An operation and maintenance data processing system and method based on natural language processing

Abstract

This invention relates to the field of power plant data processing, and discloses a natural language processing-based operation and maintenance data processing system and method to address the problems of existing technologies relying on big data, lacking interpretability, and being unable to quantitatively integrate textual experience with physical laws. The system includes parsing operation and maintenance text into structured semantic units; quantifying these units into continuous influencing factors driving the physical model; using a non-machine learning theoretical physical model for state inference and prediction; comparing the prediction results with historical real states to generate difference data; and finally, optimizing the semantic quantification rules based on the difference data. This system achieves interpretable prediction and continuous self-evolution under small sample conditions, effectively transforming qualitative operation and maintenance experience into quantitative decision-making basis, and improving the reliability of operation and maintenance decisions and the ability to accumulate knowledge.

Description

Technical Field

[0001] This invention relates to the field of power plant data processing, and more particularly to an operation and maintenance data processing system and method based on natural language processing. Background Technology

[0002] With the deepening of digital transformation and intelligent operation and maintenance in the power industry, the data accumulated by power plants in daily operation is growing exponentially. This data includes not only massive amounts of structured time-series data from sensors and control systems, but also largely untapped unstructured text data, such as operation logs, equipment inspection reports, maintenance work orders, and accident analysis reports. Currently, the industry's utilization of operation and maintenance text data still has the following shortcomings: 1. It can only achieve structured extraction and simple classification of information, but cannot transform the "degree" described in the text into a continuous variable that can be used for quantitative analysis; 2. It heavily relies on large-scale, high-quality labeled data. There are very few severe failure samples of many key equipment in power plants, which makes it difficult for the model to learn effective prediction patterns and causes it to fail in small-sample or zero-sample failure scenarios. 3. When the model issues an early warning, maintenance personnel cannot directly link it to the physical principles of the equipment, making it difficult to assess the risks and make credible decisions.

[0003] Therefore, we propose an operation and maintenance data processing system and method based on natural language processing to solve the above problems. Summary of the Invention

[0004] This invention provides an operation and maintenance data processing system and method based on natural language processing, which solves the problems of existing technologies relying on big data, lacking interpretability, and being unable to quantitatively integrate textual experience with physical laws.

[0005] The first aspect of this invention provides a method for processing operation and maintenance data based on natural language processing. This method includes: parsing and extracting input power plant text data to generate a set of structured semantic units; quantizing the set of structured semantic units into continuous influencing factors based on a domain knowledge mapping table to form a set of equipment status influencing factors; inputting the set of equipment status influencing factors into a pre-stored non-machine learning theoretical physics prediction model to output equipment status prediction results; comparing the equipment status prediction results with historical real states to generate prediction-actual state difference data; and adjusting the quantization correspondence in the domain knowledge mapping table in reverse based on the prediction-actual state difference data to generate semantic quantization rules.

[0006] Optionally, in the first implementation of the first aspect of the present invention, the power plant text data is read, and based on a power plant domain entity dictionary, the power plant text data is annotated, words are identified, and a word sequence is generated; the word sequence is matched according to a syntactic pattern rule base to identify the relationships between words; the identified equipment entities, status attributes, and operation and maintenance actions are verified and disambiguated to generate a verified relationship set; the verified relationship set is encapsulated to generate structured semantic units, and the structured semantic units of all equipment entities together constitute a structured semantic unit set.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the following steps are taken: receiving the set of structured semantic units; querying the domain knowledge mapping table to obtain the basic physical action type and basic influence intensity value; generating a mapping query result; querying the device-state-action correspondence table based on the device entity type and state attribute of the structured semantic units to determine the physical state variable and generate a state variable index; using the basic influence intensity value as input, converting the lexical description into a precise quantitative value through calculation; combining the precise quantitative value with the state variable index and attaching the timestamp information corresponding to the semantic unit to generate a device state influence factor; processing the timestamp information to calculate the attenuation ratio and correcting the device state influence factor to generate the final influence factor; traversing all structured semantic units and repeating the steps to form a set of device state influence factors.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the set of equipment state influencing factors is received, the measured physical state values ​​corresponding to the set of equipment state influencing factors are obtained, and current initial condition data is generated; based on the state variable index in the set of equipment state influencing factors, the influencing factors are grouped and aggregated to generate a subset of grouped influencing factors; according to the type of the physical state variable, the corresponding theoretical prediction sub-model in the pre-stored non-machine learning theoretical physics prediction model is called; the current initial condition data is input into the corresponding theoretical prediction sub-model as the starting point of the calculation, and the subset of grouped influencing factors is used as input to drive the model to perform time-step state extrapolation calculation; after the theoretical prediction sub-model completes the extrapolation calculation, it outputs a numerical sequence.

[0009] Optionally, in the fourth implementation of the first aspect of the present invention, the device state prediction result is received, which is a predicted numerical sequence; a specified historical time interval is obtained, which corresponds to the prediction result; based on the device entity identifier and physical state variable index associated with the device state prediction result, the actual measurement value sequence is retrieved to generate a historical real state data sequence; the predicted numerical sequence and the historical real state data sequence are aligned on the time axis and compared to obtain numerical differences, forming an instantaneous difference value sequence; the instantaneous difference value sequence is analyzed to identify abnormal time points, calculate the average deviation, and generate a difference feature set; the instantaneous difference value sequence and the difference feature set are integrated to constitute the predicted and actual state difference data.

[0010] Optionally, in the fifth implementation of the first aspect of the present invention, the following steps are taken: receiving the difference data between the predicted and actual states, obtaining the original input, parsing the set of difference features, identifying the overall deviation direction, analyzing the time points of occurrence of key anomalies, matching them, and locating specific semantic descriptions; adjusting the domain knowledge mapping table based on the analysis results of the overall deviation direction; analyzing the context of specific semantic units in conjunction with the specific semantic descriptions, generating a mapping table modification instruction set; executing the mapping table modification instruction set to update the domain knowledge mapping table and generate semantic quantification rules.

[0011] Optionally, in the sixth implementation of the first aspect of the present invention, the equipment status prediction results are received and analyzed to identify time points and generate predictive early warning time points; based on the predictive early warning time points and the associated types of out-of-limit physical state variables, preventive operation and maintenance suggestions are matched to generate a set of operation and maintenance strategy suggestions; the set of operation and maintenance strategy suggestions is output to the power plant operation and maintenance management system to generate preventive maintenance work orders or operation guidance documents and record them; after the observation period, equipment monitoring data is collected and compared with the predicted trajectory to generate actual effect evaluation data after strategy intervention; the actual effect evaluation data after strategy intervention is used as new feedback information and input into the semantic quantification rule optimization process to feed the effect of actual operation and maintenance actions back into the continuous optimization process of the semantic quantification rules.

[0012] A second aspect of this invention provides an operation and maintenance data processing system based on natural language processing. The system includes: a parsing module for parsing and extracting input power plant text data to generate a set of structured semantic units; a quantization module for quantizing the set of structured semantic units into continuous influencing factors based on a domain knowledge mapping table, forming a set of equipment status influencing factors; a deduction module for inputting the set of equipment status influencing factors into a pre-stored non-machine learning theoretical physics prediction model and outputting equipment status prediction results; a difference module for comparing the equipment status prediction results with historical real states to generate prediction-actual state difference data; and an optimization module for adjusting the quantization correspondence in the domain knowledge mapping table in reverse based on the prediction-actual state difference data to generate semantic quantization rules.

[0013] The mechanism of this invention is as follows: Vague "degree descriptions" and "operational actions" in text are mapped to continuous, dimensionless correction factors affecting specific physical state variables, thereby driving state deduction of a deterministic model based on classical physics or system theory. The system traces deviations in the deduction results back to the original textual semantic elements that generated the effect, and adjusts the mapping relationship from semantics to physical effects accordingly, thus achieving precise alignment and co-evolution of operational knowledge and the laws of the physical world in iterative feedback.

[0014] Beneficial effects: It abandons the approach of fitting a black-box model from massive amounts of data; its core predictive capability stems from a pre-defined first-principles physical model. It can automatically reverse-calibrate the mapping relationship from "textual vocabulary" to "physical impact" using each prediction deviation. This allows the system to continuously evolve through feedback from daily operational texts, becoming more accurate with use, even in the absence of labeled samples. Based on deterministic physical laws, the predictions are fully traceable and interpretable. This provides a solid theoretical basis for high-risk decision-making, greatly enhances reliable decision-making capabilities in data-scarce scenarios, and allows successful operational actions to be back-injected into semantic quantification rules. This experience can be structured and quantified for future reference by all similar equipment. By sensing the state from text, recognizing trends through physical models, outputting specific early warning time points and operation and maintenance strategy suggestions, driving actual maintenance actions, and ultimately optimizing its own cognitive rules based on the results of these actions, this closed loop allows on-site human experience to be automatically absorbed, verified, and refined, ultimately building a living system that continuously accumulates and refines power plant-specific knowledge over time. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of an embodiment of an operation and maintenance data processing method based on natural language processing according to the present invention; Figure 2 This is a schematic diagram of semantic element parsing and relation extraction. Figure 3 This is a schematic diagram of an embodiment of an operation and maintenance data processing system based on natural language processing according to an embodiment of the present invention. Detailed Implementation

[0016] This invention provides an operation and maintenance data processing system and method based on natural language processing, addressing the problems of existing technologies relying on big data, lacking interpretability, and being unable to quantitatively integrate textual experience with physical laws. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0017] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 and Figure 2 An embodiment of an operation and maintenance data processing method based on natural language processing in this invention includes: 101. Perform semantic element parsing and relation extraction on the input power plant operation and maintenance text data to generate a set of structured semantic units containing equipment entities, status attributes, operation and maintenance actions, and degree modifiers.

[0018] It is understood that the executing entity of this invention can be an operation and maintenance data processing device based on natural language processing, or it can be a terminal or a server; the specific implementation is not limited here. This embodiment of the invention will be described using a server as an example.

[0019] Specifically, the text data to be processed is read from the power plant operation and maintenance data, which includes at least one of the following: inspection logs, maintenance reports, or operation records. Based on a pre-defined entity dictionary for the power plant domain, the text data is segmented and entity-labeled to identify words in the text that belong to equipment entities, status attributes, operation and maintenance actions, or degree modifiers, and to generate a preliminary word sequence with type labels. Based on a pre-defined syntactic pattern rule base, the sequence of lexical units with type labels is matched to identify the relationships between lexical units. These relationships include: descriptive relationships describing the state attributes of a specific device entity, and application relationships that perform specific maintenance actions on a specific device entity. Based on the device context constraints in the preset domain knowledge base, consistency verification and disambiguation are performed on the identified device entities, status attributes and operation and maintenance actions to determine the unique status attributes and operation and maintenance actions associated with each device entity and generate a set of verified relationships. The validated set of relationships is encapsulated in a structured manner, generating an independent structured semantic unit for each device entity. This unit explicitly contains at least one device entity, at least one associated state attribute and degree modifier, or at least one associated operation and maintenance action and degree modifier. The structured semantic units of all device entities together constitute a set of structured semantic units.

[0020] 102. Based on a predefined domain knowledge mapping table, the degree modifiers and operation and maintenance actions in the structured semantic unit set are quantified into dimensionless continuous influence factors corresponding to specific physical state variables, forming a set of equipment state influence factors.

[0021] Specifically, the set of structured semantic units is received; for each structured semantic unit in the set, the predefined domain knowledge mapping table is queried using the operation and maintenance action type and degree modifier contained therein as the query key to obtain the basic physical action type and basic influence intensity value corresponding to the semantic unit, and a mapping query result for each semantic unit is generated. Based on the device entity type and state attributes contained in the structured semantic unit, and combined with the basic physical action type, the predefined device-state-action correspondence table is queried to determine the physical state variables specifically affected by the semantic unit, and a state variable index is generated; physical state variables include vibration amplitude, thermodynamic efficiency or electrical stability parameters. Using the basic influence strength value as input, combined with the specific vocabulary of the degree modifier, the discrete vocabulary description is transformed into a dimensionless, precise quantified value that varies within a preset continuous interval by a predefined dimensionless continuous normalization function. The precise quantized value is combined with the state variable index and the timestamp information corresponding to the semantic unit is attached to generate an initial device state influence factor. Based on the predefined time-sensitivity decay rule, the timestamp information is processed to calculate the impact decay ratio from the time the event occurred to the current analysis time. This decay ratio is then used to correct the precise quantification value in the initial equipment status impact factor, generating the final impact factor with time-sensitivity weight. Traverse all structured semantic units, repeat the above steps, and aggregate all generated final influencing factors to form a set of equipment state influencing factors, which will serve as the original input for driving the physical prediction model to perform state calculations in subsequent steps.

[0022] 103. Input the set of equipment status influencing factors into the pre-stored non-machine learning theoretical physics prediction model, drive the model to perform state calculations, and output the equipment status prediction results for a specific future time period.

[0023] Specifically, the system receives the set of equipment status influencing factors and simultaneously obtains the measured physical status values ​​at the current moment corresponding to the equipment entities targeted by the set of influencing factors from the power plant real-time monitoring database, thereby generating the current initial condition data for the state calculation. Based on the state variable index carried by each impact factor in the impact factor set, the impact factors are grouped and aggregated according to the physical state variables they affect, generating grouped impact factor subsets for different physical state variables; Based on the type of physical state variable, the corresponding theoretical prediction sub-model in the pre-stored non-machine learning theoretical physics prediction model is called; the theoretical prediction sub-model is built based on the first principles of physics or classical system theory, and its built-in parameters are preset according to the equipment model. The current initial condition data is input into the corresponding theoretical prediction sub-model as the starting point of the calculation, and the grouped impact factor subset is input as the dynamic correction term. Within the model, the text impact factor is quantified into a continuous adjustment amount for specific parameters or variables within the model according to its theoretical framework, driving the model to perform time-step state extrapolation calculation. After the theoretical prediction sub-model completes the extrapolation calculations for all time steps within a specific future period, it outputs a numerical sequence arranged in chronological order that describes the trajectory of change of a specific physical state variable within the future period. This sequence is the equipment state prediction result and serves as the original input for comparison with the historical true state in subsequent steps.

[0024] 104. Compare the equipment status prediction results with the historical real status of the corresponding equipment in the same period extracted from the actual operation and maintenance records to generate data on the difference between the prediction and the actual status.

[0025] Specifically, the system receives the device state prediction results, which are a sequence of predicted values ​​arranged in chronological order and describing the trajectory of changes of specific physical state variables in the future period. At the same time, it acquires the specified historical time interval, which corresponds exactly to the "specific future period" set when the prediction results are generated in terms of time length and start time. Based on the equipment entity identifier and physical state variable index associated with the equipment status prediction results, the actual measurement value sequence of the equipment entity with respect to the physical state variables within the specified historical time interval is retrieved from the power plant historical archive database or historical monitoring data warehouse to generate a historical real state data sequence. The predicted numerical sequence is aligned with the historical real state data sequence on the time axis. Based on a unified timestamp benchmark, the predicted value and the actual value at the same moment in the two sequences are compared and calculated point by point to obtain the numerical difference at each time point, forming a sequence of instantaneous difference values ​​arranged by time. Feature analysis is performed on the instantaneous difference value sequence to identify abnormal time points in the sequence where the difference exceeds a preset threshold, and the average deviation between the predicted value and the actual value is calculated throughout the entire comparison period to generate a set of difference features that includes the location of key abnormal points and the overall degree of deviation. The instantaneous difference value sequence is integrated with the difference feature set to form the difference data between the prediction and the actual state. This difference data will serve as the original basis for adjusting the semantic quantization rules in subsequent steps.

[0026] 105. Based on the difference between the predicted and actual states, the quantitative correspondence between semantic elements and physical influence factors in the domain knowledge mapping table is adjusted in reverse to generate updated semantic quantification rules, which are then used for subsequent processing of new operation and maintenance text data.

[0027] Specifically, the system receives data on the difference between the predicted and actual states, which includes a sequence of instantaneous difference values ​​and a set of difference features consisting of the overall degree of deviation and the location of key anomalies. Simultaneously, it obtains the original input associated with generating the difference data, namely the set of structured semantic units used for the query and its corresponding set of equipment state influencing factors. The set of differential features is analyzed to identify the overall deviation direction: if the predicted value is systematically higher than the actual value, it is determined that the original quantitative relationship caused the influencing factor to be overestimated; if it is systematically lower than the actual value, it is determined to be underestimated; at the same time, the time points of the occurrence of key anomalies are analyzed and matched with specific structured semantic units that take effect at similar time points in the original input to locate specific semantic descriptions that may be directly related to the prediction deviation. Based on the analysis results of the overall deviation direction, the "basic influence intensity value" of the relevant entries in the domain knowledge mapping table is adjusted in a directional manner: for cases that are judged to be overestimated, the basic intensity value corresponding to the relevance modifier or operation and maintenance action is reduced according to the preset rules; for cases that are underestimated, the corresponding value is increased, and a preliminary draft of the mapping table entries is generated. Based on the specific semantic descriptions located and associated with key anomalies, the draft of the preliminary map entries is refined: the context of these specific semantic units is analyzed, and if the event described is the key reason for the predicted mutation but the actual stability, the adjustment range of the corresponding entries is significantly increased; if the effect is confirmed to be unrelated to the bias, the adjustment is reduced or canceled, and the final set of map modification instructions is generated. The execution of the mapping table modification instruction set overwrites and updates the values ​​of specific entries in the predefined domain knowledge mapping table, thereby generating updated semantic quantization rules. These updated rules will serve as the new "predefined domain knowledge mapping table" to process subsequent new input operation and maintenance text data, achieving closed-loop continuous optimization of semantic quantization relationships based on prediction feedback.

[0028] 106. Predictive early warnings trigger preventive operation and maintenance suggestions, evaluate the actual effect after implementation, and feed the effect evaluation back to semantic quantification rules to form a closed-loop feedback from prediction to action to knowledge optimization.

[0029] Specifically, the system receives the equipment status prediction results and analyzes the numerical sequence in the prediction results according to the predefined equipment health status threshold rules. It identifies the time point when the physical state variable in the prediction trajectory first exceeds or is expected to exceed its preset normal range boundary, and generates a predictive warning time point for the specific equipment. Based on the predictive warning time point and the associated over-limit physical state variable type, the pre-set operation and maintenance strategy knowledge base is queried to match the preventive operation and maintenance operation suggestions applicable to the device, the type of abnormal state and to be executed before the warning time point, and generate a structured set of operation and maintenance strategy suggestions. The set of operation and maintenance strategy recommendations is output to the power plant operation and maintenance management system, which triggers the generation of corresponding preventive maintenance work orders or operation guidance documents. The actual execution status, including the execution time, the specific operation content, and the equipment status retest data after execution, is recorded as the original input for subsequent effect verification. After the actual preventive maintenance operation is completed and a preset observation period has elapsed, the actual physical status monitoring data of the equipment during the observation period is collected. This data is then compared with the predicted trajectory of the equipment in the same time period under the condition that the operation was not performed, based on the original equipment status prediction results, to generate actual effect evaluation data after the strategy intervention. The actual effect evaluation data after the strategy intervention is used as new feedback information to input into the semantic quantification rule optimization process: if the actual effect is significantly better than the original prediction (i.e. the fault is effectively avoided or postponed), the quantification relationship of the influence factor corresponding to the original text semantic unit that caused the warning is strengthened; if the effect is not significant, it is weakened, thereby accumulating the knowledge of the effect of the actual operation and maintenance actions and feeding it back into the continuous optimization process of the semantic quantification rule.

[0030] In this invention, experiential knowledge is quantified by transforming vague descriptions such as "slight" and "serious" in text into continuous correction quantities driving the physical model. It possesses self-evolutionary capabilities even with small sample sizes; the system does not rely on historical large-scale data training but automatically reverse-calibrates the semantic-to-physical impact mapping rules based on the deviation between predicted results and actual states, making the system increasingly accurate with use. It ensures end-to-end interpretability of the prediction process; any warning can be clearly traced back to specific textual descriptions and physical evolution laws, providing a reliable basis for high-risk decision-making, especially suitable for warnings of extreme faults with limited data. It establishes a continuous accumulation mechanism for operational knowledge, feeding back the actual effects of each preventative maintenance to the system, thereby transforming personal experience into reusable quantitative rules and achieving automated accumulation and refinement of enterprise knowledge. This invention provides a new path for intelligent operation and maintenance of power plants that does not rely on black-box models and possesses both physical reliability and self-optimization capabilities.

[0031] Please see Figure 3 An embodiment of an operation and maintenance data processing system based on natural language processing according to the present invention includes: 201. Parsing module: Parses and extracts the input power plant text data to generate a set of structured semantic units; 202. Quantization module: Based on the domain knowledge mapping table, quantizes the set of structured semantic units into continuous influencing factors to form a set of equipment status influencing factors; 203. Inference module: Inputs the set of equipment status influencing factors into a pre-stored non-machine learning theoretical physics prediction model and outputs the equipment status prediction result; 204. Difference module: Compares the equipment status prediction result with the historical real state to generate prediction-actual state difference data; 205. Optimization module: Based on the prediction-actual state difference data, reversely adjusts the quantization correspondence in the domain knowledge mapping table to generate semantic quantization rules.

[0032] In this embodiment of the invention, without relying on historical big data training, fuzzy textual experience can be transformed into precise input to drive the physical model, enabling interpretable prediction under small sample conditions; and by automatically optimizing semantic quantification rules through prediction bias, the system has the ability to continuously self-evolve and accumulate knowledge, fundamentally solving the problems of data dependence, black box operation and knowledge solidification of traditional methods.

[0033] The present invention also provides an operation and maintenance data processing device based on natural language processing. The operation and maintenance data processing device based on natural language processing includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor performs the steps of the operation and maintenance data processing method based on natural language processing in the above embodiments.

[0034] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the operation and maintenance data processing method based on natural language processing.

[0035] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0036] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0037] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for processing operation and maintenance data based on natural language processing, characterized in that, The operation and maintenance data processing method based on natural language processing includes: The input power plant text data is parsed and extracted to generate a set of structured semantic units; Based on the domain knowledge mapping table, the set of structured semantic units is quantified into continuous influence factors to form a set of equipment status influence factors; The set of equipment status influencing factors is input into a pre-stored non-machine learning theoretical physics prediction model, and the equipment status prediction result is output. The predicted equipment status is compared with the historical actual status to generate data on the difference between the predicted and actual status. Based on the difference between the prediction and the actual state data, the quantitative correspondence in the domain knowledge mapping table is adjusted in reverse to generate semantic quantization rules.

2. The operation and maintenance data processing method based on natural language processing according to claim 1, characterized in that, include: Read the power plant text data, annotate the power plant text data based on the power plant domain entity dictionary, identify words, and generate word sequence; Based on the syntactic pattern rule base, the word sequence is matched to identify the association between words; The identified device entities, status attributes, and maintenance actions are verified and disambiguated to generate a set of verified relationships. The verified set of relationships is encapsulated to generate structured semantic units. The structured semantic units of all device entities together constitute a set of structured semantic units.

3. The operation and maintenance data processing method based on natural language processing according to claim 2, characterized in that, include: Receive the set of structured semantic units, query the domain knowledge mapping table, obtain the basic physical action type and basic influence intensity value, and generate mapping query results; Based on the device entity type and state attributes of the structured semantic unit, query the device-state-function correspondence table, determine the physical state variables, and generate a state variable index; Using the aforementioned basic influence intensity value as input, the lexical description is converted into a precise quantitative value through calculation; The precise quantized value is combined with the state variable index and the timestamp information corresponding to the semantic unit is attached to generate the device state influence factor. The timestamp information is processed to calculate the attenuation ratio, and the device status impact factor is corrected to generate the final impact factor. Traverse all structured semantic units, repeat the steps, and form a set of device status influencing factors.

4. The operation and maintenance data processing method based on natural language processing according to claim 3, characterized in that, include: Receive the set of equipment status influencing factors, obtain the measured physical status values ​​corresponding to the set of equipment status influencing factors, and generate the current initial condition data; Based on the state variable index in the set of equipment state influencing factors, the influencing factors are grouped and aggregated to generate a subset of grouped influencing factors. Based on the type of the physical state variable, the corresponding theoretical prediction sub-model in the pre-stored non-machine learning theoretical physics prediction model is invoked; The current initial condition data is input into the corresponding theoretical prediction sub-model as the starting point of the calculation, and the grouped influence factor subset is used as input to drive the model to perform time-step state extrapolation calculation; After the theoretical prediction sub-model completes the deduction calculation, it outputs a numerical sequence.

5. The operation and maintenance data processing method based on natural language processing according to claim 4, characterized in that, include: Receive the device status prediction result, which is a predicted numerical sequence, and obtain a specified historical time interval, which corresponds to the prediction result; Based on the equipment entity identifier and physical state variable index associated with the equipment state prediction results, the actual measurement value sequence is retrieved to generate a historical real state data sequence. The predicted numerical sequence is aligned with the historical real state data sequence on the time axis, and the comparison calculation is performed to obtain the numerical difference, forming an instantaneous difference value sequence. The instantaneous difference value sequence is analyzed to identify abnormal time points, calculate the average deviation, and generate a set of difference features. The instantaneous difference value sequence is integrated with the difference feature set to form the difference data between the predicted and actual states.

6. The operation and maintenance data processing method based on natural language processing according to claim 5, characterized in that, include: Receive the difference data between the prediction and the actual state, obtain the original input, parse the set of difference features, identify the overall deviation direction, analyze the time points of key anomalies, match them, and locate specific semantic descriptions. Based on the analysis results of the overall deviation direction, the domain knowledge mapping table is adjusted; Based on the specific semantic description, the context of the specific semantic unit is analyzed to generate a mapping table modification instruction set; The mapping table modification instruction set is executed to update the domain knowledge mapping table and generate semantic quantization rules.

7. The operation and maintenance data processing method based on natural language processing according to claim 6, characterized in that, Also includes: The device status prediction results are received, analyzed, and time points are identified to generate predictive early warning time points. Based on the predictive early warning time point and the associated over-limit physical state variable type, preventive operation and maintenance operation suggestions are matched to generate a set of operation and maintenance strategy suggestions; The aforementioned set of operation and maintenance strategy recommendations is output to the power plant operation and maintenance management system to generate preventive maintenance work orders or operation guidance documents and is recorded. After the observation period, the collected equipment monitoring data is compared with the predicted trajectory to generate actual effect evaluation data after the strategy intervention; The actual effect evaluation data after the intervention of the strategy is used as new feedback information and input into the semantic quantification rule optimization process, so that the effect of the actual operation and maintenance actions is fed back into the continuous optimization process of the semantic quantification rules.

8. A maintenance data processing system based on natural language processing, characterized in that, The operation and maintenance data processing system based on natural language processing includes: The parsing module parses and extracts the input power plant text data to generate a set of structured semantic units; The quantization module, based on the domain knowledge mapping table, quantizes the set of structured semantic units into continuous influence factors, forming a set of equipment status influence factors; The inference module inputs the set of equipment status influencing factors into a pre-stored non-machine learning theoretical physics prediction model and outputs the equipment status prediction results. The difference module compares the predicted device status with the historical actual status to generate data on the difference between the predicted and actual status. The optimization module, based on the difference between the prediction and the actual state data, reversely adjusts the quantitative correspondence in the domain knowledge mapping table to generate semantic quantization rules.

Images