Intelligent early warning system for predictive maintenance of thermal power generating unit based on big data driving

Abstract

The application relates to the technical field of thermal power unit monitoring, and discloses a predictive maintenance intelligent early warning system for a thermal power unit based on big data driving, which realizes real-time acquisition of multidimensional sensor parameters through a data acquisition unit, realizes data and experience collaborative modeling through parameter correlation mining and combination of experience characteristics by a parameter prediction unit, realizes three-level health state evaluation from the safety and economy angles by an equipment health evaluation unit, embeds an experience correction mechanism, generates multistage alarms according to the deviation and the health state by an equipment state alarm unit, converts implicit experience into structured characteristics by an experience knowledge base unit, and continuously absorbs new experience by an experience feedback closed-loop unit through incremental learning and model retraining, so that the system realizes the transformation from passive response to active prediction, improves the accuracy and timeliness of equipment fault early warning, continuously optimizes the system capacity through man-machine cooperation, and effectively reduces the operation risk and operation and maintenance cost.

Description

Technical Field

[0001] This invention relates to the field of thermal power unit monitoring technology, and in particular to a big data-driven predictive maintenance intelligent early warning system for thermal power units. Background Technology

[0002] As the core equipment for power generation, the safe, stable, and efficient operation of thermal power units is crucial for ensuring energy supply. However, thermal power units are diverse in type, complex in structure, and subject to varied operating conditions. During long-term operation, problems such as wear, aging, and corrosion are inevitable, making them prone to failure, leading to shutdowns, power generation losses, and even safety accidents.

[0003] Traditional maintenance models for thermal power units mainly include reactive maintenance and periodic preventive maintenance. Reactive maintenance involves emergency repairs after equipment failure, which often results in long downtime, high maintenance costs, and the potential to trigger a chain reaction of failures. Periodic preventive maintenance involves inspections based on equipment operating time or fixed cycles. While this method can mitigate sudden failures to some extent, it carries the risk of over-maintenance or under-maintenance. This means that unnecessary maintenance may be performed on healthy equipment, leading to wasted resources; or failures may occur within the maintenance cycle, failing to be effectively prevented.

[0004] In recent years, with the development of industrial big data and artificial intelligence technologies, predictive maintenance has gradually become a research hotspot in the field of thermal power unit maintenance. Predictive maintenance aims to predict the future health status and potential failures of equipment by monitoring equipment status data in real time and using data analysis and modeling techniques. For example, CN111077876B discloses a method, device and system for intelligent evaluation and early warning of power plant equipment status, and CN114488996A discloses a method and system for equipment health monitoring and early warning.

[0005] The aforementioned and existing related technologies often suffer from the following shortcomings: Traditional equipment operation monitoring methods in thermal power plants primarily focus on setpoint alarms, paying less attention to parameter fluctuation ranges or deterioration trends. By the time equipment triggers parameter alarms or thermal protection actions, significant deterioration and failure have already occurred. Relying solely on structured data to build models fails to transform the implicit experience of equipment management personnel into quantifiable model inputs, hindering collaborative modeling between data-driven and experience-driven approaches, resulting in a one-sided reliance on data models. Furthermore, the models lack dynamic learning and adaptive optimization capabilities. Once determined based on historical data, they cannot automatically absorb new experience accumulated by management personnel, making it difficult to adjust evaluation parameters according to new experience. This means that the reliance on static manual experience cannot be completely eliminated, and evaluation biases due to the solidification or lack of experience still exist. Summary of the Invention

[0006] The technical problem to be solved by this invention is that the existing technology has the disadvantages of traditional monitoring being limited and the model lacking dynamic integration with experience. To address this, we propose a big data-driven predictive maintenance intelligent early warning system for thermal power units.

[0007] To achieve the above objectives, this application adopts the following technical solution: a big data-driven predictive maintenance intelligent early warning system for thermal power units, comprising: The data acquisition unit is used to collect sensor monitoring parameters of thermal power plant equipment in real time. The parameter prediction unit is used to use data mining technology to extract the correlation and coupling between equipment parameters from the monitoring parameter data, establish a regression model based on neural network algorithm, and use the regression model to predict and output the normal value of the monitoring parameters in real time based on the currently collected monitoring parameters. The equipment health assessment unit is used to analyze and evaluate the status of equipment parameter groups from the perspectives of safety and economy, and to derive a three-level health status based on the analysis and evaluation results. The equipment status alarm unit is used to generate equipment monitoring results and alarm information based on the deviation between the predicted normal value and the real-time collected monitoring parameters, combined with the three-level health status. An experience knowledge base unit is used to store implicit experience data input by equipment managers and convert it into structured feature labels through natural language parsing technology to provide input features to the parameter prediction unit and the equipment health evaluation unit. The experience feedback closed-loop unit is used to transform manually verified conclusions into standardized experience labels and feed them back to the experience knowledge base unit, triggering the incremental learning mechanism of the parameter prediction unit and the health evaluation unit to dynamically optimize the model.

[0008] Furthermore, the parameter prediction unit is also used for: The historical monitoring parameter data is preprocessed, including data cleaning, missing value handling, and outlier removal; Feature engineering methods are used to extract feature sets for training the neural network regression model from preprocessed historical data. The neural network algorithm uses a multilayer perceptron or a long short-term memory network, and its input layer dimension includes sensor parameter dimension and empirical feature label dimension. The regression model is updated and optimized regularly using new historical monitoring parameter data; When training and updating the regression model, the structured feature labels generated by the experience knowledge base unit are accessed simultaneously, quantifying manual experience rules into model input features, thereby realizing collaborative modeling driven by data and experience.

[0009] Furthermore, it also includes a data discrimination unit, used to receive real-time monitoring parameters collected by the data acquisition unit and compare the real-time monitoring parameters with the corresponding predicted normal values ​​output by the parameter prediction unit; When the deviation between the real-time monitoring parameter and the predicted normal value exceeds the preset first threshold, it is determined that the data of the measuring point is abnormal, and a real-time and historical list of abnormal devices containing abnormal measuring point information is generated. The device status alarm unit is also used to receive the list of abnormal devices and to assist in the inspection work accordingly.

[0010] Furthermore, the three-level health status of the equipment status alarm unit are specifically equipment-level, system-level, and unit-level. The equipment status alarm unit is also specifically used for: Configure multiple alarm modes, including: When the deviation between the real-time monitored parameters and the predicted normal values ​​exceeds the first threshold but is less than the preset second threshold, an early degradation warning alarm is triggered. When the deviation between the real-time monitoring parameter and the predicted normal value exceeds the second threshold, a fault early warning alarm is triggered. When the real-time monitored parameters reach the preset alarm limit, the alarm is triggered. The alarm information includes alarm level, alarm device, alarm parameters, and deviation value.

[0011] Furthermore, the equipment health evaluation unit includes: The equipment parameter group status evaluation module is used to score the status of related parameter groups that characterize the key status of the same equipment based on preset safety thresholds and economic thresholds, combined with real-time monitoring parameters and their deviations from predicted normal values. The equipment-level health status assessment module is used to comprehensively evaluate the status of all key parameter groups of the equipment to obtain the health status score and level of the equipment; The system-level health status evaluation module is used to derive the system's health status score and level based on the health status scores of all key devices that make up the system, combined with the system topology logic and weights. The unit-level health status assessment module is used to derive the unit's health status score and level based on the health status scores of all key systems that constitute the unit, combined with the unit's operating characteristics and weights. The evaluation modules all set scoring rules based on the security analysis and economic analysis; The scoring rules incorporate an experience correction factor. When equipment managers manually correct the evaluation results, the system automatically records the deviation between the corrected value and the original value, generates an experience correction rule, and feeds it back to the experience knowledge base unit.

[0012] Furthermore, the process of generating experience-based correction rules and feeding them back to the experience knowledge base unit includes: When equipment managers manually correct the evaluation results, the current operating conditions of the equipment are automatically associated with them. The operating conditions include the load rate range, ambient temperature group, and equipment life stage code. Based on the deviation between the corrected value and the original value, and combined with the operating conditions, three basic correction values ​​are decomposed: threshold offset, parameter weight error, and associated parameter coupling coefficient error. The three types of basic correction quantities are transformed into conditional rules with operating condition constraints, including: The data identification unit generates dynamic threshold adjustment rules based on the threshold offset. The rule format is: when the operating conditions meet the preset conditions, adjust the baseline value of the safety threshold or economic threshold. The data identification unit generates a scoring weight adjustment rule based on the parameter weight error. The rule is in the form of: when the operating conditions meet the preset conditions, adjust the weight coefficient of the economic scoring index. The data identification unit generates parameter association correction rules based on the error of the coupling coefficient of the associated parameters. The rule is in the form of: when the deviation of a specific parameter exceeds the critical value, the scoring conversion coefficient of the associated parameter is adjusted. The conditional rules are written into the experience knowledge base unit in real time via encrypted data streams.

[0013] Furthermore, the equipment health evaluation unit is also used for: Store historical evaluation results to generate historical trend charts of the health status of equipment, systems, and units; Based on the historical trend chart, identify the deterioration trend of health status; The identified degradation trend information is fed back to the device status alarm unit to supplement the generation of alarm information.

[0014] Furthermore, the experience knowledge base unit includes: The implicit experience storage module is used to receive unstructured fault judgment logic, operating rules, and factors affecting equipment lifespan input by equipment management personnel. The natural language parsing module is used to parse the implicit experience data into standardized text fragments through semantic analysis technology; The feature label generation module is used to extract key feature words based on the parsing results and generate structured feature labels with weighted coefficients. The collaborative modeling interface module is used to synchronize structured feature labels to the parameter prediction unit and the equipment health evaluation unit in real time, serving as input features for the neural network regression model and health status scoring rules.

[0015] Furthermore, the experience feedback closed-loop unit is specifically used for: Receive manual verification conclusions from equipment management personnel regarding alarm information or health evaluation results; When the human conclusion differs from the system's judgment, extract the decision-making basis and key evidence data input by the human. The decision-making basis is transformed into standardized experience tags and associated with the corresponding failure mode or life model in the experience knowledge base unit. The incremental learning mechanism of the trigger parameter prediction unit and the equipment health evaluation unit is used to optimize model parameters by integrating new experience labels according to a preset cycle.

[0016] Furthermore, the incremental learning mechanism specifically includes: When the accumulation of novel experience labels for a specific fault mode in the experience knowledge base reaches the confidence threshold, the model retraining is automatically initiated. An adaptive module for empirical feature channel weights is added to the neural network regression model to dynamically adjust the contribution of empirical labels to the prediction results. A dynamic threshold adjustment algorithm is inserted into the equipment health status scoring rules. The safety / economic thresholds are corrected in real time based on the lifespan influencing factors updated in the experience knowledge base. After the model is retrained, convergence verification is performed.

[0017] The technical effects and advantages of this invention are as follows: In this invention, the data acquisition unit acquires multi-dimensional sensor parameters in real time, providing comprehensive basic data for the system and solving the problems of delayed and incomplete data acquisition; the parameter prediction unit mines parameter correlations through neural network models and combines them with empirical features to achieve collaborative modeling of data and experience, overcoming the limitations of a simple data model; the equipment health evaluation unit forms a three-level health status evaluation from the perspectives of safety and economy, embedding an experience correction mechanism to solve the problems of single evaluation dimensions and susceptibility to the influence of solidified experience; the equipment status alarm unit generates multi-level alarms based on deviations and health status, solving the defects of traditional alarm methods that are single and have delayed early warning.

[0018] In this invention, the experience knowledge base unit transforms implicit experience into structured features, eliminating the pain point of the difficulty in quantifying and reusing manual experience; the experience feedback closed-loop unit, through incremental learning and model retraining, allows the system to continuously absorb new experience, solving the problem of static and rigid models that cannot adapt to changes in operating conditions. Overall, this achieves a shift from passive response to proactive prediction, improving the accuracy and timeliness of equipment fault warnings, and continuously optimizing system capabilities through human-machine collaboration, effectively reducing operational risks and maintenance costs, and providing strong support for the safe and economical operation of thermal power units. Attached Figure Description

[0019] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts: Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0020] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0021] Reference Figure 1 As shown, this embodiment of the invention provides a big data-driven predictive maintenance intelligent early warning system for thermal power units. This system aims to achieve intelligent predictive maintenance of thermal power unit equipment by integrating functional modules such as data acquisition, parameter prediction, health assessment, alarms, experiential knowledge, and feedback closed-loop. The system mainly includes a data acquisition unit, a parameter prediction unit, an equipment health assessment unit, an equipment status alarm unit, an experiential knowledge base unit, and an experiential feedback closed-loop unit.

[0022] The data acquisition unit is used for: The system collects sensor monitoring parameters from thermal power plant equipment in real time. These parameters include, but are not limited to, temperature, pressure, flow rate, vibration, current, voltage, rotational speed, and flue gas composition. The data acquisition unit 100 connects to field sensors, DCS, SIS, etc., via various industrial communication protocols (such as Modbus TCP / IP, OPCUA, Profibus, etc.) to ensure the real-time and accurate acquisition of monitoring parameters. The collected data undergoes preliminary timestamp marking and data format encapsulation in preparation for subsequent processing.

[0023] Specifically, the data acquisition unit collects real-time monitoring parameters from various sensors in the power plant equipment, such as temperature and pressure. It connects with field sensors, DCS, and SIS systems via industrial communication protocols like Modbus TCP / IP to ensure accurate and real-time parameter acquisition. After timestamping and data formatting, the data provides foundational data for subsequent parameter prediction and equipment health assessment. This comprehensive and timely collection of equipment operating status data supports subsequent data mining, model building, and intelligent early warning functions, forming the data-driven foundation of the entire predictive maintenance intelligent early warning system.

[0024] Parameter prediction unit, used for Data mining techniques are used to extract the correlation and coupling between equipment parameters from monitoring parameter data, and a regression model based on neural network algorithm is established. The regression model is used to predict and output the normal values ​​of the monitoring parameters in real time based on the currently collected monitoring parameters. Specifically, the historical monitoring parameter data is preprocessed, including data cleaning, missing value handling, and outlier removal. Feature engineering methods are used to extract feature sets from the preprocessed historical data to train the neural network regression model. The neural network algorithm uses a multilayer perceptron or long short-term memory network, and its input layer dimension includes sensor parameter dimension and empirical feature label dimension. The regression model is updated and optimized regularly using new historical monitoring parameter data. When training and updating the regression model, the structured feature labels generated by the experience knowledge base unit are accessed simultaneously, quantifying manual experience rules into model input features, and realizing collaborative modeling driven by data and experience.

[0025] Specifically, the parameter prediction unit leverages data mining techniques to uncover correlations among equipment parameters. After preprocessing historical data and extracting feature sets, a regression model is constructed using a multilayer perceptron or long short-term memory network. Through regular updates and optimizations, combined with structured feature labels from an experience knowledge base, data-driven and experience-based collaborative modeling is achieved. This enables accurate prediction of normal values ​​for monitoring parameters, providing a reliable basis for subsequent equipment status assessments and fault warnings. It effectively enhances the ability to predict equipment operating trends, allowing the model to adapt to complex equipment operating conditions and contributing to the construction of a more intelligent and accurate equipment management and prediction system, supporting the system in efficiently carrying out predictive maintenance and other tasks.

[0026] The equipment health assessment unit is used for: The status of equipment parameter groups is analyzed and evaluated from the perspectives of safety and economy, and equipment-level, system-level, and unit-level results are derived based on the analysis and evaluation results. Specifically, the equipment health assessment unit includes: The equipment parameter group status evaluation module is used to score the status of related parameter groups that characterize the key status of the same equipment based on preset safety thresholds and economic thresholds, combined with real-time monitoring parameters and their deviations from predicted normal values. The equipment-level health status assessment module is used to comprehensively evaluate the status of all key parameter groups of the equipment to obtain the health status score and level of the equipment; The system-level health status evaluation module is used to derive the system's health status score and level based on the health status scores of all key devices that make up the system, combined with the system topology logic and weights. The unit-level health status assessment module is used to derive the unit's health status score and level based on the health status scores of all key systems that constitute the unit, combined with the unit's operating characteristics and weights. The evaluation modules all set scoring rules based on safety analysis and economic analysis; The scoring rules incorporate experience correction factors. When equipment managers manually correct the evaluation results, the system automatically records the deviation between the corrected value and the original value, generates experience correction rules, and feeds them back to the experience knowledge base unit.

[0027] Specifically, the equipment health assessment unit constructs a multi-dimensional, multi-level assessment system to achieve full-level health status evaluation from equipment parameter groups to the entire unit. From both safety and economic perspectives, based on preset thresholds and considering the deviation between real-time monitored parameters and predicted normal values, it accurately scores related parameter groups characterizing key equipment states. This is then integrated level by level to form equipment-level, system-level, and unit-level health scores and grades. This hierarchical evaluation mechanism not only details subtle changes in specific parameter groups but also provides a macro-level understanding of the overall unit status. Experience-based correction factors embedded in the scoring rules generate new rules by recording manual corrections of deviations and feeding them back to the knowledge base, achieving dynamic optimization of the assessment system. This process not only ensures the comprehensiveness and accuracy of health assessments, providing precise basis for equipment maintenance, but also continuously improves assessment capabilities through human-machine collaboration, effectively enhancing unit operation safety, reducing maintenance costs, and improving overall economic efficiency.

[0028] The process of generating experience-based correction rules and feeding them back to the experience knowledge base unit includes: When equipment managers manually correct the evaluation results, the current operating conditions of the equipment are automatically associated with them. The operating conditions include the load rate range, ambient temperature group, and equipment life stage code. Based on the deviation between the corrected value and the original value, and combined with the operating conditions, three basic correction quantities are decomposed: threshold offset, parameter weight error, and related parameter coupling coefficient error. The three types of basic correction quantities are transformed into conditional rules with operating condition constraints, including: Dynamic threshold adjustment rules are generated based on threshold offset. The rule format is: when the operating conditions meet the preset conditions, the baseline value of the safety threshold or economic threshold is adjusted. The data identification unit generates scoring weight adjustment rules based on parameter weight error. The rule format is: when the operating conditions meet the preset conditions, the weight coefficient of the economic scoring index is adjusted. The data identification unit generates parameter association correction rules based on the coupling coefficient error of the associated parameters. The rule format is: when the deviation of a specific parameter exceeds the critical value, the scoring conversion coefficient of the associated parameter is adjusted. The conditional rules are written into the experience knowledge base unit in real time through encrypted data stream and take effect in the next evaluation cycle.

[0029] Specifically, in the equipment health assessment unit, the process of generating experience-based correction rules and feeding them back to the experience knowledge base, by associating them with the equipment's operating conditions, decomposes manual correction deviations into three types of basic correction quantities, transforming them into conditional rules with operating condition constraints. Dynamic threshold adjustments, scoring weight adjustments, and parameter-related correction rules are encrypted, written into the knowledge base, and take effect quickly, enabling dynamic optimization of assessment rules. This allows health assessments to adapt to diverse operating conditions, improves the accuracy of equipment status evaluation, and helps build a more realistic, intelligent, and self-optimizing equipment health assessment system, enhancing the scientific rigor and reliability of equipment health assessments.

[0030] Furthermore, it should be added that the equipment health assessment unit is also used for: The system stores historical evaluation results to form historical trend charts of the health status of equipment, systems, and units. Based on these historical trend charts, it identifies trends in health status deterioration and feeds back the identified deterioration trend information to the equipment status alarm unit to supplement the generation of alarm information.

[0031] The data screening unit is used for: The system receives real-time monitoring parameters collected by the data acquisition unit and compares them with the corresponding predicted normal values ​​output by the parameter prediction unit. When the deviation between the real-time monitoring parameters and the predicted normal values ​​exceeds a preset first threshold, it determines that the data at the measurement point is abnormal and generates a real-time and historical list of abnormal devices containing information on abnormal measurement points. The device status alarm unit is also used to receive the list of abnormal devices and to assist in the inspection work accordingly.

[0032] Specifically, the data identification unit compares real-time monitoring parameters with predicted normal values, and determines that the measuring point is abnormal when the deviation exceeds a first threshold, generating a real-time and historical list of abnormal equipment. The equipment status alarm unit receives the list to assist in inspection, solving the problems of low efficiency and delayed anomaly detection in manual inspection, achieving accurate anomaly identification and timely early warning, providing clear targets for inspection, and improving the ability to detect and handle equipment failures early.

[0033] The equipment status alarm unit is used for: Based on the deviation between the predicted normal value and the real-time collected monitoring parameters, and combining equipment-level, system-level, and unit-level data, equipment monitoring results and alarm information are generated. Specifically, set up multiple alarm modes, including: When the deviation between the real-time monitoring parameter and the predicted normal value exceeds the first threshold but is less than the preset second threshold, an early degradation warning alarm is triggered. When the deviation between the real-time monitoring parameter and the predicted normal value exceeds the second threshold, a fault warning alarm is triggered. When the real-time monitoring parameter reaches the preset set alarm limit, a set alarm is triggered. The alarm information includes the alarm level, alarm device, alarm parameter, and deviation value.

[0034] Experience knowledge base units are used for: The system stores implicit experience data input by equipment managers and transforms it into structured feature labels using natural language processing technology to provide input features to the parameter prediction unit and the equipment health evaluation unit.

[0035] Specifically, the equipment status alarm unit, by setting multiple alarm modes, triggers different levels of alarms based on the degree of parameter deviation. Combined with multi-level health assessments, it generates accurate alarm information, solving the problems of traditional alarm methods being singular and having delayed warnings. This achieves graded early warning from early deterioration to failure. The experience knowledge base unit transforms the implicit experience of management personnel into structured feature labels, providing input for parameter prediction and health assessment, solving the problem of the difficulty in quantifying and applying human experience. Together, these two units improve the timeliness of fault response through graded alarms and enhance the applicability of the model through experience transformation, comprehensively improving the accuracy of equipment status monitoring and the effectiveness of early warnings.

[0036] The experience knowledge base unit includes: The implicit experience storage module is used to receive unstructured fault judgment logic, operating rules, and factors affecting equipment lifespan input by equipment management personnel. The natural language parsing module is used to parse implicit experience data into standardized text fragments through semantic analysis techniques; The feature label generation module is used to extract key feature words based on the parsing results and generate structured feature labels with weighted coefficients. The collaborative modeling interface module is used to synchronize structured feature labels to the parameter prediction unit and the equipment health evaluation unit in real time, serving as input features for the neural network regression model and health status scoring rules.

[0037] Specifically, the experience knowledge base unit, through the collaborative operation of four modules, effectively addresses the core problem of the difficulty in transforming and reusing implicit human experience in traditional equipment management. The implicit experience storage module systematically receives unstructured experience accumulated by equipment managers, including fault diagnosis logic, operating rules, and factors influencing equipment lifespan, solving the problem of scattered storage and easy loss of this valuable experience. The natural language parsing module, using semantic analysis technology, transforms vague and non-standardized experience expressions into standardized text fragments, overcoming the obstacles of inconsistent experience descriptions and difficulty in direct application. The feature label generation module further extracts key feature words from the parsing results and assigns weight coefficients to generate structured feature labels, realizing the quantitative processing of implicit experience and solving the pain point of experience being difficult to integrate into data models. The collaborative modeling interface module synchronizes these structured feature labels to the parameter prediction unit and the equipment health evaluation unit in real time, serving as input features for the neural network regression model and health status scoring rules, allowing human experience to directly participate in the model construction and evaluation process, effectively compensating for the limitations of purely data-driven models under complex operating conditions. This complete closed loop not only achieves the standardization, structuring, and reuse of tacit experience, but also promotes the deep integration of data and experience, significantly improving the accuracy and adaptability of equipment parameter prediction and health assessment, and providing more comprehensive and reliable support for equipment management decisions.

[0038] The experience feedback closed-loop unit is used for: The conclusions of manual verification are transformed into standardized experience labels and fed back to the experience knowledge base unit, triggering the incremental learning mechanism of the parameter prediction unit and the health evaluation unit to dynamically optimize the model.

[0039] Specifically, the receiving equipment management personnel will manually verify the alarm information or health evaluation results. When human conclusions differ from system judgments, the decision-making basis and key evidence data input by humans are extracted, the decision-making basis is transformed into standardized experience labels, and associated with the corresponding fault mode or life model in the experience knowledge base unit. This triggers the incremental learning mechanism of the parameter prediction unit and the equipment health evaluation unit, and the new experience labels are integrated to optimize the model parameters according to a preset cycle.

[0040] Furthermore, when the accumulation of novel experience labels for a specific fault mode in the experience knowledge base reaches a confidence threshold, where the number of similar labels is ≥50 and the error rate is <5%, the model is automatically retrained. An adaptive module for empirical feature channel weights is added to the neural network regression model to dynamically adjust the contribution of empirical labels to the prediction results. A dynamic threshold adjustment algorithm is inserted into the equipment health status scoring rules. The safety / economic thresholds are corrected in real time based on the lifespan influencing factors updated in the experience knowledge base. After the model is retrained, convergence verification is performed: the model is considered to have converged when the rate of change of the loss function is ≤0.1% for three consecutive iterations.

[0041] Specifically, the experience feedback closed-loop unit addresses the problems of system models struggling to dynamically adapt to changes in actual operating conditions and the inability of human verification conclusions to effectively feed back into the system. By receiving human verification conclusions, when they differ from the system's judgment, the decision-making basis is transformed into standardized experience labels and linked to the knowledge base, triggering incremental learning in relevant units and periodically optimizing model parameters. When a new experience label for a specific fault mode reaches a confidence threshold, retraining is automatically initiated. Combining adaptive adjustment of experience feature channel weights and a dynamic threshold algorithm, model parameters and thresholds are corrected in real time. A convergence verification mechanism after model retraining ensures stable optimization results. This closed loop achieves the continuous transformation of human experience into system capabilities, improving the model's adaptability to complex operating conditions and enhancing the accuracy of equipment status assessment and early warning.

[0042] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A big data-driven predictive maintenance intelligent early warning system for thermal power units, characterized in that, include: The data acquisition unit is used to collect sensor monitoring parameters of thermal power plant equipment in real time. The parameter prediction unit is used to use data mining technology to mine the correlation and coupling between equipment parameters from the monitoring parameter data, establish a regression model based on neural network algorithm, the input layer dimension of the regression model includes sensor parameter dimension and empirical feature label dimension, and use the regression model to predict and output the predicted normal value of the corresponding monitoring parameter in real time based on the currently collected sensor monitoring parameters. The data discrimination unit is used to receive the real-time monitoring parameters collected by the data acquisition unit and compare the real-time monitoring parameters with the corresponding predicted normal values ​​output by the parameter prediction unit. The equipment health evaluation unit is used to score the status of related parameter groups representing the critical status of the same equipment from the perspectives of safety and economy, based on preset safety and economic thresholds, combined with real-time monitoring parameters and their deviations from predicted normal values, and to obtain equipment-level health status, system-level health status and unit-level health status based on the status scoring results. The scoring rules of the equipment health evaluation unit embed an experience correction factor. When the equipment manager manually corrects the evaluation results, the system automatically records the deviation between the corrected value and the original value, generates experience correction rules, and feeds them back to the experience knowledge base unit. The equipment status alarm unit is used to generate equipment monitoring results and alarm information based on the deviation between the predicted normal value and the real-time collected monitoring parameters, combined with the equipment-level health status, system-level health status and unit-level health status. An experience knowledge base unit is used to store implicit experience data input by equipment managers and convert it into structured feature labels through natural language parsing technology to provide input features to the parameter prediction unit and the equipment health evaluation unit. The experience feedback closed-loop unit is used to convert manual verification conclusions into standardized experience labels and feed them back to the experience knowledge base unit, triggering the incremental learning mechanism of the parameter prediction unit and the equipment health evaluation unit to dynamically optimize the model. The process of generating experience-based correction rules and feeding them back to the experience knowledge base unit includes: When equipment managers manually correct the evaluation results, the current operating conditions of the equipment are automatically associated with them. The operating conditions include the load rate range, ambient temperature group, and equipment life stage code. Based on the deviation between the corrected value and the original value, and combined with the operating conditions, three basic correction values ​​are decomposed: threshold offset, parameter weight error, and associated parameter coupling coefficient error. The three types of basic correction quantities are transformed into conditional rules with operating condition constraints, including: The data identification unit generates dynamic threshold adjustment rules based on the threshold offset. The dynamic threshold adjustment rules are as follows: when the operating conditions meet the preset conditions, the baseline value of the safety threshold or economic threshold is adjusted. The data identification unit generates a scoring weight adjustment rule based on the parameter weight error. The scoring weight adjustment rule is as follows: when the operating conditions meet the preset conditions, the weight coefficient of the economic scoring index is adjusted. The data identification unit generates parameter association correction rules based on the error of the coupling coefficient of the associated parameters. The parameter association correction rules are as follows: when the deviation of a specific parameter exceeds the critical value, the scoring conversion coefficient of the associated parameter is adjusted. The conditional rules are written into the experience knowledge base unit in real time via encrypted data streams.

2. The intelligent early warning system for predictive maintenance of thermal power units based on big data as described in claim 1, characterized in that: The parameter prediction unit is also used for: Preprocessing of historical monitoring parameter data includes data cleaning, handling of missing values, and removal of outliers; Feature engineering is used to extract a feature set for training the neural network regression model from preprocessed historical monitoring parameter data. The neural network algorithm uses a multilayer perceptron or a long short-term memory network. The regression model is updated and optimized regularly using new historical monitoring parameter data; When training and updating the regression model, the structured feature labels generated by the experience knowledge base unit are accessed simultaneously, quantifying manual experience rules into model input features, thereby realizing collaborative modeling driven by data and experience.

3. The intelligent early warning system for predictive maintenance of thermal power units based on big data as described in claim 2, characterized in that: The data screening unit is also used for: When the deviation between the real-time monitoring parameter and the corresponding predicted normal value exceeds the preset first threshold, it is determined that the measurement point data corresponding to the real-time monitoring parameter is abnormal, and a real-time abnormal device list and a historical abnormal device list containing abnormal measurement point information are generated. The device status alarm unit is also used to receive the real-time abnormal device list and the historical abnormal device list, and to assist in the inspection work accordingly.

4. The intelligent early warning system for predictive maintenance of thermal power units based on big data as described in claim 3, characterized in that: The device status alarm unit is also used to set multiple alarm modes, the multiple alarm modes including: When the deviation between the real-time monitored parameter and the corresponding predicted normal value exceeds the first threshold but is less than the preset second threshold, an early degradation warning is triggered. When the deviation between the real-time monitored parameter and the corresponding predicted normal value exceeds the second threshold, a fault early warning alarm is triggered. When the real-time monitored parameter reaches the preset alarm limit, the alarm is triggered. The alarm information includes alarm level, alarm device, alarm parameters, and deviation value.

5. The intelligent early warning system for predictive maintenance of thermal power units based on big data as described in claim 4, characterized in that: The equipment health assessment unit includes: The equipment parameter group status evaluation module is used to score the status of related parameter groups that characterize the key status of the same equipment based on preset safety thresholds and economic thresholds, combined with real-time monitoring parameters and their deviations from the corresponding predicted normal values. The equipment-level health status assessment module is used to comprehensively evaluate the status of all key parameter groups of the equipment to obtain the health status score and level of the equipment; The system-level health status evaluation module is used to derive the system's health status score and level based on the health status scores of all key devices that make up the system, combined with the system topology logic and weights. The unit-level health status assessment module is used to derive the unit's health status score and level based on the health status scores of all key systems that constitute the unit, combined with the unit's operating characteristics and weights. The evaluation modules all set scoring rules based on safety and economy.

6. The intelligent early warning system for predictive maintenance of thermal power units based on big data as described in claim 5, characterized in that: The equipment health assessment unit is also used for: Store historical evaluation results to generate historical trend charts of the health status of equipment, systems, and units; Based on the historical trend chart, identify the deterioration trend of health status; The identified degradation trend information is fed back to the device status alarm unit to supplement the generation of alarm information.

7. The intelligent early warning system for predictive maintenance of thermal power units based on big data as described in claim 6, characterized in that: The experience knowledge base unit includes: The implicit experience storage module is used to receive unstructured fault judgment logic, operating rules, and factors affecting equipment lifespan input by equipment management personnel. The natural language parsing module is used to parse the implicit experience data into standardized text fragments through semantic analysis technology; The feature label generation module is used to extract key feature words based on the parsing results and generate structured feature labels with weighted coefficients. The collaborative modeling interface module is used to synchronize structured feature labels to the parameter prediction unit and the equipment health evaluation unit in real time, serving as input features for the neural network regression model and health status scoring rules.

8. The intelligent early warning system for predictive maintenance of thermal power units based on big data as described in claim 7, characterized in that: The experience feedback closed-loop unit is specifically used for: Receive manual verification conclusions from equipment management personnel regarding alarm information or health evaluation results; When the human conclusion differs from the system's judgment, extract the decision-making basis and key evidence data input by the human. The decision-making basis is transformed into standardized experience tags and associated with the corresponding failure mode or life model in the experience knowledge base unit. The incremental learning mechanism of the trigger parameter prediction unit and the equipment health evaluation unit is used to optimize model parameters by integrating new experience labels according to a preset cycle.

9. The intelligent early warning system for predictive maintenance of thermal power units based on big data as described in claim 8, characterized in that: The incremental learning mechanism specifically includes: When the accumulation of novel experience labels for a specific fault mode in the experience knowledge base reaches the confidence threshold, the model retraining is automatically initiated. An adaptive module for empirical feature channel weights is added to the neural network regression model to dynamically adjust the contribution of empirical labels to the prediction results. A dynamic threshold adjustment algorithm is inserted into the equipment health status scoring rules. The safety / economic thresholds are corrected in real time based on the lifespan influencing factors updated in the experience knowledge base. After the model is retrained, convergence verification is performed.

Images