Method and system for generating automatic report of power plant in credential and credential environment based on AI

Abstract

The invention relates to the technical field of power plant report generation methods, in particular to an AI-based power plant automatic report generation method and system in a credential environment. According to the method, operation parameters, environment parameters and archive data of power plant equipment are acquired in real time through data acquisition equipment, multi-dimensional feature extraction is performed by using a lightweight neural network, and a feature data packet including a time sequence feature matrix, a working condition feature vector and data quality identification information is generated. And based on the feature data packet, an improved streaming clustering algorithm is adopted to dynamically divide equipment operation condition modes and establish a health state reference benchmark, an adaptive threshold range is generated by combining environment deviation compensation and health attenuation calculation, and accurate assessment of the equipment health state is realized. And generating a dynamic health reference data packet through time sequence feature analysis and working condition feature processing. According to the invention, full-process intelligent processing from data acquisition to report generation is realized, and the accuracy and operation and maintenance efficiency of power plant equipment health monitoring are remarkably improved.

Description

Technical Field

[0001] This invention relates to the technical field of power plant report generation methods, and more specifically, to an AI-based method and system for automatically generating power plant reports in a domestically developed information technology environment. Background Technology

[0002] With the deepening of the information technology application innovation strategy in the energy industry, the intelligent transformation of power plants has placed higher demands on independent and controllable technologies. Against this backdrop, AI-based automated reporting systems are gradually becoming a key technological means to improve power plant operational efficiency. Especially in the field of equipment health monitoring, existing technologies typically employ rule-based conditional judgment methods to achieve anomaly detection: by pre-defining fixed threshold ranges under typical operating conditions, real-time collected operating parameters are compared with preset thresholds to generate equipment status analysis reports.

[0003] While this fixed-threshold-based analysis method simplifies the report generation process to some extent, it still has significant shortcomings in adaptability to actual operating conditions. Specifically, the threshold settings lack dynamic adjustment capabilities. Because equipment operating status continuously changes with environmental conditions, usage time, and other factors, fixed judgment criteria cannot accurately reflect the true health level of the equipment, leading to false or missed reports in the generated reports. This contradiction between the static analysis method and the dynamic operating environment seriously affects the accuracy of equipment status assessment. Summary of the Invention

[0004] This invention provides an AI-based method and system for automatically generating reports in power plants under a domestically developed information technology environment. It constructs a data package of equipment operation features through multi-dimensional data collection and feature extraction, and establishes a dynamic health assessment model based on this data package using a streaming clustering algorithm. This enables adaptive threshold generation and accurate assessment of equipment health status, thereby solving the problems mentioned in the background art.

[0005] In power plant equipment health monitoring, false alarms and missed alarms are caused by fixed thresholds that cannot dynamically adapt to changes in equipment status.

[0006] To achieve the above objectives, the report generation method includes the following steps:

[0007] S1. Collect equipment operating parameters, environmental parameters and equipment file data, and extract features from them to generate a feature data package containing a time-series feature matrix, operating condition feature vector and data quality identification information;

[0008] S2. Construct a dynamic health assessment model based on feature data packets. The steps are as follows:

[0009] S2.1 Analyze the operating condition feature vectors using a streaming clustering algorithm, classify the equipment operating condition modes, establish health status reference benchmarks for each operating condition, and generate operating condition quality identifiers.

[0010] S2.2. Use the time series feature matrix to train the time series anomaly detection model and output the real-time health status score of the device;

[0011] S2.3. Combine data quality identification information with health status reference benchmarks to calculate and generate dynamic threshold ranges;

[0012] S2.4. Integrate real-time health status scores, dynamic threshold ranges, and operating condition quality indicators to generate dynamic health benchmark data packages and assess the health status of power plant equipment.

[0013] The design concept of the above technical solution stems from an understanding of the unique characteristics of power plant equipment health monitoring: traditional fixed threshold methods cannot cope with the complex and variable operating conditions of equipment and the nonlinear characteristics of performance degradation. Specifically, the use of streaming clustering algorithms (rather than static operating condition classification) can dynamically adapt to equipment load fluctuations and environmental changes. Without this design, the system will be unable to identify transitional operating conditions, leading to misjudgments. The introduction of data quality identification (rather than simple data cleaning) can distinguish between sensor anomalies and actual equipment failures. The lack of this mechanism will reduce the reliability of anomaly detection. The design that integrates health status scoring, dynamic thresholds, and operating condition quality (rather than judging by a single indicator) solves the problem of insufficient threshold adaptability of traditional methods in the equipment aging stage through multi-dimensional cross-validation. This architectural design ensures that the system can capture both instantaneous anomalies in equipment status and track long-term performance degradation trends, realizing a paradigm upgrade from "static threshold alarm" to "dynamic health management".

[0014] Based on this, the streaming clustering algorithm adopts adaptive neighborhood radius technology to dynamically adjust the clustering sensitivity according to the stability of the device's operating status.

[0015] In another technical solution, the calculation of the dynamic threshold range incorporates a dynamic compensation mechanism for environmental deviation and a nonlinear calculation method for health degradation.

[0016] This technical solution overcomes the contradiction between accuracy and stability in traditional fixed-radius clustering by employing adaptive neighborhood radius technology. When equipment operation fluctuates significantly, automatically expanding the neighborhood radius avoids invalid operating condition classifications due to instantaneous fluctuations (such as during unit start-up and shutdown transitions). Conversely, reducing the radius during stable operation allows for the identification of subtle differences in operating conditions. The dynamic environmental deviation compensation mechanism (compared to traditional fixed environmental coefficients) addresses the issue of poor threshold applicability for the same equipment in different seasons and regions by real-time sensing of changes in environmental parameters such as temperature and humidity. Furthermore, the nonlinear calculation of health degradation (unlike linear lifespan models) more accurately depicts the objective law of exponential decline in equipment performance over time. Using a simple linear model would severely underestimate the performance degradation rate of aging equipment.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0018] 1. This invention constructs a closed-loop processing flow of "data acquisition - feature extraction - dynamic modeling - intelligent decision-making," achieving a three-dimensional assessment of equipment health status through deep coupling of each stage. Unlike traditional single-point threshold detection methods, this solution organically combines time-series feature analysis, operating condition pattern recognition, and quality assessment, forming a dynamic assessment system with self-learning capabilities. This enables the system to continuously optimize the health assessment model and gradually improve prediction accuracy.

[0019] 2. This invention innovatively integrates the adaptation to the domestic IT environment with the design of intelligent algorithms. Through optimization methods such as lightweight network design and hybrid precision computing, it ensures the accuracy of the algorithm while meeting the computing power constraints of domestic hardware platforms. This achieves a deep integration of independent and controllable technology with advanced intelligent algorithms, providing safe and reliable technical support for the intelligent operation and maintenance of key equipment in the energy industry. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall process structure of the AI-based automatic report generation method for power plants in the context of information technology innovation.

[0021] Figure 2 This is a schematic diagram of the specific process of step S1 of the present invention;

[0022] Figure 3 This is a schematic diagram of the specific process of step S2 of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Currently, in power plant equipment health monitoring, the inability of fixed thresholds to dynamically adapt to changes in equipment status leads to false alarms and missed alarms. One objective of this invention is to provide an AI-based method for automatically generating power plant reports in a domestically developed information technology environment. (See [link to relevant documentation]). Figure 1 As shown, it includes the following steps:

[0025] S1. Collect equipment operating parameters, environmental parameters and equipment file data, and extract features from them to generate a feature data package containing a time-series feature matrix, operating condition feature vector and data quality identification information;

[0026] S2. Construct a dynamic health assessment model based on feature data packets. The steps are as follows:

[0027] S2.1 Analyze the operating condition feature vectors using a streaming clustering algorithm, classify the equipment operating condition modes, establish health status reference benchmarks for each operating condition, and generate operating condition quality identifiers.

[0028] S2.2. Use the time series feature matrix to train the time series anomaly detection model and output the real-time health status score of the device;

[0029] S2.3. Combine data quality identification information with health status reference benchmarks to calculate and generate dynamic threshold ranges;

[0030] S2.4. Integrate real-time health status scores, dynamic threshold ranges, and operating condition quality indicators to generate dynamic health benchmark data packages and assess the health status of power plant equipment.

[0031] By adopting dynamic health benchmark modeling to achieve adaptive adjustment of thresholds and performing anomaly detection based on multi-source data fusion analysis, an intelligent report containing credibility assessment is finally generated, realizing accurate assessment and visualization of the health status of power plant equipment in the information technology innovation environment.

[0032] See Figure 2 As shown, step S1 of this invention involves real-time acquisition of multi-dimensional operational data and extraction of operating condition features. Specifically, the primary task of this step is to achieve comprehensive monitoring of the operating status of key power plant equipment through a data acquisition system adapted to the domestic IT environment. This acquisition system uses domestically produced hardware and a dedicated communication protocol to ensure safe and stable operation in the domestic IT environment.

[0033] For acquiring equipment operating parameters, the system obtains real-time monitoring data through a network of intelligent sensors installed on key equipment. Vibration data is collected by accelerometers deployed on the equipment bearing housings and casing. These sensors utilize MEMS technology and support synchronous measurement of triaxial vibration signals. Temperature monitoring is achieved through an array of temperature sensors embedded in key parts of the equipment. These sensors employ industrial-grade PT100 platinum resistance thermometers, and data conversion is performed by a signal conditioning unit. Current parameters are acquired through high-precision current transformers, which can capture the current waveform characteristics of the equipment in real time. All operating parameters are transmitted to edge computing nodes via an industrial fieldbus, where preliminary data verification and time-scale alignment are performed within the nodes.

[0034] The environmental parameter acquisition system is implemented by environmental monitoring terminals deployed around the equipment. Each terminal integrates temperature and humidity sensors, air pressure sensors, and noise sensors, forming a distributed monitoring network through wireless self-organizing network technology. The system employs a time-division multiplexing mechanism to ensure the synchronous acquisition of environmental data and equipment operation data. The environmental monitoring terminals have built-in low-power processors that can preprocess the raw data before transmitting it to the central acquisition server via an encrypted channel.

[0035] Equipment data is obtained by connecting to the power plant's existing asset management system. A dedicated data interface is developed for the system, establishing a secure connection with the database to periodically extract static parameters and maintenance records of the equipment. Key information collected includes equipment model, commissioning date, cumulative runtime, and historical maintenance records. To ensure data timeliness, the system is equipped with an automatic trigger mechanism that immediately updates the relevant data when equipment status changes or maintenance operations occur.

[0036] The rigorously validated multi-source data then proceeds to the feature extraction stage. The system employs a lightweight neural network model based on an AI framework to perform deep feature mining on the collected equipment operating parameter matrix, environmental parameter vector, and equipment archive dataset. Specifically, the equipment operating parameter matrix is ​​used to extract a time-series feature matrix through a time-frequency analysis network, while the environmental parameter vector and equipment archive data are processed by a feature fusion network to generate a condition feature vector. Both types of data are then input into a quality assessment subnetwork, which outputs data quality identification information. This model can automatically identify and extract engineering-significant feature parameters, including key indicators such as the spectral characteristics of vibration signals, the trend characteristics of temperature changes, and the distortion characteristics of current waveforms. Simultaneously, the system constructs a comprehensive feature vector characterizing the equipment's operating conditions by comprehensively analyzing multiple dimensions such as runtime, load rate, and environmental factors.

[0037] During feature extraction, the system places particular emphasis on the engineering applicability of features and their compatibility with subsequent modules. All extracted feature parameters are standardized and accompanied by complete timestamps and source identifiers to ensure traceability to the original collected data. The system's built-in quality control module monitors the integrity of the feature data in real time, automatically marking and compensating for abnormal feature values. Finally, the rigorously quality-verified feature dataset is packaged into a standardized feature data package. This package contains a normalized time-series feature matrix, operating condition feature vectors, and data quality identification information, and is transmitted via an encrypted data channel to the subsequent dynamic health benchmark modeling module, providing reliable input data for adaptive threshold adjustment.

[0038] like Figure 2 As shown, step S2 of this invention is dynamic health baseline modeling and threshold generation. Specifically, this step constructs a dynamically changing health assessment baseline and adaptive threshold based on the time-series feature matrix, operating condition feature vector, and data quality identification information passed from the previous step. The system first analyzes the operating condition feature vector and automatically identifies different operating condition modes of the equipment through an improved streaming clustering algorithm. This algorithm uses a sliding window mechanism to process real-time data streams and performs dynamic clustering by calculating feature similarity. It can automatically create new operating condition clusters or merge similar operating condition clusters, while periodically cleaning up outdated operating condition clusters that have not been updated for a long time to ensure the timeliness and accuracy of operating condition classification. In particular, the clustering algorithm of this invention introduces an adaptive neighborhood radius technology, which can automatically adjust the clustering sensitivity according to the stability of the equipment's operating status. When the equipment's operating status fluctuates greatly, the neighborhood radius is automatically expanded to avoid generating too many meaningless operating condition classifications; when the equipment's operation is stable, the neighborhood radius is reduced to improve the precision of operating condition identification. A corresponding health status reference baseline is established for each operating condition mode. This process fully considers the differences in the operating characteristics of equipment under different loads and environmental conditions, ensuring the accuracy of health benchmarks.

[0039] During the dynamic threshold generation phase, the system employs an innovative calculation method, specifically determining the real-time changing threshold range using the following formula:

[0040] T = μ + k(σ + αE + βH);

[0041] In the formula, T represents the dynamic threshold;

[0042] μ represents the historical average health data;

[0043] k is the adjustment coefficient (range 1.5-3.0);

[0044] σ represents the standard deviation of historical health data;

[0045] α represents the environmental factor weighting coefficient (values ​​range from 0.1 to 0.3);

[0046] E represents the environmental deviation degree (0-1 standardized value);

[0047] β represents the health decay weighting coefficient (value ranges from 0.05 to 0.2);

[0048] H represents the amount of health decline.

[0049] This calculation method comprehensively considers the statistical characteristics of historical health data, the specific impact of the current operating environment, and the cumulative wear and tear of the equipment, enabling it to automatically adapt to changes in equipment status. Compared with traditional fixed threshold methods, this dynamic adjustment mechanism has three significant innovations: First, it introduces a dynamic compensation mechanism for environmental deviation; when a sudden change in environmental conditions is detected, the system automatically increases the weight coefficient of environmental factors, making the threshold more sensitive to environmental changes. Second, the health degradation is calculated using a non-linear method, which can more accurately reflect the performance degradation pattern during the equipment aging process. Finally, all coefficients have online self-optimization capabilities, automatically adjusting parameter values ​​based on actual operating results. This dynamic adjustment mechanism significantly improves the accuracy of equipment health status assessment.

[0050] The system employs a dual-channel collaborative architecture to achieve the aforementioned functions. The first channel analyzes the time-series feature matrix, extracts the time-series features of equipment operation through a deep neural network, and outputs the current health status score. This neural network uses a special residual attention structure, capable of simultaneously capturing both long-term trends and short-term fluctuations in equipment operation. The second channel processes the operating condition feature vector, combines it with data quality identification information, and calculates the most suitable threshold adjustment strategy for the current operating condition. This channel innovatively introduces an operating condition memory pool mechanism, which can store and recall processing experience from similar operating conditions in the past, significantly improving the rationality of threshold calculation. The outputs of the two channels are intelligently fused at the decision-making level, generating a dynamic health benchmark data package containing three core indicators through a credibility-weighted algorithm: a real-time health score (generated by the time-series analysis channel), a dynamic threshold range (calculated using a threshold formula), and an operating condition quality identifier (generated during the operating condition clustering process). This dual-channel architecture ensures both the real-time performance of the calculations and the scientific rigor of the decision-making process.

[0051] During model operation, the system continuously monitors data quality indicators. When data anomalies or missing data are detected, a three-tiered data compensation mechanism is automatically activated: for short-term minor anomalies, sliding window mean compensation is used; for persistent anomalies, historical data compensation based on similar operating conditions is switched to; and for severe anomalies, an expert knowledge base is activated to assist decision-making. Simultaneously, the system possesses intelligent online learning capabilities, employing a combination of incremental learning and transfer learning. It periodically fine-tunes model parameters using the latest collected data and can quickly transfer learning experience from other similar devices when significant changes in operating modes are detected. All computations are performed on an AI-accelerated chip, with a dedicated computational pipeline design ensuring high efficiency and real-time performance.

[0052] The dynamic health baseline data package generated in this step, after undergoing multiple encryption and verification processes, is then passed to the multi-source data fusion and analysis module in the next step. The dynamic health baseline data package effectively solves the threshold rigidity problem in the following ways: by using real-time updated operating condition feature vectors and environmental deviation parameters, the threshold can automatically adapt to changes in the equipment's operating environment; by calculating health decay, the threshold can dynamically adjust along with the equipment's aging process; and by continuously improving each weight coefficient through an online self-optimization mechanism, ensuring that the threshold always remains in an optimal state. This dynamic adjustment mechanism completely overcomes the shortcomings of traditional fixed thresholds, which cannot adapt to changes in equipment status. The entire process establishes a complete digital twin archive, recording in detail the decision-making basis for all threshold adjustments, the parameter change process, and the learning evolution trajectory, providing comprehensive data support for subsequent intelligent operation and maintenance.

[0053] Step S3 of this invention involves multi-source data fusion analysis and anomaly decision-making. Specifically, the core task of this step is to perform in-depth analysis and anomaly judgment on the dynamic health benchmark data packet transmitted in step S2. The system first establishes a data preprocessing channel to clean and verify the received real-time health scores, dynamic threshold ranges, and operating condition quality indicators. The cleaning process includes outlier removal, data smoothing, and missing value compensation to ensure the quality of the analyzed data. The verification stage uses a cross-validation algorithm to check the consistency between different data sources and mark suspicious data points.

[0054] During the data analysis phase, the system employs a three-tiered analysis architecture. The first tier analyzes real-time health scores, using an improved sliding window algorithm to detect short-term abnormal fluctuations. This algorithm adaptively adjusts the window size, using a larger window to smooth random fluctuations when the equipment is stable, and automatically shrinking the window to improve detection sensitivity when the status changes drastically. The second tier handles dynamic threshold ranges, establishing a deviation assessment model based on fuzzy logic. This model not only determines whether parameters exceed thresholds but also quantifies the degree of exceedance and the trend of change, outputting a deviation level and trend score. The third tier focuses on operational condition quality indicators, implementing rule-based logical verification and correlation analysis to identify potential sensor faults or data transmission problems.

[0055] After the above analysis and processing, this step finally generates an equipment anomaly diagnostic report. This report contains four core components: anomaly level assessment (divided into three levels: Attention, Warning, and Emergency), anomaly type classification (including mechanical failure, electrical failure, environmental anomalies, etc.), root cause analysis results (identifying key parameters leading to the anomaly and their impact paths), and preliminary handling recommendations (preferred handling measures based on the anomaly characteristics). This anomaly diagnostic report is transmitted to the intelligent report generation module in step S4 via an encrypted data channel, providing professional diagnostic basis for the generation of the final report.

[0056] Step S4 of this invention is intelligent report generation and output. Specifically, this step, as the final stage of this embodiment, is responsible for transforming the analysis results of the preceding steps into structured, operable intelligent reports. The system first establishes a report generation engine, which consists of three core components: a data integrator, a content generator, and a visualization renderer.

[0057] The data integrator receives the equipment anomaly diagnosis report from step S3 and the dynamic health baseline data package from step S2, and performs deep data fusion. This component first performs structured parsing of the anomaly level, type, root cause analysis, and handling recommendations in the anomaly diagnosis report to establish an anomaly event knowledge graph. Simultaneously, it extracts equipment health score curves, parameter change trends, and operating condition characteristics from the dynamic health baseline data package to construct a time-series database of equipment status. These two data sources are precisely correlated through unique equipment identifiers and timestamps to form a complete analysis dataset.

[0058] The content generator operates based on a predefined report template system. The system maintains a template library containing 12 basic templates, covering periodic reports such as daily, weekly, and monthly reports, as well as special report types such as emergency reports and special analysis reports. The content generator automatically selects the most suitable template based on the anomaly level and operational needs, and then performs intelligent population: converting the anomaly event knowledge graph into textual descriptions, automatically generating professional analysis content including root causes, scope of impact, and evolution trends; extracting key indicators from the equipment status time-series database to generate data summaries and trend predictions; and expanding handling suggestions into specific operational steps and precautions based on power plant operation and maintenance procedures.

[0059] The visualization renderer is responsible for transforming structured data into intuitive visual presentations. This component employs a multi-layered visualization strategy: the top layer displays the overall health status of the device through a five-level warning light and a health score curve; the middle layer shows the anomaly propagation path and key influencing parameters through a parameter relationship network diagram; and the bottom layer provides detailed parameter change curves and threshold comparison charts. All visualization elements support interactive operations, allowing users to drill down and view detailed data. The system also innovatively introduces AR visualization assistance, allowing users to view operating parameters and anomaly prompts overlaid on the real-world device on their mobile devices by scanning the device's QR code.

[0060] The final generated intelligent report contains five standard units: the execution summary unit provides core conclusions and action recommendations; the data analysis unit displays the detailed diagnostic process and evidence; the trend prediction unit predicts equipment status changes over the next 24 hours; the operation and maintenance guidance unit provides step-by-step operation guidelines; and the appendix unit contains the original data index and technical specifications. Before outputting the report, the system automatically performs a three-level quality check: format verification ensures report standardization, logical verification ensures content consistency, and security verification confirms data integrity. Reports that pass verification are encrypted using national cryptographic algorithms and digitally signed before being simultaneously output through three channels: pushed to the main database of the power plant production management system, sent to the mobile terminals of relevant responsible persons, and archived in the distributed document management system.

[0061] This invention solves the problem that traditional fixed threshold methods cannot adapt to dynamic changes in equipment status by constructing a dynamic health assessment model, realizing adaptive threshold adjustment, innovating multi-source data fusion analysis methods, and intelligent report generation technology. It achieves full-process intelligent management of power plant equipment health status, including accurate monitoring, intelligent diagnosis, and automated report generation in a domestically developed information technology environment.

[0062] The second objective of this embodiment is to provide an AI-based automatic report generation system for power plants under a domestically developed information technology environment. This system includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement all the steps described in the foregoing method embodiments. Specifically, this system includes the following four core components.

[0063] The data acquisition and feature extraction module is responsible for acquiring real-time operating status information of power plant equipment. This module has built-in data acquisition hardware and connects to various sensors and monitoring devices via dedicated interfaces, enabling simultaneous acquisition of key parameters such as vibration, temperature, and current. The module integrates a data cleaning and quality control unit to handle outliers and compensate for missing values ​​in the raw data, ensuring the reliability of subsequent analysis. After standardization, the acquired data is fed into the feature extraction engine, which uses a lightweight neural network model to automatically extract engineering-significant feature parameters and operating condition feature vectors.

[0064] The dynamic health benchmark modeling module is the core computing unit of this system. Running on an AI-accelerated chip, it comprises two main components: a working condition clustering analyzer and a dynamic threshold calculator. The working condition clustering analyzer, based on an improved streaming clustering algorithm, identifies the operating condition patterns of equipment in real time. The dynamic threshold calculator, through an innovative calculation formula, generates an adaptive threshold range that changes with the equipment's state. The module integrates an online learning mechanism, enabling continuous optimization of model parameters based on the latest data.

[0065] The multi-source data fusion analysis module receives the processing results from the preceding modules and performs comprehensive analysis and anomaly detection. This module comprises three parallel analysis units: a time-series feature analysis unit to detect short-term abnormal fluctuations, a threshold deviation assessment unit to quantify the degree of parameter anomalies, and a condition quality analysis unit to identify data acquisition problems. The outputs of each analysis unit are used to generate a diagnostic report containing anomaly level, type, and root cause analysis through an intelligent fusion algorithm.

[0066] The intelligent report generation and output module is the system's final output interface. This module comprises three functional layers: template management, content generation, and visualization rendering. The template management layer maintains a standardized report template library and can automatically match the most suitable template framework based on the anomaly type. The content generation layer transforms analysis results into structured report content, automatically generating professional technical analysis and operation and maintenance suggestions. The visualization rendering layer is responsible for converting data into intuitive charts and graphs, supporting various interactive display methods. The final generated report, after encryption and digital signature, is output to the power plant management system through a secure interface.

[0067] The system adopts a layered architecture, with modules communicating via a secure data bus. All computations are performed on a hardware platform within a domestically developed IT environment. A dynamic resource scheduling mechanism is specifically designed to automatically adjust computing resource allocation based on the processing load, ensuring the real-time requirements of critical tasks. Simultaneously, the system incorporates comprehensive security auditing functions, recording all data processing and system operations to meet the stringent data security and operational traceability requirements of power plants. Through the collaborative work of these modules, the system achieves full automation and intelligence in the monitoring of power plant equipment health status and report generation.

[0068] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for automatically generating power plant reports based on AI in a domestically developed information technology environment, characterized in that, Includes the following steps: S1. Collect equipment operating parameters, environmental parameters and equipment file data, and extract features from them to generate a feature data package containing a time-series feature matrix, operating condition feature vector and data quality identification information; S2. Construct a dynamic health assessment model based on feature data packets. The steps are as follows: S2.1 Analyze the operating condition feature vectors using a streaming clustering algorithm, classify the equipment operating condition modes, establish health status reference benchmarks for each operating condition, and generate operating condition quality identifiers. S2.

2. Use the time series feature matrix to train the time series anomaly detection model and output the real-time health status score of the device; S2.

3. Combine data quality identification information with health status reference benchmarks to calculate and generate dynamic threshold ranges; S2.

4. Integrate real-time health status scores, dynamic threshold ranges, and operating condition quality indicators to generate dynamic health benchmark data packages and assess the health status of power plant equipment.

2. The method for automatically generating power plant reports based on AI in a domestically developed information technology environment, as described in claim 1, is characterized in that: The device's operating parameters are collected through an intelligent sensor network, including vibration signals, temperature data, and current waveform data.

3. The method for automatically generating power plant reports based on AI in a domestically developed information technology environment, as described in claim 1, is characterized in that: The feature extraction is implemented using a lightweight neural network model based on an AI framework, including a time-frequency analysis network and a feature fusion network.

4. The method for automatically generating power plant reports based on AI in a domestically developed information technology environment according to claim 1, characterized in that: The streaming clustering algorithm employs adaptive neighborhood radius technology, dynamically adjusting the clustering sensitivity based on the stability of the device's operating status.

5. The method for automatically generating power plant reports based on AI in a domestically developed information technology environment according to claim 1, characterized in that: The calculation of the dynamic threshold range incorporates a dynamic compensation mechanism for environmental deviation and a nonlinear calculation method for health decay.

6. The method for automatically generating power plant reports based on AI in a domestically developed information technology environment according to claim 1, characterized in that: Multi-source data fusion analysis is performed on dynamic health baseline data packages to generate an anomaly diagnosis report that includes anomaly level, anomaly type, and root cause analysis.

7. The method for automatically generating power plant reports based on AI in a domestically developed information technology environment according to claim 6, characterized in that: The multi-source data fusion analysis adopts a three-level analysis architecture, including short-term abnormal fluctuation detection, threshold deviation assessment, and operating condition quality verification.

8. The method for automatically generating power plant reports based on AI in a domestically developed information technology environment according to claim 6, characterized in that: Intelligent reports are generated based on the anomaly diagnosis report and dynamic health benchmark data package. The intelligent reports include an execution summary unit, a data analysis unit, and an operation and maintenance guidance unit.

9. The method for automatically generating power plant reports based on AI in a domestically developed information technology environment, as described in claim 8, is characterized in that: The intelligent report generation adopts a combination of template-based and dynamic generation, and supports interactive and visual display.

10. An AI-based automatic report generation system for power plants under a domestic IT innovation environment, characterized in that: Includes a memory and a processor for using the AI-based automatic report generation method for power plants in a domestic IT innovation environment as described in any one of claims 1-9.

Images