Intelligent correlation analysis and visual display method for production data of thermal power plant
By combining multimodal data acquisition and processing, dynamic correlation model construction and intelligent visualization with edge computing and augmented reality technologies, the problem of multimodal fusion and dynamic correlation of thermal power plant production data has been solved, enabling efficient and accurate production decision-making and optimization.
Patent Information
- Application Number
- CN202510965390.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-14
- Publication Date
- 2025-10-28
AI Technical Summary
Existing technologies for correlation analysis and visualization of thermal power plant production data suffer from problems such as difficulty in multimodal data fusion, inconsistent data frequencies, static models that cannot adapt to changes in operating conditions, and limited visualization effects, leading to delayed decision-making and misjudgments.
By employing multimodal data acquisition and preprocessing, dynamic correlation model construction, intelligent visualization and interactive design, predictive analysis and closed-loop feedback control, and combining graph neural networks, edge computing, federated learning and augmented reality technologies, we can achieve spatiotemporal alignment, dynamic correlation and intelligent decision-making of data.
It improved the efficiency and accuracy of data utilization, realized dynamic correlation analysis of equipment, process and energy consumption, improved the interactive experience and prediction accuracy, reduced unplanned downtime and false alarms, and optimized production decision-making.
Smart Images
Figure CN120851369A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of thermal power plant technology, and more specifically, to a method for intelligent correlation analysis and visualization of thermal power plant production data. Background Technology
[0002] In the modern energy system, thermal power plants, as important power suppliers, bear the key responsibility of ensuring a stable energy supply. With the continuous expansion of installed capacity and the increasing complexity of unit operation, massive and complex data are generated during their production and operation. This data covers all key aspects of unit operation, including but not limited to boiler combustion parameters (such as fuel quantity, air volume, furnace temperature, etc.), turbine operating parameters (such as steam pressure, temperature, speed, etc.), generator electrical parameters (such as voltage, current, power factor, etc.), and operating data of various auxiliary equipment (such as the operating frequency and pressure of feedwater pumps and condensate pumps).
[0003] With the accelerated intelligent transformation of the thermal power generation industry, the efficient utilization of thermal power plant production data has become a core requirement for improving operational efficiency and reducing energy consumption and emissions. However, existing technical solutions still have significant limitations in data correlation analysis, visualization interaction, and decision-making closed loops, specifically in the following aspects:
[0004] 1. Traditional solutions often focus on the analysis of a single data source (such as DCS sensor data), neglecting the value of integrating multimodal data such as video surveillance, environmental parameters, and production process logs. The sampling frequency of data from different devices varies greatly (such as second-level DCS data and minute-level environmental monitoring data), making it difficult to align the spatiotemporal dimensions and limiting the comprehensiveness and accuracy of data correlation analysis.
[0005] 2. Existing methods mostly adopt rule-based or static threshold-based monitoring modes, which cannot dynamically adapt to changes in operating conditions (such as load fluctuations and changes in ambient temperature). Analysis such as equipment failure prediction and energy consumption optimization relies on historical experience or simplified mathematical models, making it difficult to capture the dynamic correlation between equipment, process and energy consumption, resulting in decision lag or misjudgment.
[0006] 3. Traditional visualization solutions mainly use two-dimensional charts, which can only display single parameters or simple statistical results, and cannot intuitively present the spatial layout of equipment, process flow and complex relationships between data. Summary of the Invention
[0007] To overcome the aforementioned shortcomings of existing technologies, this invention provides a method for intelligent correlation analysis and visualization of thermal power plant production data.
[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent correlation analysis and visualization of thermal power plant production data, comprising the following steps:
[0009] S1. Multimodal data acquisition and preprocessing: Acquire equipment sensor data, video surveillance streams, environmental parameters, and production process logs; Employ a spatiotemporal feature alignment algorithm to synchronize data sources with different sampling frequencies in terms of time and space.
[0010] In the multimodal data acquisition of S1, environmental parameters include air quality, temperature and humidity, and outside temperature, while the production process log covers the entire life cycle data of coal procurement, transportation, combustion, and waste disposal.
[0011] The data collection in S1 or the dynamic correlation analysis in S2 also includes data security and privacy protection. The data security and privacy protection adopts a federated learning architecture, which allows different thermal power plants to train models locally and share only gradient parameters instead of raw data; and integrates a differential privacy module to add Laplace noise during data correlation analysis to prevent sensitive production information from being inferred from the visualization results.
[0012] The data preprocessing stage in S1 also includes edge computing and lightweight deployment. Edge computing and lightweight deployment are used to deploy edge computing boxes on-site in thermal power plants to preprocess high-bandwidth data and transmit only feature vectors to the cloud; and to develop model compression algorithms to quantize deep learning models to INT8 precision to adapt to the limited computing resources of industrial PLCs.
[0013] S2. Dynamic Relationship Model Construction: Based on graph neural networks, a dynamic relationship graph of equipment-process-energy consumption is constructed, and the weights of nodes and edges are updated in real time; a causal reasoning mechanism is introduced, and the correlation and causality between data are distinguished through Granger causality test or counterfactual analysis.
[0014] In the construction of the S2 dynamic association model, the nodes of the graph neural network include equipment, process parameters and energy consumption indicators, the edges represent the association relationship between nodes, and the weights are dynamically updated through real-time data streams.
[0015] S3. Intelligent Visual Interactive Design: Develop a 3D visualization platform based on digital twins to support 3D modeling and data overlay for all scenarios in thermal power plants; integrate an AR augmented reality module to enable real-time data projection and anomaly annotation during on-site inspections via handheld devices;
[0016] In the intelligent visualization interaction design of S3, the 3D visualization platform supports gesture / voice dual-modal interaction, and users can zoom the 3D model with gestures or retrieve specific data association paths with voice commands.
[0017] S4. Predictive Analysis and Health Assessment: Using an LSTM-Transformer hybrid network, combining historical equipment fault data with real-time operating parameters, the remaining lifespan of critical equipment is predicted; a dynamic threshold adjustment algorithm is developed to automatically correct equipment anomaly alarm thresholds based on environmental parameters.
[0018] In the predictive analysis of S4, the dynamic threshold adjustment algorithm establishes a dynamic adjustment model for alarm thresholds by mapping environmental parameters to the operating status of the equipment, thereby reducing the false alarm rate.
[0019] S5. Closed-loop feedback control and optimization: The analysis results are fed back to the DCS system through the OPC UA protocol to automatically adjust the combustion ratio and production parameters; a multi-objective optimization algorithm is introduced to meet environmental emission constraints while reducing coal consumption.
[0020] In the closed-loop feedback control step of S5, the multi-objective optimization algorithm takes the minimum coal consumption and emission compliance as optimization objectives, and solves the optimal parameter combination through genetic algorithm or particle swarm algorithm.
[0021] This invention utilizes multimodal data fusion, dynamic correlation graph construction, intelligent visualization interaction, predictive analysis and decision-making, and edge computing technologies to achieve efficient utilization and intelligent decision-making of thermal power plant production data. Specific beneficial effects include:
[0022] 1. Data utilization efficiency is significantly improved. By integrating heterogeneous data from multiple sources such as equipment, video, environment, and processes, the one-sidedness of traditional analysis is solved. The spatiotemporal feature alignment algorithm ensures accurate correlation of data at different frequencies.
[0023] 2. Improved analysis accuracy: The dynamic correlation graph updates the complex relationship between equipment, process, and energy consumption in real time, and the causal reasoning mechanism avoids misjudgments caused by false correlations, thus improving the reliability of fault prediction and energy consumption optimization.
[0024] 3. Enhanced interactive experience: Utilizing 3D digital twin and AR augmented reality technologies, data is intuitively mapped to the device space, enabling natural operation through gesture / voice interaction and overcoming the limitations of traditional two-dimensional charts in information expression;
[0025] 4. The LSTM-Transformer model predicts equipment lifespan in advance and dynamically adjusts thresholds to adapt to changes in operating conditions, transforming maintenance from "regular inspections" to "pre-emptive intervention" and reducing unplanned downtime. Attached Figure Description
[0026] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation
[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.
[0028] As attached Figure 1 The method for intelligent correlation analysis and visualization of thermal power plant production data, as shown, includes the following steps:
[0029] S1. Multimodal data acquisition and preprocessing: Acquire equipment sensor data, video surveillance streams, environmental parameters, and production process logs; Employ a spatiotemporal feature alignment algorithm to synchronize data sources with different sampling frequencies in terms of time and space.
[0030] S2. Dynamic Relationship Model Construction: Based on graph neural networks, a dynamic relationship graph of equipment-process-energy consumption is constructed, and the weights of nodes and edges are updated in real time; a causal reasoning mechanism is introduced, and the correlation and causality between data are distinguished through Granger causality test or counterfactual analysis.
[0031] S3. Intelligent Visual Interactive Design: Develop a 3D visualization platform based on digital twins to support 3D modeling and data overlay for all scenarios in thermal power plants; integrate an AR augmented reality module to enable real-time data projection and anomaly annotation during on-site inspections via handheld devices;
[0032] S4. Predictive Analysis and Health Assessment: Using an LSTM-Transformer hybrid network, combining historical equipment fault data with real-time operating parameters, the remaining lifespan of critical equipment is predicted; a dynamic threshold adjustment algorithm is developed to automatically correct equipment anomaly alarm thresholds based on environmental parameters.
[0033] S5. Closed-loop feedback control and optimization: The analysis results are fed back to the DCS system through the OPC UA protocol to automatically adjust the combustion ratio and production parameters; a multi-objective optimization algorithm is introduced to meet environmental emission constraints while reducing coal consumption.
[0034] Among them, the S1 multimodal data acquisition includes environmental parameters such as air quality, temperature and humidity and outside temperature. The production process log covers the entire life cycle data of coal procurement, transportation, combustion and waste treatment. The entire life cycle data covers coal procurement (supplier information, calorific value), transportation (vehicle trajectory, loss), combustion (ratio, efficiency) and waste treatment (ash and slag destination, secondary utilization), providing a foundation for carbon footprint tracking.
[0035] Specifically, the spatiotemporal feature alignment algorithm in S1 is implemented as follows:
[0036] a. Time dimension alignment: A timestamp-based interpolation algorithm is used to resample the second-level DCS data and the minute-level environmental monitoring data to ensure data correlation within the same time window. For example, the environmental monitoring data (minute level) is extended to the second level through linear interpolation to match the timestamp of the DCS data.
[0037] b. Spatial Dimension Mapping: By mapping device IDs to spatial coordinates, sensor data (such as pipe temperature) is associated with specific locations in the 3D model. For example, a unique device ID is assigned to each temperature sensor, and the correspondence between the ID and the coordinates of the 3D model is established in the digital twin platform.
[0038] In the construction of the S2 dynamic association model, the nodes of the graph neural network include equipment, process parameters and energy consumption indicators, the edges represent the association relationship between nodes, the weights are dynamically updated through real-time data streams, and the node weights are dynamically calculated through equipment running time, process parameter fluctuation range and energy consumption indicator change rate to ensure the real-time performance and accuracy of the association graph.
[0039] Specifically, the implementation method for dynamically updating node and edge weights in the graph neural network of S2 is as follows:
[0040] a. Node feature extraction: Extract static features such as operating time and fault history, as well as dynamic features such as real-time temperature and pressure, from equipment nodes (such as boilers); extract features such as historical average and fluctuation range from process parameter nodes (such as combustion efficiency);
[0041] b. Edge weight update mechanism: An attention mechanism is used to calculate the correlation strength between nodes. For example, the edge weight between boiler temperature and turbine vibration is dynamically adjusted based on the Pearson correlation coefficient of their real-time data. The edge weight is increased when the coefficient is higher than a threshold, and decreased when it is lower.
[0042] In the S3's intelligent visualization interaction design, the 3D visualization platform supports gesture / voice dual-modal interaction. Users can zoom the 3D model with gestures or retrieve specific data association paths with voice commands. Gesture interaction supports operations such as zooming, rotating, and panning, while voice commands support natural language commands such as "display boiler temperature trend" and "retrieve turbine vibration association path".
[0043] The specific integration method of the AR augmented reality module and the 3D platform in S3 is as follows:
[0044] a. Equipment identification and positioning: Scan the QR code of the equipment on site or identify the shape of the equipment through AR devices (such as HoloLens), and call the digital twin platform to obtain the equipment ID and real-time data;
[0045] b. Real-time data projection: Real-time data associated with the device ID (such as vibration value) is rendered into 3D annotations using the AR engine and superimposed onto the device surface. For example, when the vibration exceeds the limit, a red warning box is projected at the bearing position and the specific value is displayed.
[0046] In the predictive analysis of S4, the dynamic threshold adjustment algorithm establishes a dynamic adjustment model for alarm thresholds by mapping environmental parameters to the operating status of equipment, thereby reducing the false alarm rate. The dynamic adjustment model is trained by historical environmental data and equipment alarm records. The corrected alarm threshold can be output by inputting the current environmental parameters (such as the outside temperature).
[0047] Specifically, the logic of the dynamic threshold adjustment algorithm in S4 is as follows:
[0048] a. Environmental parameter mapping model: Establish the mapping relationship between environmental parameters (such as ambient temperature) and the normal operating range of equipment. For example, by training a regression model with historical data, the normal threshold range of equipment vibration, temperature, etc. can be output by inputting the current temperature.
[0049] b. Real-time correction mechanism: The threshold range output by the model is compared with the real-time equipment data, and the alarm threshold is dynamically adjusted. For example, in low temperatures in winter, the alarm lower limit of the equipment lubricating oil temperature is automatically relaxed to avoid false alarms.
[0050] In the closed-loop feedback control step of S5, the multi-objective optimization algorithm takes the minimum coal consumption and emission compliance as the optimization objectives. It solves the optimal parameter combination through genetic algorithm or particle swarm algorithm. The multi-objective optimization algorithm introduces constraints (such as unit start-up and shutdown time, environmental protection equipment operation status) during the solution process to ensure that the optimization results conform to the actual production rules.
[0051] Specifically, the implementation of the solution strategy for the multi-objective optimization algorithm in S5 is as follows:
[0052] a. Objective function design: Transform the goals of minimizing coal consumption and meeting emission standards into mathematical objectives, such as minimizing unit coal consumption for power generation (g / kWh) and minimizing SO2 emission concentration (mg / m³). 3 );
[0053] b. Constraint handling: Introduce constraints such as unit start-up and shutdown time and environmental protection equipment operating status, and integrate the constraints into the objective function through the penalty function method. For example, if the environmental protection equipment is not running, a high penalty coefficient is applied to the emission target.
[0054] c. Algorithm selection: An improved particle swarm optimization algorithm is adopted, which introduces a dynamic adjustment mechanism for inertia weight to balance global search and local convergence capabilities.
[0055] Preferably, after data collection in S1 or before dynamic correlation analysis in S2, data security and privacy protection are also included. The data security and privacy protection adopts a federated learning architecture, which allows different thermal power plants to train models locally and share only gradient parameters rather than the original data. It also integrates a differential privacy module to add Laplace noise during data correlation analysis to prevent sensitive production information from being inferred from the visualization results. The federated learning architecture achieves cross-power plant collaboration by encrypting gradient parameters, and the differential privacy module ensures that even if the parameters are intercepted, the original production data cannot be restored.
[0056] Specifically, the implementation of the federated learning architecture and the method of adding differential privacy noise are as follows:
[0057] a. Federated learning process: Each thermal power plant trains its model (such as a fault prediction model) locally using historical data, and only encrypts the model gradient parameters before uploading them to the central server; the server aggregates the gradients and updates the global model, and then distributes it to each power plant for the next round of training;
[0058] b. Differential privacy noise addition: During data association analysis, Laplace noise is added to the feature vectors. For example, noise following a Laplace(0, b) distribution is added to the equipment vibration feature values, where b is determined by the privacy budget ε (b = Δf / ε, Δf is the sensitivity).
[0059] Preferably, the data preprocessing stage in S1 also includes edge computing and lightweight deployment. Edge computing and lightweight deployment are used to deploy edge computing boxes on-site in the thermal power plant to preprocess high-bandwidth data and transmit only feature vectors to the cloud. A model compression algorithm is developed to quantize the deep learning model to INT8 precision to adapt to the limited computing resources of industrial PLCs. When the edge computing box preprocesses the video stream, it extracts key frame features (such as flame color and pipeline leak points) and transmits only feature vectors instead of the original video, reducing network bandwidth requirements.
[0060] Specifically, the methods and model compression algorithms for preprocessing video streams at edge computing nodes are as follows:
[0061] a. Video stream preprocessing: Deploy object detection models (such as YOLOv5) in edge computing boxes to identify key areas in the video in real time (such as pipe leaks and abnormal flame colors), extract feature vectors (such as leak coordinates and flame RGB mean) of these areas, and transmit them to the cloud.
[0062] b. Model quantization method: The INT8 quantization technique is used to convert the weights and activation values of the deep learning model (such as LSTM-Transformer) from FP32 to INT8 format. For example, the quantization scale is determined by the KL divergence calibration method to minimize the difference in model output before and after quantization.
[0063] In summary, this invention can integrate equipment, video, environmental, and process data, solving the limitations of traditional single-source data analysis (such as using only sensor data while ignoring the impact of the environment on equipment); and through algorithms, it synchronizes data at different frequencies (such as second-level DCS and minute-level environmental data), ensuring the spatiotemporal consistency of correlation analysis.
[0064] By using dynamic correlation graphs to update the correlation between equipment, process and energy consumption in real time, the problem of static models being unable to adapt to changes in operating conditions can be avoided. Granger tests are used to distinguish between correlation and causation, reducing misjudgments caused by false correlations (such as mistakenly associating ambient temperature fluctuations with equipment failures).
[0065] By mapping equipment parameters to 3D models (such as displaying pipe temperature as changes in the color of the model surface), spatial relationships that are difficult to express in traditional two-dimensional charts can be intuitively displayed. By using augmented reality (AR), real-time data can be projected onto the surface of the equipment through handheld devices during inspections, and abnormal points (such as the location of bearings with excessive vibration) can be directly marked.
[0066] By using the LSTM-Transformer model to predict the remaining lifespan of equipment in advance, maintenance can be transformed from "periodic inspection" to "pre-intervention". The alarm threshold can be automatically adjusted according to environmental parameters (such as outside temperature) to reduce false alarms caused by changes in operating conditions (such as the expansion of the normal vibration range of equipment in low winter temperatures).
[0067] The analysis results directly adjust DCS parameters (such as combustion ratio) to achieve a fully automated closed loop of "analysis-decision-execution". This reduces coal consumption while meeting environmental constraints (such as SO2 / NOx concentration limits), solving the problem of traditional single-objective optimization that suffers from one aspect while another.
[0068] Federated learning uses shared models across power plants instead of raw data to solve data silos while protecting trade secrets; edge computing preprocesses high-bandwidth data such as video streams on-site and transmits only feature vectors to the cloud, reducing network dependence and computing latency.
[0069] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for intelligent correlation analysis and visualization of thermal power plant production data, characterized in that, Includes the following steps: S1. Multimodal data acquisition and preprocessing: Acquire equipment sensor data, video surveillance streams, environmental parameters, and production process logs; Employ a spatiotemporal feature alignment algorithm to synchronize data sources with different sampling frequencies in terms of time and space. S2. Dynamic Relationship Model Construction: Based on graph neural networks, a dynamic relationship graph of equipment-process-energy consumption is constructed, and the weights of nodes and edges are updated in real time. Introduce a causal reasoning mechanism and distinguish between correlation and causation between data through Granger causality test or counterfactual analysis; S3. Intelligent Visual Interactive Design: Develop a 3D visualization platform based on digital twins to support 3D modeling and data overlay for all scenarios in thermal power plants; integrate an AR augmented reality module to enable real-time data projection and anomaly annotation during on-site inspections via handheld devices; S4. Predictive Analysis and Health Assessment: Using an LSTM-Transformer hybrid network, combining historical equipment fault data with real-time operating parameters, the remaining lifespan of critical equipment is predicted; a dynamic threshold adjustment algorithm is developed to automatically correct equipment anomaly alarm thresholds based on environmental parameters. S5. Closed-loop feedback control and optimization: The analysis results are fed back to the DCS system through the OPC UA protocol to automatically adjust the combustion ratio and production parameters; a multi-objective optimization algorithm is introduced to meet environmental emission constraints while reducing coal consumption.
2. The method for intelligent correlation analysis and visualization of thermal power plant production data according to claim 1, characterized in that: In the multimodal data acquisition of S1, environmental parameters include air quality, temperature and humidity, and outside temperature, while the production process log covers the entire life cycle data of coal procurement, transportation, combustion, and waste disposal.
3. The method for intelligent correlation analysis and visualization of thermal power plant production data according to claim 1, characterized in that: In the construction of the S2 dynamic association model, the nodes of the graph neural network include equipment, process parameters and energy consumption indicators, the edges represent the association relationship between nodes, and the weights are dynamically updated through real-time data streams.
4. The method for intelligent correlation analysis and visualization of thermal power plant production data according to claim 1, characterized in that: In the intelligent visualization interaction design of S3, the 3D visualization platform supports gesture / voice dual-modal interaction, allowing users to zoom the 3D model with gestures or retrieve specific data association paths with voice commands.
5. The method for intelligent correlation analysis and visualization of thermal power plant production data according to claim 1, characterized in that: In the predictive analysis of S4, the dynamic threshold adjustment algorithm establishes a dynamic adjustment model for alarm thresholds by mapping environmental parameters to the operating status of the equipment.
6. The method for intelligent correlation analysis and visualization of thermal power plant production data according to claim 1, characterized in that: The data collection in S1 or the dynamic correlation analysis in S2 also includes data security and privacy protection. The data security and privacy protection adopts a federated learning architecture, which allows different thermal power plants to train models locally and share only gradient parameters rather than the original data; and integrates a differential privacy module to add Laplace noise during data correlation analysis.
7. The method for intelligent correlation analysis and visualization of thermal power plant production data according to claim 1, characterized in that: The data preprocessing stage in S1 includes edge computing and lightweight deployment. Edge computing and lightweight deployment are used to deploy edge computing boxes on-site at thermal power plants to preprocess high-bandwidth data and transmit only feature vectors to the cloud. Develop a model compression algorithm to quantize deep learning models to INT8 precision, adapting to the limited computing resources of industrial PLCs.
8. The method for intelligent correlation analysis and visualization of thermal power plant production data according to claim 1, characterized in that: In the closed-loop feedback control step of S5, the multi-objective optimization algorithm takes the minimum coal consumption and emission compliance as optimization objectives, and solves the optimal parameter combination through genetic algorithm or particle swarm algorithm.