Boiler operation intelligent diagnosis platform based on multi-source data fusion

Abstract

The invention discloses a boiler operation intelligent diagnosis platform based on multi-source data fusion. The boiler operation intelligent diagnosis platform comprises a data acquisition and preprocessing module, a multi-source data fusion and feature extraction module, an intelligent diagnosis and decision support module and a man-machine interaction and alarm module. According to the invention, a modularized multi-source data fusion intelligent diagnosis platform is constructed, various heterogeneous data are uniformly processed through a data acquisition and preprocessing module, and a high-quality data basis is provided for upper-layer analysis; a multi-source data fusion and feature extraction module is used for deeply mining associated features among data, and a fusion feature set capable of comprehensively reflecting the boiler health state is constructed; and finally, through an intelligent diagnosis and decision support module, in combination with historical data and real-time fusion features, early warning, accurate positioning and root analysis of boiler faults are realized, an operation and maintenance mode is converted into predictive maintenance from post-maintenance and regular maintenance, and the safety and reliability of boiler operation are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of boiler technology, and more specifically to an intelligent diagnostic platform for boiler operation based on multi-source data fusion. Background Technology

[0002] Traditional boiler operation monitoring and diagnostic systems primarily rely on alarms based on a limited number of key parameter thresholds, such as temperature and pressure, which trigger alarms when single-point data exceed limits. This approach has significant limitations: First, it is typically reactive, only issuing warnings when a fault has already occurred or is about to occur, lacking predictive capability; second, different sensors and data systems operate independently, forming "information silos," making it impossible to analyze the operational status from a global perspective, and resulting in insufficient diagnostic capabilities for slowly evolving latent faults or complex faults caused by the coupling of multiple systems.

[0003] With the development of IoT and big data technologies, boiler systems can collect multi-source heterogeneous data, including thermodynamic parameters, combustion images, acoustic vibrations, water quality tests, fuel characteristics, and control system commands. However, existing technologies lack effective means to deeply integrate and intelligently analyze this data, thus failing to achieve accurate assessment of boiler operating status, intelligent identification of early faults, and comprehensive optimization of operating efficiency. Summary of the Invention

[0004] To address these issues, this invention provides an intelligent diagnostic platform for boiler operation based on multi-source data fusion.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] The intelligent diagnostic platform for boiler operation based on multi-source data fusion includes a data acquisition and preprocessing module, a multi-source data fusion and feature extraction module, an intelligent diagnosis and decision support module, and a human-computer interaction and alarm module.

[0007] The data acquisition and preprocessing module is used to acquire multi-source heterogeneous operating data in real time from the distributed sensor network, control system database, and manual input interface of the boiler and its auxiliary equipment system; the data acquisition and preprocessing module has built-in data cleaning unit, format standardization unit, and time sequence alignment unit;

[0008] The multi-source data fusion and feature extraction module is connected to the data acquisition and preprocessing module and receives the standardized multi-source time-series data stream; the multi-source data fusion and feature extraction module includes a data-level fusion unit, a feature-level fusion unit, and a database;

[0009] The intelligent diagnosis and decision support module is connected to the multi-source data fusion and feature extraction module, and it integrates a machine learning-based health status assessment model and a fault diagnosis model.

[0010] The human-machine interaction and alarm module is connected to the intelligent diagnosis and decision support module. It includes a human-machine interaction unit and an alarm unit. The human-machine interaction unit is used to dynamically display the boiler's three-dimensional model, real-time operating parameters, health index change curves, fault diagnosis results, and decision support reports in the form of a graphical interface. The alarm unit is used to automatically trigger different levels of audible and visual alarm information according to the severity of the fault.

[0011] Preferably, the data-level fusion unit is used to perform spatial gridding interpolation fusion on homogeneous sensor data with spatial correlation to generate boiler heating surface temperature field cloud map and pressure distribution field; the feature-level fusion unit uses feature engineering methods to extract time-domain, frequency-domain and time-frequency-domain features from time-series data streams and fused field data, and applies principal component analysis and mutual information theory to perform feature dimensionality reduction and screening, mining deep correlation features related to boiler coking, ash accumulation, efficiency decline, unstable combustion, and tube wall overheating faults from multi-source data, and constructing a high-dimensional fused health state feature vector; the database is used to store the benchmark feature vector library under historical normal operating conditions and feature vector templates for typical fault modes;

[0012] The intelligent diagnosis and decision support module receives the fused health status feature vector generated in real time and compares it with the benchmark feature vector library in the database. It then calculates the comprehensive health index of the current boiler system using the health status assessment model. When the health index is lower than a preset threshold or the similarity between the fused health status feature vector and the feature vector template of any fault mode exceeds a set threshold, the fault diagnosis model is triggered. The fault diagnosis model comprehensively applies convolutional neural networks to perform pattern recognition on flame images, applies long short-term memory networks to predict trends in time-series features, and combines expert system rule bases to output specific fault types, fault locations, severity levels, and possible cause analyses. Simultaneously, the intelligent diagnosis and decision support module also includes a decision support unit, which generates a diagnostic report based on the diagnostic results, including operational suggestions, maintenance strategies, and optimized operating parameters.

[0013] Preferably, the data cleaning unit in the data acquisition and preprocessing module specifically performs the following steps:

[0014] A joint algorithm based on sliding window and empirical rules is used to identify and eliminate gross errors in sensor data under steady-state operating conditions.

[0015] For the repair of missing data, depending on the duration and type of missing data, linear interpolation, K-nearest neighbor regression algorithm based on historical data from the same period, or long short-term memory network prediction model are used to fill the missing data.

[0016] The timing alignment unit uses a method based on data valid timestamps and interpolation resampling to unify data streams with different sampling frequencies to a preset highest common frequency or a specified standard frequency.

[0017] Preferably, the feature-level fusion unit, when extracting time-domain features, calculates not only the mean, variance, kurtosis, and skewness, but also waveform indices, impulse indices, and margin indices; when extracting frequency-domain features, it performs Fast Fourier Transform on the vibration and noise signals to extract the amplitudes and total power spectral density of the first N natural frequencies; when extracting time-frequency domain features, it performs wavelet packet transform on the non-stationary signals to extract the energy proportion of each frequency band as features; the feature dimensionality reduction and screening adopts a feature selection algorithm based on the maximum correlation-minimum redundancy criterion to screen out the feature subset that is most relevant to the target fault and has the minimum redundancy between them.

[0018] Preferably, the health status assessment model uses a support vector machine or deep autoencoder to construct a descriptive model of the boiler's normal operating status, and quantifies the comprehensive health index by calculating the deviation between the real-time fused health status feature vector and the boundary of the descriptive model.

[0019] Preferably, it also includes a model self-learning and optimization module, which is connected to the databases of the intelligent diagnosis and decision support module and the multi-source data fusion and feature extraction module. The model self-learning and optimization module continuously collects new normal operation data and manually confirmed fault case data, and periodically or triggeredly performs incremental learning and optimization adjustment on the algorithm parameters in the health status assessment model and fault diagnosis model, and uses the newly accumulated fault case data to expand and update the feature vector template of the typical fault mode.

[0020] Preferably, the model self-learning and optimization module uses an online sequential extreme learning machine algorithm to update a type of support vector machine model in the health status assessment model online, in order to adapt to the slow performance drift of the boiler equipment caused by long-term operation.

[0021] Preferably, the graphical interface of the human-computer interaction and alarm module includes a virtual reality display unit; the virtual reality display unit can overlay and render the three-dimensional model of the boiler with real-time fusion data, including temperature field cloud map, pressure distribution field, and health status labels of key components.

[0022] This invention has the following advantages: By constructing a modular multi-source data fusion intelligent diagnostic platform, it effectively overcomes the shortcomings of existing technologies. The platform uses a data acquisition and preprocessing module to uniformly process various heterogeneous data, providing a high-quality data foundation for upper-level analysis; it utilizes a multi-source data fusion and feature extraction module to deeply mine the correlation features between data, constructing a fusion feature set that comprehensively reflects the boiler's health status; finally, through an intelligent diagnosis and decision support module, combining historical data with real-time fusion features, it achieves early warning, precise location, and root cause analysis of boiler faults, transforming the operation and maintenance mode from reactive maintenance and periodic inspections to predictive maintenance, significantly improving the safety and reliability of boiler operation.

[0023] This platform can also provide operational optimization suggestions based on diagnostic results, guiding operators to adjust parameters, thereby effectively improving boiler thermal efficiency, reducing fuel consumption and pollutant emissions, and achieving the dual goals of safe and economical operation. Attached Figure Description

[0024] To more intuitively illustrate the prior art and this application, exemplary drawings are provided below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing this application; for example, based on the technical concept disclosed in this application and the exemplary drawings, those skilled in the art are able to easily make conventional adjustments or further optimizations to the addition / reduction / classification, specific shapes, positional relationships, connection methods, size ratios, etc. of certain units (components).

[0025] Figure 1 A block diagram of the intelligent diagnostic platform for boiler operation based on multi-source data fusion provided in the embodiments of this application. Detailed Implementation

[0026] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these embodiments are merely for further explanation of the present invention and should not be construed as limiting the scope of protection of the present invention. Technical engineers in the field can make some non-essential improvements and adjustments to the present invention based on the above-described content. 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.

[0027] Please see Figure 1 The intelligent diagnostic platform for boiler operation based on multi-source data fusion includes a data acquisition and preprocessing module, a multi-source data fusion and feature extraction module, an intelligent diagnosis and decision support module, and a human-computer interaction and alarm module.

[0028] The data acquisition and preprocessing module is used to collect multi-source heterogeneous operating data in real time from the distributed sensor network, control system database, and manual input interface of the boiler and its auxiliary systems. The multi-source heterogeneous operating data includes at least boiler body temperature field data, flue gas composition and pressure data of each stage of flue, burner flame image sequence, furnace and fan and pump vibration and noise data, feedwater and steam quality test data, fuel industry analysis data, and control system setpoints and feedback values. The data acquisition and preprocessing module has built-in data cleaning unit, format standardization unit, and time sequence alignment unit to identify and remove outliers, interpolate and repair missing values, standardize data format, and synchronize all data to a unified timestamp to form a standardized multi-source time sequence data stream.

[0029] The multi-source data fusion and feature extraction module is connected to the data acquisition and preprocessing module and receives the standardized multi-source time-series data stream. The multi-source data fusion and feature extraction module includes a data-level fusion unit, a feature-level fusion unit, and a database. The data-level fusion unit performs spatial gridding interpolation fusion on spatially correlated homogeneous sensor data to generate a boiler heating surface temperature field cloud map and pressure distribution field. The feature-level fusion unit uses feature engineering methods to extract time-domain, frequency-domain, and time-frequency-domain features from the time-series data stream and the fused field data, and applies principal component analysis and mutual information theory for feature dimensionality reduction and screening. It mines deep correlation features related to boiler coking, ash accumulation, efficiency decay, unstable combustion, and tube wall overheating faults from the multi-source data, constructing a high-dimensional fused health state feature vector. The database stores a baseline feature vector library under historical normal operating conditions and feature vector templates for typical fault modes.

[0030] The intelligent diagnosis and decision support module is connected to the multi-source data fusion and feature extraction module. Internally, it integrates a machine learning-based health status assessment model and a fault diagnosis model. The intelligent diagnosis and decision support module receives the real-time generated fused health status feature vector and compares it with a benchmark feature vector library in the database. It then calculates the comprehensive health index of the current boiler system using the health status assessment model. When the health index falls below a preset threshold or the similarity between the fused health status feature vector and the feature vector template of any fault mode exceeds a set threshold, the fault diagnosis model is triggered. This model comprehensively applies convolutional neural networks for pattern recognition of flame images, applies long short-term memory networks for trend prediction of temporal features, and combines this with an expert system rule base to output specific fault types, fault locations, severity levels, and possible cause analyses. Simultaneously, the intelligent diagnosis and decision support module also includes a decision support unit, which generates a diagnostic report based on the diagnostic results, including operational suggestions, maintenance strategies, and optimized operating parameters.

[0031] The human-machine interaction and alarm module is connected to the intelligent diagnosis and decision support module. It is used to dynamically display the boiler's three-dimensional model, real-time operating parameters, health index change curves, fault diagnosis results, and decision support reports in the form of a graphical interface. The module is also equipped with a multi-level alarm management unit, which automatically triggers different levels of audible and visual alarm information according to the severity of the fault, and supports notification to operation and maintenance personnel via SMS, email, or mobile application push.

[0032] In implementing the technical solution of this invention, firstly, the data acquisition and preprocessing module, acting as the system's "sensory nerves," extensively collects multi-source heterogeneous data from the boiler, including temperature, pressure, images, vibration, and water quality. Through data cleaning, standardization, and time-series alignment, this raw data is organized into a clean and unified time-series data stream, laying the foundation for subsequent analysis. Next, the multi-source data fusion and feature extraction module, acting as the system's "cerebral cortex," deeply processes the preprocessed data. At the data level, it synthesizes discrete points into intuitive images such as temperature and pressure fields. At the feature level, it extracts deep-fused feature vectors in the time and frequency domains that are strongly correlated with the boiler's health status from massive amounts of data, thus transforming multi-source information into a "language" that can be understood by the intelligent model. Then, the intelligent diagnosis and decision support module, acting as the system's "expert think tank," uses machine learning models (such as support vector machines, CNNs, and LSTMs) and expert systems to compare real-time features with historical normal states and fault templates, calculating a comprehensive health index. This enables early warning, accurate identification, and root cause analysis of faults, and generates specific operation and maintenance decision recommendations. Finally, the human-computer interaction and alarm module, as the system's "interactive interface," presents complex analysis results to maintenance personnel in a graphical or even immersive 3D manner, and issues alarms in multiple ways according to the fault level, ensuring that diagnostic information can be understood and executed in a timely and accurate manner, thereby completing a complete intelligent diagnostic closed loop from data to decision.

[0033] The data cleaning unit in the data acquisition and preprocessing module specifically performs the following steps:

[0034] A joint algorithm based on sliding window and empirical rules is used to identify and remove gross errors in sensor data under steady-state operating conditions, which is very effective against noise that drifts slowly or changes abruptly.

[0035] For the repair of missing data, depending on the duration and type of missing data, linear interpolation, K-nearest neighbor regression algorithm based on historical data from the same period, or long short-term memory network prediction model are used to fill the missing data.

[0036] The time-series alignment unit employs a method based on valid data timestamps and interpolation resampling to unify data streams with different sampling frequencies to a preset highest common frequency or a specified standard frequency. Depending on the complexity of the missing data, it specifies different repair strategies, ranging from simple linear interpolation to complex LSTM prediction models, ensuring the stability and adaptability of the preprocessing stage and providing higher-quality data for subsequent analysis.

[0037] The feature-level fusion unit in the multi-source data fusion and feature extraction module calculates not only the mean, variance, kurtosis, and skewness when extracting time-domain features, but also waveform indices, impulse indices, and margin indices. When extracting frequency-domain features, it performs Fast Fourier Transform on vibration and noise signals to extract the amplitude and total power spectral density of the first N natural frequencies. When extracting time-frequency domain features, it performs wavelet packet transform on non-stationary signals to extract the energy proportion of each frequency band as features. The feature dimensionality reduction and screening adopts a feature selection algorithm based on the maximum correlation-minimum redundancy criterion to screen out the feature subset that is most relevant to the target fault and has the minimum redundancy between them.

[0038] In its implementation, it not only extracts conventional statistical features, but also includes "pulse indicators" and "margin indicators" that reflect the waveform's impact characteristics, as well as frequency band energy features that are sensitive to the signal's frequency structure. Furthermore, by using the "maximum correlation-minimum redundancy" algorithm to filter features, it effectively avoids information overlap, constructs a more concise and representative feature vector, and improves model efficiency and diagnostic accuracy.

[0039] The health status assessment model in the intelligent diagnosis and decision support module uses a support vector machine or deep autoencoder to construct a descriptive model of the boiler's normal operating status. The comprehensive health index is quantified by calculating the deviation between the real-time fused health status feature vector and the boundary of this descriptive model. The fault diagnosis model is a hybrid intelligent model. It first uses a convolutional neural network to perform deep learning on the burner flame image sequence to identify flame morphology, brightness distribution, and stability characteristics, thus determining the combustion state. Simultaneously, it uses a long short-term memory network to perform multi-step trend prediction on key thermodynamic parameters and the principal components of the fused health status feature vector. Finally, the image recognition results, trend prediction results, and the current feature vector are input into an expert system containing IF-THEN rules and fuzzy reasoning for comprehensive decision-making, ultimately determining the fault type and root cause.

[0040] This solution employs an unsupervised learning method, "Support Vector Machine," to construct a health baseline, making it suitable for industrial scenarios where obtaining all fault samples is difficult. The fault diagnosis model is a hybrid intelligent model, combining the image recognition capabilities of CNNs, the temporal prediction capabilities of LSTMs, and the knowledge reasoning capabilities of expert systems. For example, when boiler efficiency declines, CNNs can identify abnormal flame morphology, LSTMs predict an upward trend in flue gas temperature, and the expert system integrates this information to determine "burner coking" and pinpoint its location, achieving multi-angle evidence fusion diagnosis.

[0041] The expert system's rule base is constructed based on boiler thermodynamics principles, equipment operating procedures, a historical fault case database, and domain expert experience. Its rules cover various typical fault modes, including coking, ash accumulation, heat exchanger tube leakage, burner blockage, fan stall, and water-cooled wall overheating. Each rule is associated with a confidence factor, used for uncertainty reasoning when multiple fault possibilities coexist. This allows for weighted reasoning when information is uncertain or multiple possibilities exist. For example, when some symptoms point to ash accumulation and others to coking, the system can output the diagnosis with the highest probability, which is more consistent with engineering practice. The confidence factor can be derived from historical data or adjusted by the user; there are no restrictions on this.

[0042] The platform also includes a model self-learning and optimization module, which enables the system to continuously optimize and update its diagnostic model and fault templates using new data generated during daily operation and human feedback. This module connects the databases of the intelligent diagnosis and decision support module and the multi-source data fusion and feature extraction module. The model self-learning and optimization module continuously collects new normal operation data and manually confirmed fault case data, and periodically or triggeredly performs incremental learning and optimization adjustments on the algorithm parameters in the health status assessment model and fault diagnosis model. It also uses newly accumulated fault case data to expand and update the feature vector templates of typical fault modes, thereby achieving continuous evolution of the platform's diagnostic capabilities.

[0043] The model self-learning and optimization module uses an online sequential extreme learning machine algorithm to update a type of support vector machine model in the health status assessment model online, in order to adapt to the slow performance drift of boiler equipment caused by long-term operation; for the convolutional neural network and long short-term memory network in the fault diagnosis model, a periodic batch retraining method is adopted to fine-tune the network weights using the accumulated new sample data.

[0044] For the health baseline model, a fast online learning algorithm is used to adapt to the slow drift of the device; for complex deep learning models, a more thorough periodic retraining is used to absorb new knowledge. This differentiated update strategy ensures both the real-time performance of the system and the thoroughness and accuracy of the model updates.

[0045] The graphical interface of the human-computer interaction and alarm module integrates a virtual reality display unit, which greatly enhances the intuitiveness of human-computer interaction. This unit can overlay and render the boiler's three-dimensional model with real-time fused data, including temperature field cloud maps, pressure distribution fields, and health status labels of key components. This allows maintenance personnel to conduct immersive inspections through VR devices, intuitively observe the operating status and fault points in areas that are not visible inside the boiler, and enable staff to intuitively view information that cannot be directly displayed by traditional interfaces, such as the internal temperature distribution of the furnace and overheating of specific pipe walls. It also helps personnel quickly locate problems and understand the spatial relationships of complex faults.

[0046] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A boiler operation intelligent diagnostic platform based on multi-source data fusion, characterized in that, It includes a data acquisition and preprocessing module, a multi-source data fusion and feature extraction module, an intelligent diagnosis and decision support module, and a human-computer interaction and alarm module; The data acquisition and preprocessing module is used to acquire multi-source heterogeneous operating data in real time from the distributed sensor network, control system database, and manual input interface of the boiler and its auxiliary equipment system; the data acquisition and preprocessing module has built-in data cleaning unit, format standardization unit, and time sequence alignment unit; The multi-source data fusion and feature extraction module is connected to the data acquisition and preprocessing module and receives the standardized multi-source time-series data stream; the multi-source data fusion and feature extraction module includes a data-level fusion unit, a feature-level fusion unit, and a database; The intelligent diagnosis and decision support module is connected to the multi-source data fusion and feature extraction module, and it integrates a machine learning-based health status assessment model and a fault diagnosis model. The human-machine interaction and alarm module is connected to the intelligent diagnosis and decision support module. It includes a human-machine interaction unit and an alarm unit. The human-machine interaction unit is used to dynamically display the boiler's three-dimensional model, real-time operating parameters, health index change curves, fault diagnosis results, and decision support reports in the form of a graphical interface. The alarm unit is used to automatically trigger different levels of audible and visual alarm information according to the severity of the fault.

2. The intelligent boiler operation diagnostic platform based on multi-source data fusion as described in claim 1, characterized in that, The data-level fusion unit is used to perform spatial gridding interpolation fusion on homogeneous sensor data with spatial correlation to generate temperature field cloud maps and pressure distribution fields of the boiler heating surface. The feature-level fusion unit uses feature engineering methods to extract time-domain, frequency-domain, and time-frequency-domain features from the time-series data stream and the fused field data, and applies principal component analysis and mutual information theory to perform feature dimensionality reduction and screening. It mines deep correlation features related to boiler coking, ash accumulation, efficiency decline, unstable combustion, and tube wall overheating faults from multi-source data, and constructs a high-dimensional fused health status feature vector. The database is used to store the benchmark feature vector library under historical normal operating conditions and the feature vector templates of typical fault modes. The intelligent diagnosis and decision support module receives the fused health status feature vector generated in real time and compares it with the benchmark feature vector library in the database. It then calculates the comprehensive health index of the current boiler system using the health status assessment model. When the health index is lower than a preset threshold or the similarity between the fused health status feature vector and the feature vector template of any fault mode exceeds a set threshold, the fault diagnosis model is triggered. The fault diagnosis model comprehensively applies convolutional neural networks to perform pattern recognition on flame images, applies long short-term memory networks to predict trends in time-series features, and combines this with an expert system rule base to output specific fault types, fault locations, severity levels, and possible cause analyses. Simultaneously, the intelligent diagnosis and decision support module also includes a decision support unit, which generates a diagnostic report based on the diagnostic results, including operational suggestions, maintenance strategies, and optimized operating parameters.

3. The intelligent boiler operation diagnostic platform based on multi-source data fusion as described in claim 1, characterized in that, The data cleaning unit in the data acquisition and preprocessing module specifically performs the following steps: A joint algorithm based on sliding window and empirical rules is used to identify and eliminate errors in sensor data under steady-state operating conditions. For the repair of missing data, depending on the duration and type of missing data, linear interpolation, K-nearest neighbor regression algorithm based on historical data from the same period, or long short-term memory network prediction model are used to fill the missing data. The timing alignment unit uses a method based on data valid timestamps and interpolation resampling to unify data streams with different sampling frequencies to a preset highest common frequency or a specified standard frequency.

4. The intelligent boiler operation diagnostic platform based on multi-source data fusion according to claim 2, characterized in that, The feature-level fusion unit, when extracting time-domain features, calculates not only the mean, variance, kurtosis, and skewness, but also waveform indices, impulse indices, and margin indices; when extracting frequency-domain features, it performs Fast Fourier Transform on vibration and noise signals to extract the amplitude and total power spectral density of the first N natural frequencies; when extracting time-frequency domain features, it performs wavelet packet transform on non-stationary signals to extract the energy proportion of each frequency band as features; the feature dimensionality reduction and screening adopts a feature selection algorithm based on the maximum correlation-minimum redundancy criterion to screen out the feature subset that is most relevant to the target fault and has the minimum redundancy between them.

5. The intelligent boiler operation diagnostic platform based on multi-source data fusion according to claim 2, characterized in that, The health status assessment model uses a support vector machine or deep autoencoder to construct a descriptive model of the boiler's normal operating status, and quantifies the comprehensive health index by calculating the deviation between the real-time fused health status feature vector and the boundary of the descriptive model.

6. The intelligent boiler operation diagnostic platform based on multi-source data fusion according to claim 1, characterized in that, It also includes a model self-learning and optimization module, which connects the databases of the intelligent diagnosis and decision support module and the multi-source data fusion and feature extraction module. The model self-learning and optimization module continuously collects new normal operation data and manually confirmed fault case data, and periodically or triggeredly performs incremental learning and optimization adjustment on the algorithm parameters in the health status assessment model and fault diagnosis model, and uses the newly accumulated fault case data to expand and update the feature vector templates of typical fault modes.

7. The intelligent boiler operation diagnostic platform based on multi-source data fusion according to claim 6, characterized in that, The model self-learning and optimization module uses an online sequential extreme learning machine algorithm to update a type of support vector machine model in the health status assessment model online, in order to adapt to the slow performance drift of boiler equipment caused by long-term operation.

8. The intelligent boiler operation diagnostic platform based on multi-source data fusion according to claim 1, characterized in that, The graphical interface of the human-computer interaction and alarm module includes a virtual reality display unit; the virtual reality display unit can overlay and render the three-dimensional model of the boiler with real-time fused data, including temperature field cloud map, pressure distribution field, and health status labels of key components.

Images