A data-driven thermal power unit equipment health early warning method and device

Abstract

The application provides a kind of data-driven thermal power unit equipment health early warning method and device, comprising: constructing historical sample data set;The feature engineering extraction is carried out to historical sample data set, determine a plurality of sensitive characteristic parameters of the equipment health state, and construct the equipment health benchmark baseline based on sensitive characteristic parameters;From real-time operation data, the real-time characteristic parameters corresponding to sensitive characteristic parameters are extracted;The multidimensional health deviation of real-time characteristic parameters relative to health benchmark baseline is calculated;Multi-dimensional health deviation is input into the health early warning model pre-trained, and the current equipment health state score and predicted degradation trend are output by health early warning model;Combined with the preset grading early warning rule, determine whether to trigger early warning and the early warning level triggered.The application can effectively solve the problem that the traditional fixed threshold method is poor in adaptability under variable working conditions, the false alarm rate is high, improve the equipment fault prediction ability and the intelligent level of operation and maintenance decision.

Description

Technical Field

[0001] This application relates to the field of intelligent operation technology for thermal power units, and more specifically, to a data-driven method and device for early warning of equipment health in thermal power units. Background Technology

[0002] As a crucial component of the power system, the health status of thermal power units directly impacts power generation efficiency, operational safety, and grid stability. With the development of thermal power units towards deeper peak shaving and flexible operation, the alternating stress on the equipment increases, raising the risk of failure and placing higher demands on the real-time and accurate nature of equipment health management.

[0003] Traditional thermal power unit equipment maintenance mainly adopts regular maintenance (planned maintenance) or post-failure maintenance (failure maintenance). Regular maintenance has the problem of "over-maintenance" or "under-maintenance", resulting in waste of manpower and material resources; post-failure maintenance may lead to unplanned shutdowns, causing huge economic losses to power generation companies. In recent years, with the development of condition monitoring technology, some power plants have begun to try equipment health management based on threshold alarms, that is, to set upper and lower limits of a single parameter to trigger an alarm for exceeding the limit. However, this method has the following defects: (1) a single parameter threshold is difficult to reflect the comprehensive health status of the equipment under the coupling effect of multiple parameters, and the false alarm rate and missed alarm rate are high; (2) the threshold is usually statically set and cannot adapt to the operating condition drift caused by changes in unit load, changes in ambient temperature, etc.; (3) it lacks the ability to predict the trend of equipment deterioration and cannot achieve early warning.

[0004] Meanwhile, with the development of the Industrial Internet and intelligent control platforms, thermal power units have accumulated massive amounts of operational data, such as DCS (Distributed Control System) data, SIS (System-on-a-Plant Monitoring Information System) data, and various sensor data. The emergence of domestically produced intelligent control platforms such as ICS (Intelligent Control Platform) has provided a technological foundation for data-driven equipment health management. Existing technologies include some data-driven equipment fault diagnosis methods, such as fault classification models based on neural networks and support vector machines. However, these methods typically only focus on diagnosis after a fault occurs, lacking dynamic assessment and early warning of the health degradation process; moreover, model training relies on a large number of fault samples, while fault samples are scarce in actual production, resulting in insufficient model generalization ability. Furthermore, existing methods often fail to consider the differences in equipment health benchmarks under different operating conditions, directly applying the same benchmark to all operating conditions, leading to frequent false alarms.

[0005] In summary, there is an urgent need for a health early warning method for thermal power unit equipment that can fully utilize the technological advantages of the ICS platform to achieve multi-source data fusion, adaptive health benchmark construction under operating conditions, multi-dimensional health deviation calculation, and degradation trend prediction. Summary of the Invention

[0006] The present invention aims to overcome the shortcomings of the prior art and provide a data-driven method and device for early warning of the health of thermal power unit equipment, so as to solve the problems of inaccurate health assessment, delayed early warning and poor adaptability in the prior art.

[0007] In a first aspect, the present invention provides a data-driven method for early warning of the health of thermal power unit equipment, the method comprising: Acquire historical operating data and corresponding fault labels of thermal power unit equipment in multiple historical periods, and construct a historical sample dataset; Feature engineering is performed on the historical sample dataset to extract multiple sensitive feature parameters that characterize the health status of the equipment, and a baseline of equipment health is constructed based on the sensitive feature parameters. Acquire real-time operating data of thermal power unit equipment, and extract real-time feature parameters corresponding to the sensitive feature parameters from the real-time operating data; The real-time feature parameters are compared with the device health baseline to calculate the multi-dimensional health deviation of the real-time feature parameters relative to the health baseline. The multi-dimensional health deviations are input into a pre-trained health early warning model, which outputs the current equipment health status score and predicted degradation trend. Based on the current device health status score and the predicted degradation trend, combined with preset graded early warning rules, it is determined whether to trigger an early warning and the level of the early warning.

[0008] Preferably, acquiring real-time operating data of thermal power unit equipment includes: Based on the standardized interface protocol of the ICS intelligent control platform, the OPC UA protocol with a sampling rate greater than 5Hz and a transmission delay of less than 100ms was selected as the data acquisition method. The real-time operating data is collected from the DCS system and third-party sensors based on the aforementioned data acquisition method; The collected real-time running data is preprocessed using the domestically produced GPU acceleration module integrated into the ICS platform.

[0009] Preferably, the feature engineering extraction of the historical sample dataset includes: A bootstrap correction algorithm is used to clean and correct the historical running data in the historical sample dataset. By combining the equipment's mechanism parameters and operating conditions, parameters in the dimensions of vibration, temperature, pressure, and current are extracted from the corrected historical operating data as the sensitive feature parameters.

[0010] Preferably, the step of constructing a device health baseline based on the sensitive feature parameters includes: Based on the sensitive feature parameters and the corresponding equipment operating conditions, the historical sample dataset is divided according to the operating condition type; For each type of operating condition, a health baseline corresponding to that operating condition is constructed for the sensitive characteristic parameters to determine the set of adaptive health baselines for the operating condition. The set of adaptive health baselines for operating conditions is updated online through the containerized environment of the ICS platform.

[0011] Preferably, the health early warning model is a bidirectional decoupled neural network model trained using the sensitive feature parameters and fault labels in the historical sample dataset, based on a data-driven algorithm. The health early warning model is deployed online through the containerized environment of the ICS platform and supports model iteration updates and one-click uploads based on TensorFlow or PyTorch.

[0012] Preferably, calculating the multidimensional health deviation of the real-time feature parameters relative to the health baseline includes: Based on the current operating conditions when the real-time operating data is acquired, a corresponding target health baseline is matched from the set of operating condition adaptive health baselines. The real-time feature parameters are compared with the target health baseline dimension by dimension, and the deviation of each dimension is calculated. The deviations of different dimensions are weighted and fused to obtain a multi-dimensional health deviation vector as the multi-dimensional health deviation.

[0013] Preferably, the preset graded early warning rules include three levels: mild early warning, moderate early warning, and severe early warning; after determining whether to trigger an early warning and the level of warning triggered based on the current equipment health status score and the predicted degradation trend, combined with the preset graded early warning rules, the method further includes: When an alert is triggered, fault mode identification and root cause analysis are performed based on data-driven methods to locate the device or subsystem from which the anomaly originates. The ICS platform's visualization SDK toolkit pushes early warning information, fault analysis results, and maintenance suggestions.

[0014] Secondly, the present invention provides a data-driven thermal power unit equipment health early warning device, comprising: The historical sample dataset construction module is used to obtain historical operating data and corresponding fault labels of thermal power unit equipment in multiple historical periods to construct a historical sample dataset. The equipment health baseline construction module is used to perform feature engineering extraction on the historical sample dataset, determine multiple sensitive feature parameters that characterize the health status of the equipment, and construct the equipment health baseline based on the sensitive feature parameters. The real-time feature parameter extraction module is used to acquire real-time operating data of thermal power unit equipment and extract real-time feature parameters corresponding to the sensitive feature parameters from the real-time operating data. A multi-dimensional health deviation calculation module is used to compare the real-time feature parameters with the device health baseline and calculate the multi-dimensional health deviation of the real-time feature parameters relative to the health baseline. The output module is used to input the multi-dimensional health deviations into a pre-trained health early warning model, and the health early warning model outputs the current equipment health status score and predicted degradation trend. The early warning triggering module is used to determine whether to trigger an early warning and the level of early warning based on the current device health status score and the predicted degradation trend, combined with preset graded early warning rules.

[0015] Thirdly, the present invention provides a readable medium including executable instructions, which, when executed by a processor of an electronic device, cause the electronic device to perform any of the methods described in the first aspect.

[0016] Fourthly, the present invention provides an electronic device including a processor and a memory storing execution instructions, wherein when the processor executes the execution instructions stored in the memory, the processor performs the method as described in any of the first aspects.

[0017] This invention provides a data-driven health early warning method and device for thermal power unit equipment. By constructing a sample dataset integrating multi-source historical data and fault labels, and extracting multi-dimensional sensitive feature parameters such as vibration, temperature, pressure, and current based on a bootstrap correction algorithm and equipment mechanisms, a set of adaptive health baselines is established. This effectively solves the problems of poor adaptability and high false alarm rate of traditional fixed threshold alarms under varying load conditions. By comparing real-time feature parameters with the corresponding baselines in multiple dimensions, a weighted and fused health deviation vector is calculated, and combined with a health early warning model trained based on a bidirectional decoupled neural network. The system outputs equipment health status scores and degradation trends, enabling quantitative assessment and trend prediction of early equipment degradation processes. This overcomes the shortcomings of existing technologies, which can only perform post-fault diagnosis and lack early warning capabilities. By introducing hierarchical early warning rules and combining them with fault mode recognition and root cause analysis after triggering an early warning, the system can accurately locate the source equipment or subsystem of the anomaly. Furthermore, it pushes maintenance suggestions through a visual SDK toolkit, significantly improving the intelligence level of thermal power unit equipment health management, reducing the risk of unplanned shutdowns, and providing strong technical support for the safe and efficient operation of thermal power units in deep peak shaving scenarios and flexible power generation under the "dual carbon" target.

[0018] The further effects of the aforementioned non-conventional preferred method will be explained below in conjunction with specific embodiments. Attached Figure Description

[0019] To more clearly illustrate the embodiments of the present invention or the existing technical solutions, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A schematic diagram of a data-driven early warning method for the health of thermal power unit equipment provided in an embodiment of the present invention; Figure 2 A schematic diagram of another data-driven early warning method for the health of thermal power unit equipment provided in an embodiment of the present invention; Figure 3 A schematic diagram illustrating the composition of a data-driven early warning device for the health of thermal power unit equipment, provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0022] See Figure 1 The image shows a specific embodiment of a data-driven early warning method for the health of thermal power unit equipment provided by the present invention. In this embodiment, the data-driven early warning method for the health of thermal power unit equipment includes:

[0023] Step 101: Obtain historical operating data and corresponding fault labels of thermal power unit equipment in multiple historical periods, and construct a historical sample dataset; Specifically, in this embodiment, historical operating data refers to multi-source time-series data recorded by the distributed control system (DCS), plant-level monitoring information system (SIS), and various online monitoring sensors during the long-term operation of thermal power units. This data covers different operating stages, including unit start-up and shutdown, variable load operation, deep peak shaving, and rated operating conditions. Historical operating data includes at least raw monitoring parameters reflecting the equipment's operating status, such as vibration amplitude, bearing temperature, medium pressure, motor current, flow rate, and power. Fault labels are equipment health status identifiers corresponding to historical operating data. By reviewing equipment maintenance records, fault reports, and operating logs, manual or semi-automatic labeling is performed to determine whether equipment faults occurred during historical operating periods, the type of fault, and the severity of the fault, forming sample labels required for supervised learning. Historical operating data is then associated and aligned with the corresponding fault labels to construct a historical sample dataset containing both normal and fault samples, providing a data foundation for subsequent feature extraction and model training.

[0024] Step 102: Perform feature engineering to extract features from the historical sample dataset, determine multiple sensitive feature parameters that characterize the health status of the equipment, and construct a baseline for equipment health based on the sensitive feature parameters; Furthermore, feature engineering extraction refers to extracting highly correlated feature parameters from raw monitoring data using signal processing, statistical analysis, and other methods. Specifically, this includes: First, using a bootstrap correction algorithm to clean and correct historical operating data in the historical sample dataset. This algorithm is a bias correction method based on resampling technology. Through iterative sampling and bias estimation, it eliminates outliers and noise data caused by factors such as sensor drift, communication interruptions, and environmental interference, while simultaneously filling in short-term data gaps and improving data quality. Second, combining equipment mechanistic parameters and operating conditions, parameters in the dimensions of vibration, temperature, pressure, and current are extracted from the corrected historical operating data as sensitive feature parameters. Equipment mechanistic parameters include the equipment's rated parameters, structural characteristics, and material properties, which guide feature selection; operating conditions include external conditions affecting equipment operating status such as unit load, ambient temperature, and fuel characteristics. Vibration dimension features include frequency amplitude, harmonic amplitude, kurtosis index, etc.; temperature dimension features include bearing temperature, medium temperature, temperature rise rate, etc.; pressure dimension features include inlet and outlet pressure, pressure difference, pressure pulsation, etc.; current dimension features include motor current, power factor, current imbalance, etc.

[0025] The construction of equipment health baselines based on sensitive feature parameters includes: First, dividing the historical sample dataset by operating condition type according to the sensitive feature parameters and corresponding equipment operating conditions. Operating condition type refers to a typical operating range defined by factors such as unit load rate, operating mode, and environmental conditions, such as below 50% load, 50%-80% load, above 80% load, and deep peak shaving. The purpose of dividing operating conditions is to eliminate the impact of operating condition changes on the equipment health baseline and to ensure data comparability within the same operating condition. Second, for each sensitive feature parameter under each operating condition type, constructing the corresponding health baseline for that operating condition to determine the set of adaptive health baselines. The health baseline refers to the statistical distribution range or standard value of the sensitive feature parameters corresponding to a specific operating condition when the equipment is in a healthy state. It can be constructed using methods such as mean, median, confidence interval, and kernel density estimation. The set of adaptive health baselines covers all typical operating conditions of the unit, providing a dynamically matched benchmark for subsequent real-time assessment. Finally, the adaptive health baseline set is iteratively updated online through the containerized environment of the ICS platform. The ICS platform is an intelligent control platform with high availability, low latency data transmission, domestically produced GPU-accelerated computing, and containerized deployment capabilities. The containerized environment includes Docker, Kubernetes, etc., which supports version management, canary releases, and online updates of the baseline model, enabling the health baseline to be continuously optimized as the equipment runs and data accumulates, adapting to the slow degradation or changes in equipment performance.

[0026] Step 103: Obtain real-time operating data of thermal power unit equipment, and extract real-time feature parameters corresponding to sensitive feature parameters from the real-time operating data; Furthermore, acquiring real-time operating data of the thermal power unit equipment specifically includes: First, based on the standardized interface protocol of the ICS intelligent control platform, the OPC UA protocol with a sampling rate greater than 5Hz and a transmission delay of less than 100ms is selected as the data acquisition method. The OPC UA protocol is a cross-platform, secure, and reliable data exchange standard for industrial automation, supporting efficient data acquisition from the DCS system and third-party sensors. A sampling rate greater than 5Hz can capture key characteristics during the dynamic changes of the equipment, and a transmission delay of less than 100ms ensures the real-time nature of the data, meeting the timeliness requirements of online health assessment. Second, real-time operating data is acquired from the DCS system and third-party sensors based on this data acquisition method. In this embodiment, the DCS system is the core control system of the thermal power unit, recording the main parameters of unit operation; third-party sensors include vibration monitoring systems, oil analysis systems, infrared temperature measurement systems, and other supplementary monitoring methods to obtain more comprehensive equipment status information. Finally, the acquired real-time operating data is preprocessed using the domestically produced GPU acceleration module integrated into the ICS platform. This domestically produced GPU acceleration module features strong parallel computing capabilities and low power consumption. It can efficiently perform preprocessing operations such as data filtering, normalization, resampling, and outlier removal, providing clean and well-organized data input for subsequent feature extraction. From the preprocessed real-time running data, based on the definition and extraction method of the sensitive feature parameters determined in step 102, the corresponding real-time feature parameters are extracted to form the device state vector at the current moment.

[0027] Step 104: Compare the real-time feature parameters with the equipment health baseline and calculate the multi-dimensional health deviation of the real-time feature parameters relative to the health baseline. Furthermore, the calculation of multi-dimensional health deviations of real-time characteristic parameters relative to the health baseline includes: First, based on the current operating conditions when acquiring real-time operating data, a corresponding target health baseline is matched from the set of adaptive health baselines. The current operating conditions can be determined online using key parameters such as unit load and main steam pressure to ensure matching to a health baseline under the same or closest operating conditions. Second, the real-time characteristic parameters are compared with the target health baseline dimension by dimension, and the deviation of each dimension is calculated. This deviation can be measured using Euclidean distance, Mahalanobis distance, relative error, Z-score, etc., to reflect the degree of difference between the real-time characteristic parameters and the health baseline. For characteristics of different dimensions such as vibration, temperature, pressure, and current, their respective deviations can be calculated separately. Finally, the deviations of different dimensions are weighted and fused to obtain a multi-dimensional health deviation vector as the multi-dimensional health deviation. The weights for weighting and fusion can be comprehensively determined based on factors such as the sensitivity of each characteristic parameter to the equipment's health status, historical fault data analysis results, and expert experience, so that the fused health deviation vector can comprehensively reflect the overall health deviation of the equipment in multiple dimensions.

[0028] Step 105: Input the multi-dimensional health deviations into the pre-trained health early warning model, and the health early warning model outputs the current equipment health status score and predicted degradation trend; Furthermore, the health early warning model in this embodiment is a data-driven algorithm-based bidirectional decoupled neural network model trained using sensitive feature parameters and fault labels from historical sample datasets. The bidirectional decoupled neural network is a deep learning architecture that integrates forward and backward propagation mechanisms. It can extract high-level features from multi-dimensional health deviations in the input, while separating common degradation features from individual fault features in the device's health status through a decoupling mechanism, thus improving the model's sensitivity to early, subtle faults and its recognition accuracy. During model training, multi-dimensional health deviations from the historical sample dataset are used as input, and the corresponding device health status score (e.g., 0-100 points, with higher scores indicating better health) and degradation trend (e.g., stable, slow degradation, accelerated degradation) are used as output targets. The network parameters are optimized through the backward propagation algorithm. The health early warning model is deployed online through the containerized environment of the ICS platform and supports iterative updates and one-click uploads of models based on TensorFlow or PyTorch. TensorFlow and PyTorch are currently the mainstream deep learning frameworks, supporting rapid model development and training. The one-click upload function allows professional engineers to package and upload optimized models to the ICS platform, enabling smooth model upgrades through a containerized environment and ensuring that the early warning model can be continuously optimized with data accumulation and technological advancements.

[0029] Step 106: Based on the current equipment health status score and predicted degradation trend, and in conjunction with the preset graded early warning rules, determine whether to trigger an early warning and the level of the early warning.

[0030] Furthermore, the preset tiered early warning rules include three levels: mild, moderate, and severe. Each warning level corresponds to a different health status score threshold range and degradation trend combination. For example, when the health status score is above 80 and the degradation trend is stable, no warning is triggered; when the health status score is in the 60-80 range and the degradation trend is slow, a mild warning is triggered; when the health status score is in the 40-60 range or the degradation trend is accelerated, a moderate warning is triggered; and when the health status score is below 40, a severe warning is triggered regardless of the degradation trend. The early warning rules can be configured and adjusted based on factors such as equipment importance, historical failure statistical analysis results, and operation and maintenance strategies.

[0031] As can be seen from the above technical solutions, the beneficial effects of this embodiment are as follows: By constructing an adaptive health benchmark under operating conditions, calculating multi-dimensional health deviations, using bidirectional decoupled neural network prediction and a hierarchical early warning triggering mechanism, dynamic and accurate assessment of the health status of thermal power unit equipment and early trend warning are achieved. This effectively solves the problems of poor adaptability and high false alarm and missed alarm rates of traditional fixed threshold methods under varying operating conditions, and significantly improves the equipment fault prediction capability and the level of intelligent operation and maintenance. Figure 1 The embodiments shown are merely basic examples of the method of the present invention. Other preferred embodiments of the method can be obtained by making certain optimizations and extensions based on them.

[0032] like Figure 2 The image shows another specific embodiment of a data-driven health early warning method for thermal power unit equipment according to the present invention. This embodiment further describes the method based on the foregoing embodiments, and includes the following steps:

[0033] Step 201: When an early warning is triggered, perform fault mode identification and root cause analysis based on data-driven methods to locate the source device or subsystem of the anomaly. Specifically, after determining that an early warning has been triggered, this embodiment further utilizes the data analysis and mining capabilities of the ICS platform to perform fault mode identification and root cause analysis based on data-driven methods, locating the source equipment or subsystem of the anomaly. Specifically, a triggered early warning refers to the warning signal generated in step 106 after determining that the conditions for a graded early warning are met based on the health status score and predicted degradation trend. This signal indicates that the equipment's health status has deviated from the normal range and there is a potential risk of failure. Data-driven refers to using statistical analysis, machine learning, and other algorithms to mine the inherent patterns in historical operating data, fault case libraries, and real-time monitoring data, rather than relying on fixed physical models or expert experience rules. Fault mode identification involves comparing and matching the multi-dimensional health deviation vector, real-time feature parameter sequence, and equipment operating conditions under the current abnormal state with the fault modes already marked in the historical fault case library to determine the fault type to which the current anomaly belongs. Specific methods for fault mode identification include, but are not limited to, cluster analysis, support vector machines, deep residual networks, and attention mechanisms. By calculating the distance or similarity between the current features and the feature centers of various fault modes, several candidate fault modes with the highest probability are output. Root cause analysis, based on the identified failure mode, traces the root cause of equipment health degradation and pinpoints the specific equipment or subsystem from which the anomaly originates. Root cause analysis can employ methods such as causal inference, association rule mining, fault tree analysis, and propagation entropy. By analyzing the propagation path and impact intensity of abnormal characteristic parameters within the equipment's interconnected network, it identifies the key nodes triggering cascading anomalies. For example, when a fan bearing temperature rises accompanied by increased vibration and current fluctuations, root cause analysis can pinpoint the anomaly as poor bearing lubrication or wear, rather than an electrical fault in the motor. The source equipment or subsystem refers to the specific equipment unit in a thermal power unit experiencing performance degradation or early failure, such as the coal mill, forced draft fan, induced draft fan, feedwater pump, high-pressure heater, condenser, generator stator winding, or its associated functional subsystems, such as the pulverizing system, flue gas system, steam-water system, and electrical system. Accurately locating the source of the anomaly provides clear direction for subsequent maintenance decisions, avoiding blind troubleshooting and wasted resources.

[0034] Step 202: Push early warning information, fault analysis results and maintenance suggestions through the visualization SDK toolkit of the ICS platform.

[0035] Furthermore, in this embodiment, the ICS platform refers to the intelligent control platform upon which this method relies. It possesses capabilities such as high availability, low-latency data transmission, domestically produced GPU-accelerated computing, and containerized deployment, providing a basic operating environment for each step of this invention. The visualization SDK toolkit refers to a software development kit integrated on the ICS platform, providing a series of functional modules for data visualization, interactive interface design, message push, and report generation. This toolkit supports presenting complex analysis results in an intuitive and easy-to-understand graphical form, such as equipment health status trend charts, multi-dimensional health deviation radar charts, fault mode matching probability distribution charts, and root cause propagation path topology charts. Simultaneously, the SDK toolkit has communication interfaces with operator terminals (such as operator stations, mobile terminals, and large-screen display systems), supporting real-time push of early warning information and diagnostic results. Early warning information includes the trigger time, warning level (mild / moderate / severe), name of the warning device or subsystem, current health status score, and description of degradation trends, aiming to inform operators of equipment anomalies as soon as possible. The fault analysis results include the fault modes identified in step 201 (such as "early bearing wear", "rotor imbalance", "decreased cooling efficiency", etc.), root cause location results (such as "#2 blower drive end bearing"), and relevant evidence parameters (such as changes in vibration octave components, temperature rise rate, current fluctuation amplitude, etc.), used to support operators' judgment on the degree of trust in the diagnostic conclusions. Maintenance recommendations refer to targeted guidance suggestions generated based on the fault modes and root cause analysis results, combined with equipment maintenance manuals, expert experience databases, and similar case handling records. Maintenance recommendations may include short-term measures (such as adjusting operating parameters, increasing monitoring frequency) and long-term measures (such as arranging shutdown for maintenance, replacing spare parts, optimizing operating modes), and can be marked with priority and completion deadlines according to the warning level. The above information is integrated and pushed through the ICS platform's visualization SDK toolkit, realizing a closed-loop information process from status perception, health assessment, warning triggering, fault diagnosis to maintenance guidance, effectively supporting the intelligent and precise decision-making of thermal power unit equipment operation and maintenance.

[0036] This invention also provides a data-driven health early warning device for thermal power unit equipment. See also Figure 3 The image shows a specific embodiment of a data-driven health early warning device for thermal power unit equipment provided by the present invention. This embodiment of the device is used to execute... Figures 1-2 The physical apparatus of the method. Its technical solution is essentially the same as the above embodiments, and the corresponding descriptions in the above embodiments also apply to this embodiment. The apparatus includes:

[0037] The historical sample dataset construction module 301 is configured to acquire historical operating data and corresponding fault labels of thermal power unit equipment in multiple historical periods to construct a historical sample dataset. The equipment health baseline construction module 302 is configured to perform feature engineering extraction on historical sample datasets, determine multiple sensitive feature parameters characterizing the health status of equipment, and construct the equipment health baseline based on the sensitive feature parameters. The real-time feature parameter extraction module 303 is configured to acquire real-time operating data of thermal power unit equipment and extract real-time feature parameters corresponding to sensitive feature parameters from the real-time operating data. The multi-dimensional health deviation calculation module 304 is configured to compare real-time feature parameters with the device health baseline and calculate the multi-dimensional health deviation of the real-time feature parameters relative to the health baseline. The output module 305 is configured to input multi-dimensional health deviations into a pre-trained health warning model, and the health warning model outputs the current device health status score and predicted degradation trend. The early warning triggering module 306 is configured to determine whether to trigger an early warning and the level of the early warning based on the current equipment health status score and predicted degradation trend, combined with preset graded early warning rules.

[0038] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. The memory may include main memory, such as high-speed random-access memory (RAM), or it may also include non-volatile memory, such as at least one disk storage device. Of course, the electronic device may also include other hardware required for other services.

[0039] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. Buses can be categorized as address buses, data buses, and other types. For ease of representation, Figure 4 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0040] Memory is used to store instructions for execution. Specifically, instructions for execution are computer programs that can be executed. Memory can include main memory and non-volatile memory, and it provides the processor with execution instructions and data.

[0041] In one possible implementation, the processor reads the corresponding execution instructions from non-volatile memory into main memory and then executes them. Alternatively, it may obtain the corresponding execution instructions from other devices to logically form a data-driven thermal power unit equipment health early warning device. The processor executes the execution instructions stored in the memory to implement a data-driven thermal power unit equipment health early warning method provided in any embodiment of the present invention.

[0042] The above is as described in the present invention. Figure 3 The method for executing a data-driven health early warning device for thermal power unit equipment provided in the illustrated embodiment can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor.

[0043] The steps of the method disclosed in the embodiments of this invention can be directly manifested as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0044] This invention also proposes a readable medium storing execution instructions. When these instructions are executed by a processor of an electronic device, the device can perform a data-driven thermal power unit equipment health early warning method provided in any embodiment of this invention, specifically for executing, for example... Figure 1 , Figure 2 The method shown.

[0045] The electronic devices in the foregoing embodiments may be computers.

[0046] Those skilled in the art will understand that embodiments of the present invention can be provided as methods or computer program products. Therefore, the present invention can be implemented in a completely hardware embodiment, a completely software embodiment, or a combination of software and hardware.

[0047] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0048] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0049] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A data-driven method for early warning of equipment health in thermal power units, characterized in that, The method includes: Acquire historical operating data and corresponding fault labels of thermal power unit equipment in multiple historical periods, and construct a historical sample dataset; Feature engineering is performed on the historical sample dataset to extract multiple sensitive feature parameters that characterize the health status of the equipment, and a baseline of equipment health is constructed based on the sensitive feature parameters. Acquire real-time operating data of thermal power unit equipment, and extract real-time feature parameters corresponding to the sensitive feature parameters from the real-time operating data; The real-time feature parameters are compared with the device health baseline to calculate the multi-dimensional health deviation of the real-time feature parameters relative to the health baseline. The multi-dimensional health deviations are input into a pre-trained health early warning model, which outputs the current equipment health status score and predicted degradation trend. Based on the current device health status score and the predicted degradation trend, combined with preset graded early warning rules, it is determined whether to trigger an early warning and the level of the early warning.

2. The method according to claim 1, characterized in that, The acquisition of real-time operating data of thermal power unit equipment includes: Based on the standardized interface protocol of the ICS intelligent control platform, the OPC UA protocol with a sampling rate greater than 5Hz and a transmission delay of less than 100ms was selected as the data acquisition method. The real-time operating data is collected from the DCS system and third-party sensors based on the aforementioned data acquisition method; The collected real-time running data is preprocessed using the domestically produced GPU acceleration module integrated into the ICS platform.

3. The method according to claim 1, characterized in that, The feature engineering extraction of the historical sample dataset includes: A bootstrap correction algorithm is used to clean and correct the historical running data in the historical sample dataset; By combining the equipment's mechanism parameters and operating conditions, parameters in the dimensions of vibration, temperature, pressure, and current are extracted from the corrected historical operating data as the sensitive feature parameters.

4. The method according to claim 1, characterized in that, The construction of the device health baseline based on the sensitive feature parameters includes: Based on the sensitive feature parameters and the corresponding equipment operating conditions, the historical sample dataset is divided according to the operating condition type; For each type of operating condition, a health baseline corresponding to that operating condition is constructed for the sensitive characteristic parameters to determine the set of adaptive health baselines for the operating condition. The set of adaptive health baselines for operating conditions is updated online through the containerized environment of the ICS platform.

5. The method according to claim 1, characterized in that, The health early warning model is a bidirectional decoupled neural network model trained using the sensitive feature parameters and fault labels in the historical sample dataset, based on a data-driven algorithm. The health early warning model is deployed online through the containerized environment of the ICS platform and supports model iteration updates and one-click uploads based on TensorFlow or PyTorch.

6. The method according to claim 1, characterized in that, The calculation of the multidimensional health deviation of the real-time feature parameters relative to the health baseline includes: Based on the current operating conditions when the real-time operating data is acquired, a corresponding target health baseline is matched from the set of operating condition adaptive health baselines. The real-time feature parameters are compared with the target health baseline dimension by dimension, and the deviation of each dimension is calculated. The deviations of different dimensions are weighted and fused to obtain a multi-dimensional health deviation vector as the multi-dimensional health deviation.

7. The method according to claim 1, characterized in that, The preset graded early warning rules include three levels: mild early warning, moderate early warning, and severe early warning. After determining whether to trigger an early warning and the level of warning triggered based on the current equipment health status score and the predicted degradation trend, combined with the preset graded early warning rules, the process further includes: When an alert is triggered, fault mode identification and root cause analysis are performed based on data-driven methods to locate the device or subsystem from which the anomaly originates. The ICS platform's visualization SDK toolkit pushes early warning information, fault analysis results, and maintenance suggestions.

8. A data-driven health early warning device for thermal power unit equipment, characterized in that, include: The historical sample dataset construction module is used to obtain historical operating data and corresponding fault labels of thermal power unit equipment in multiple historical periods to construct a historical sample dataset. The equipment health baseline construction module is used to perform feature engineering extraction on the historical sample dataset, determine multiple sensitive feature parameters that characterize the health status of the equipment, and construct the equipment health baseline based on the sensitive feature parameters. The real-time feature parameter extraction module is used to acquire real-time operating data of thermal power unit equipment and extract real-time feature parameters corresponding to the sensitive feature parameters from the real-time operating data. A multi-dimensional health deviation calculation module is used to compare the real-time feature parameters with the device health baseline and calculate the multi-dimensional health deviation of the real-time feature parameters relative to the health baseline. The output module is used to input the multi-dimensional health deviations into a pre-trained health early warning model, and the health early warning model outputs the current equipment health status score and predicted degradation trend. The early warning triggering module is used to determine whether to trigger an early warning and the level of early warning based on the current device health status score and the predicted degradation trend, combined with preset graded early warning rules.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the method of any one of claims 1 to 7.

10. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the method described in any one of claims 1 to 7.

Images