment value of the target object under different evaluation dimensions based on the contextual attributes corresponding to the feature items and the preset degradation knowledge rules. The update module 400 is used to cumulatively update the state scores of each evaluation dimension of the target object based on the risk increment value, so as to obtain the dynamic state of the target object. The monitoring module 500 is used to monitor the evolution sequence of the state scores of each evaluation dimension in a dynamic state over time. The matching module 600 is used to match the evolution sequence with preset progressive failure modes and output risk warning information for the target object when the match is successful.

[0104] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0105] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A line loss intelligent analysis method based on NLP, characterized in that, Includes the following steps: S1. Obtain monitoring information for the target object; S2. Standardize the monitoring information and associate the processed monitoring information with the target object to construct an information sequence of the target object in the time dimension; S3. Analyze the feature items in the information sequence, and calculate the risk increment value of the target object under different evaluation dimensions based on the contextual attributes corresponding to the feature items and the preset degradation knowledge rules; S4. Based on the risk increment value, the state scores of each evaluation dimension of the target object are cumulatively updated to obtain the dynamic state of the target object; S5. Monitor the evolution sequence of the state scores of each evaluation dimension in the dynamic state over time; S6. Match the evolution sequence with a preset progressive failure mode, and output risk warning information for the target object when the match is successful.

2. The NLP-based intelligent line loss analysis method according to claim 1, characterized in that, The monitoring information includes unstructured text data describing the operational status of the target object.

3. The NLP-based intelligent line loss analysis method according to claim 2, characterized in that, Step S2, the step of standardizing the monitoring information, includes: S21A1. Perform text cleaning, typo correction, and industry terminology standardization on the monitoring information; wherein, when performing industry terminology standardization, identify intensity indicator words in the monitoring information that describe the equipment status or the degree of risk urgency; S21A2. Based on the preset intensity level mapping rules, the intensity indicator words are quantified into intensity values; S21A3. The intensity value is attached as an attribute to the standardized industry terminology to obtain the standardized industry terminology for the attached intensity value, and used as the processed monitoring information.

4. The NLP-based intelligent line loss analysis method according to claim 1, characterized in that, Step S2, which involves associating the processed monitoring information with the target object to construct an information sequence of the target object in the time dimension, includes: S21B1. Identify power system topology change event information; S21B2. When the topology change event information is detected, a new entity identity version is created for the affected target object; the new entity identity version is associated with the old entity identity version, and the change time is recorded; S21B3. Based on the timestamp of the processed monitoring information and the entity identity version record, determine whether the processed monitoring information should be associated with the current entity identity version or its historical entity identity version, and associate the processed monitoring information with the target object corresponding to the determined entity identity version; S21B4. If the target object is changed, the information sequence of the old entity identity version and the information sequence of the new entity identity version are logically integrated to construct a continuous information sequence of the target object in the time dimension.

5. The NLP-based intelligent line loss analysis method according to claim 4, characterized in that, The topology change event information includes equipment replacement records, line modification records, or temporary bypass operation records.

6. The NLP-based intelligent line loss analysis method according to claim 1, characterized in that, Step S2, which involves associating the processed monitoring information with the target object to construct an information sequence of the target object in the time dimension, includes: S21C1. Establish a hierarchical topology for power equipment; the hierarchical topology clearly defines the inclusion relationships between equipment at each level; S21C3. Identify the hierarchical granularity to which the device entity described by the processed monitoring information belongs; S21C4. Based on the hierarchical granularity of the device entity, perform cross-level information influence transmission to associate the processed monitoring information with the target object and construct the information sequence of the target object in the time dimension; the specific steps of the cross-level information influence transmission include: S21C41. If the processed monitoring information describes a low-level device, the impact of the processed monitoring information on the state of the low-level device is accumulated to the state information sequence of the direct superior device of the low-level device, and further accumulated upwards to the state information sequence of higher-level devices. S21C42. If the processed monitoring information describes a high-level device, then the impact of the processed monitoring information on the state of the high-level device is decomposed and transmitted to the state information sequence of the direct subordinate devices of the high-level device, and further transmitted down to the state information sequence of lower-level devices. S21C5. Maintain an independent but interconnected information sequence for each level of device entity; the information sequence reflects its own state changes and synchronously reflects the information influence from other levels.

7. The NLP-based intelligent line loss analysis method according to claim 1, characterized in that, The specific steps in step S3 include: S31. Analyze the feature items in the information sequence and identify the contextual attributes of the feature items; S32. If multiple contextual attributes are identified for the same feature item, and the contextual attributes are contradictory or ambiguous, then the following steps are performed: S321. Obtain the report sources corresponding to the multiple contextual attributes, and determine the authority level corresponding to each report source; S322. Select the contextual attribute based on the authority level; S323. If multiple context attributes have the same level of authority, the context attribute is selected based on the timestamp of the report corresponding to the context attribute. S324. If multiple contextual attributes still exist with the same highest authority level and latest timestamp, then select the contextual attribute according to the risk level indicated by the contextual attribute; S325. The selected contextual attributes are taken as the final contextual attributes of the feature terms; S33. Calculate the risk increment value of the target object under different evaluation dimensions based on the final contextual attributes of the feature items and the preset degradation knowledge rules.

8. The NLP-based intelligent line loss analysis method according to claim 1, characterized in that, The specific steps in step S4 include: S41. Determine the influence weight corresponding to the risk increment value; the influence weight reflects the degree of influence of the risk increment value on the state scores of each evaluation dimension of the target object; S42. Determine the time decay factor corresponding to the risk increment value; the time decay factor reflects the degree to which the impact of the risk increment value on the state scores of each evaluation dimension of the target object weakens over time; S43. Based on the influence weight and the time decay factor, the risk increment value is weighted and decayed to obtain the adjusted risk increment value; S44. The adjusted risk increment value is added to the state score of each evaluation dimension of the target object to achieve cumulative update of the state score and obtain the dynamic state of the target object.

9. The NLP-based intelligent line loss analysis method according to claim 1, characterized in that, The specific steps in step S5 include: S51. Establish a dynamic reference baseline for the state scores of each evaluation dimension; the dynamic reference baseline is obtained by smoothing the state scores within a preset time window. S52. Monitor the deviation between the state score and the dynamic reference baseline; S53. When the duration of the deviation between the state score and the dynamic reference baseline reaches a preset duration and the deviation exceeds a preset threshold, it is judged as a deterioration trend; S54. If the duration of the deviation does not reach the preset duration, or the magnitude of the deviation does not exceed the preset threshold, it is determined to be a non-faulty fluctuation.

10. A line loss intelligent analysis system based on NLP, characterized in that, include: The acquisition module is used to acquire monitoring information for the target object; A construction module is used to standardize the monitoring information and associate the processed monitoring information with the target object to construct an information sequence of the target object in the time dimension. The calculation module is used to parse the feature items in the information sequence and calculate the risk increment value of the target object under different evaluation dimensions based on the contextual attributes corresponding to the feature items and the preset degradation knowledge rules. The update module is used to cumulatively update the state scores of each evaluation dimension of the target object based on the risk increment value, so as to obtain the dynamic state of the target object; The monitoring module is used to monitor the evolution sequence of the state scores of each evaluation dimension in the dynamic state over time. The matching module is used to match the evolution sequence with a preset progressive failure mode, and output risk warning information for the target object when the match is successful.

Images