Modularized high-precision performance online calculation platform for operation process of thermal power generating unit

Abstract

The invention discloses a modular high-precision performance online calculation platform for a thermal power generating unit operation process, and relates to the technical field of thermal power generating unit performance online calculation. Comprising a real-time data credibility sensing and marking subsystem, a micro-service and container-based dynamic reconfigurable computing subsystem, a graphical process arrangement and credibility visualization subsystem, a credible data driving bus and a cloud edge collaborative architecture. The real-time data credibility sensing and marking subsystem performs multi-modal coordination processing on original measuring point data, binds and quantifies an uncertainty interval, and generates a standard data stream with a credibility label; the dynamic reconfigurable computing subsystem realizes dynamic management of computing micro-services through a service arrangement and hot deployment engine; the graphical process arrangement and credibility visualization subsystem supports self-defined calculation workflow and full-link credibility traceability visualization; the trusted data driving bus guarantees data transmission and self-adaptive scheduling, and the reliability of a calculation result and the flexibility of the system are improved.

Description

Technical Field

[0001] This invention relates to the field of online performance calculation technology for thermal power units, specifically a modular high-precision online performance calculation platform for the operation process of thermal power units. Background Technology

[0002] With the deepening of the power system reform and the continuous improvement of energy conservation and emission reduction requirements, the economic efficiency of thermal power unit operation has become a key factor in the core competitiveness of power generation enterprises. To further explore energy-saving potential, online performance calculation and monitoring systems based on real-time data have become an indispensable technical tool for modern thermal power plants. These systems collect information from data sources such as the unit's distributed control system (DCS) to perform real-time calculations and analyses of key economic indicators such as boiler efficiency, turbine heat consumption, and plant power consumption rate, aiming to provide immediate guidance for operational optimization.

[0003] Currently, relevant technical solutions in this field mainly focus on improving the accuracy of computational models or expanding analytical functions. For example, the invention patent CN114609926B, "A Dynamic Online Simulation Method for Thermal Power Plants Based on a Thermal Power Simulation Platform," discloses a method for improving the fitting accuracy of the simulation model to the actual unit operating state by establishing a mathematical simulation model and using DCS data for dynamic online back-calculation correction. Another example is some advanced intelligent operation and management platforms that integrate big data analysis, machine learning, and other technologies to achieve advanced functions such as equipment characteristic analysis, risk warning, and operation optimization guidance. In addition, there are also modular platforms focused on thermal system simulation, allowing users to build system models in a modular fashion for variable operating condition analysis and optimization during the design phase.

[0004] However, while pursuing powerful functionality and computational accuracy, existing technical solutions generally suffer from an architectural limitation: the systems typically employ a tightly coupled, integrated, or fixed-layered architecture. In this architecture, data preprocessing, core computation, and result output are deeply integrated. Any optimization or upgrade of a single computational algorithm, or the addition of a new analysis module, can have far-reaching consequences, requiring system shutdown, modification, debugging, and redeployment. This rigid architecture makes performance computing platforms unable to respond flexibly and quickly to technological iterations and changes in business needs. System maintenance and functional expansion are costly, ultimately resulting in insufficient long-term viability and continuous evolution capabilities. This has become a bottleneck restricting the development of online performance computing technology for thermal power units towards greater agility and intelligence. This invention is proposed against this backdrop. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a modular high-precision online performance calculation platform for the operation of thermal power units. Through modular architecture and trustworthiness awareness technology, it can achieve non-stop upgrades and full-link trustworthiness traceability, thereby improving the reliability of calculation results, reducing maintenance costs, and supporting flexible function expansion and multi-power plant collaborative optimization.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a modular high-precision online performance calculation platform for the operation of thermal power units, the platform comprising:

[0007] The real-time data credibility perception and labeling subsystem is used to obtain raw measurement point data from power plant data sources, perform multimodal data coordination processing, and calculate and bind a quantified uncertainty range for each frame of processed data based on data coordination residuals and measurement instrument accuracy, generating a standard data stream with credibility labels.

[0008] The dynamic reconfigurable computing subsystem based on microservices and containers includes multiple computing microservices that encapsulate specific performance computing functions. Each computing microservice is deployed in an independent container, and its service contract clearly defines the format of input and output data and the required data trustworthiness requirements. The dynamic reconfigurable computing subsystem also includes a service orchestration and hot deployment engine, which is used to dynamically manage the lifecycle of the computing microservices without downtime of the machine group, including blue-green deployment or canary release of services based on container images, and realize online loading, replacement and unloading of computing microservices.

[0009] The graphical workflow orchestration and credibility visualization subsystem provides a graphical interface for users to customize the orchestration of computing workflows, including data sources, data processing filters, and one or more computing microservices, by dragging and dropping components and connecting them. It also provides full-link credibility traceability visualization of the final performance indicators based on the uncertainty range.

[0010] The trusted data-driven bus connects the real-time data trustworthiness perception and labeling subsystem, the dynamic reconfigurable computing subsystem, and the graphical process orchestration and visualization subsystem. It is responsible for transmitting the standard data stream with trustworthiness labels and triggering the adaptive scheduling strategy of computing microservices based on data quality events.

[0011] Furthermore, the multimodal data coordination processing in the real-time data credibility perception and labeling subsystem specifically performs the following steps:

[0012] Receive raw measurement point datasets from the distributed control system. ,in Indicates the first The original measured values ​​of each measuring point This represents the total number of measurement points.

[0013] According to the preset unit operating condition identification rules, a target coordination strategy is selected from multiple preset data coordination strategies. The data coordination strategy includes at least a coordination strategy based on mechanism constraints and a coordination strategy based on neural network residual prediction.

[0014] Apply the target coordination strategy to the original measurement point dataset The data is processed to obtain the reconciled dataset. and the corresponding residual vector ,in The observation matrix is ​​constructed based on the correlation between measurement points, where the coordination strategy based on mechanism constraints is solved by solving a constraint optimization problem. Implementation, in which The optimization process must satisfy a set of equality constraints consisting of mass conservation and energy conservation equations, given the covariance matrix of the measurement error. .

[0015] Furthermore, a quantized uncertainty range is calculated and bound to the processed data of each frame in real time, which is achieved through the following process:

[0016] For the coordinated data Calculate its combined standard uncertainty. The calculation formula is: ,in For the Type B standard uncertainty component determined according to the measuring instrument calibration certificate, Based on the residual vector The Type A standard uncertainty components determined by statistical characteristics are obtained through... Calculation, where For the residual vector corresponding to the first The residuals at each measuring point The standard deviation within the sliding time window. This represents the number of valid samples within the sliding time window.

[0017] Based on the combined standard uncertainty and the preset inclusion factor Calculate the expanded uncertainty The expanded uncertainty The uncertainty interval is defined. ;

[0018] The coordinated data and its corresponding expanded uncertainty The data is encapsulated into a unified format data packet to form the standard data stream with the credibility label.

[0019] Furthermore, the input data reliability requirements defined in the service contract of the computational microservice include the maximum allowed extended uncertainty threshold for the input data item. ;

[0020] When executing the blue-green deployment or canary release, the service orchestration and hot deployment engine simultaneously routes the standard data stream with the trustworthiness label to both the old and new versions of the computing microservice instance. It then verifies the new version of the service and makes traffic switching decisions by comparing the consistency of the output results of the old and new instances within the overlapping range of uncertainty intervals and determining whether the uncertainty of the output results themselves meets the preset convergence conditions.

[0021] Furthermore, the dynamic reconfigurable computing subsystem also includes a data quality monitoring module, which continuously monitors the uncertainty range change trend of the data at each measurement point output by the real-time data credibility perception and labeling subsystem;

[0022] An alarm is triggered when the uncertainty of data at a specific measurement point continuously exceeds a preset first threshold.

[0023] When the uncertainty exceeds a higher second threshold, the data quality monitoring module sends a data quality degradation event to the service orchestration and hot deployment engine, triggering the service orchestration and hot deployment engine to switch the computing microservice instance that depends on the measurement point data to a pre-configured backup computing model with lower requirements for input data quality.

[0024] Furthermore, the platform adopts a cloud-edge collaborative architecture, with the core components of the real-time data credibility perception and labeling subsystem, the dynamic reconfigurable computing subsystem, and the graphical process orchestration and credibility visualization subsystem deployed on the edge computing nodes local to the power plant.

[0025] The platform also includes a cloud-based cross-platform integration and intelligent application ecosystem subsystem, which includes an algorithm model library, a federated learning coordination server, and third-party intelligent application encapsulation specifications.

[0026] The computing microservices on the edge computing nodes interact with the model parameter gradients of similar computing microservices on other power plant edge nodes based on the federated learning coordination server, so as to collaboratively train and generate a global optimization model.

[0027] The service orchestration and hot deployment engine dynamically updates local computing microservice instances based on the global optimization model.

[0028] Furthermore, the components provided by the graphical process orchestration and credibility visualization subsystem include a data source component, a credibility filtering component, a computational microservice component, and a result output component.

[0029] The credibility filtering component allows users to set logical rules to route data streams to different downstream computing microservice component branches based on the size of the uncertainty range of the input data.

[0030] Furthermore, the platform includes at least one internal algorithm of the aforementioned computational microservice, which performs probability calculations using the uncertainty interval carried by the input data to achieve uncertainty propagation, for functions with the following relationships: The output of the computational microservice Combined standard uncertainty Propagation calculations are performed using the following formula:

[0031] in, For function For input quantity The partial derivatives, Input quantity The combined standard uncertainty, Input quantity and The estimated correlation coefficient between them;

[0032] The computational microservice ultimately outputs the computation result. and its expanded uncertainty .

[0033] Furthermore, when the service orchestration and hot deployment engine performs dynamic management of computing microservices, the container images it relies on are stored in a private image repository integrated with a version control system.

[0034] When the algorithm code library in the version control system is updated and the build pipeline is triggered, a new container image is automatically generated and pushed to the private image repository;

[0035] After the service orchestration and hot deployment engine detects a new image event, it automatically starts the blue-green deployment or canary release process.

[0036] Furthermore, the third-party intelligent application encapsulation specification specifically stipulates the metadata that the computing microservices accessing the platform must provide. The metadata includes at least: a unique identifier for the microservice, a functional description, an input / output data pattern definition, input data credibility requirements, and a resource requirement description.

[0037] The intelligent application algorithms encapsulated based on this specification are built into container images and uploaded to the algorithm model library of the cross-platform integration and intelligent application ecosystem subsystem.

[0038] The graphical process orchestration and credibility visualization subsystem provides an interface for authorized users to subscribe to specified container images from the algorithm model library and instantiate and deploy them to specified edge computing nodes for operation.

[0039] Compared with existing technologies, this modular high-precision online performance calculation platform for the operation of thermal power units has the following advantages:

[0040] I. This invention, by adopting a microservice and containerized architecture, encapsulates specific performance computing functions into independent computing microservices and deploys them in containers. Combined with service orchestration and hot deployment engines, it enables blue-green deployment or canary release of computing microservices. This allows for online loading, replacement, and unloading of computing microservices without downtime of the system, thereby breaking the limitations of the tightly coupled integrated architecture of existing technologies. It avoids changes to the entire system when optimizing or adding a single function, significantly reduces system maintenance and function expansion costs, effectively improves the platform's responsiveness to technological iterations and changes in business needs, and enhances the platform's long-term vitality and continuous evolution capabilities.

[0041] Second, this invention achieves multimodal data coordination and processing and quantification uncertainty interval binding through a real-time data credibility perception and labeling subsystem. Combined with the uncertainty propagation calculation of computing microservices and the graphical full-link credibility tracing function, it can provide a reliable credibility basis for performance calculation results. At the same time, the cooperation between the data quality monitoring module and the adaptive scheduling strategy can switch to the backup calculation model when the data quality of some measurement points is poor, ensuring the stable operation of the system.

[0042] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0044] Figure 1 This is a schematic diagram of the overall architecture and data flow of the system in the platform of this invention;

[0045] Figure 2 This is a schematic diagram of the credibility data processing and propagation chain of the present invention;

[0046] Figure 3This is a schematic diagram of the dynamic reconstruction and emergency scheduling of computing microservices according to the present invention. Detailed Implementation

[0047] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0048] Example

[0049] like Figures 1 to 3 As shown, this embodiment uses a 300MW coal-fired power unit as the application object to describe in detail the specific implementation of the modular high-precision online performance calculation platform for the operation process of the power unit of the present invention. The implementation process of each part is described in detail below.

[0050] In this embodiment, the platform adopts a cloud-edge collaborative architecture. Edge computing nodes are deployed in the power plant's local control room, configured with two industrial servers (Intel Xeon Gold 6330 CPU, 128GB RAM, 2TB SSD) to run the core components of the real-time data credibility perception and labeling subsystem, the dynamic reconfigurable computing subsystem, and the graphical process orchestration and credibility visualization subsystem. The cloud is deployed on a public cloud server, configured with 8 cores, 16GB RAM, and a 500GB cloud disk, running the cross-platform integration and intelligent application ecosystem subsystem. The edge computing nodes and the cloud are connected via the power plant's dedicated fiber optic network with a bandwidth of 1000Mbps, ensuring the real-time performance and stability of data transmission.

[0051] The trusted data-driven bus is implemented using a message queue based on the MQTT protocol. It is responsible for data transmission between subsystems within the edge node and between the edge and the cloud. The transmission latency is controlled within 100ms to meet the real-time computing requirements.

[0052] In this embodiment, the real-time data credibility perception and labeling subsystem is deployed on one of the industrial servers of the edge computing node. It obtains raw measurement point data from the distributed control system of the unit through the OPCUA protocol. The total number of measurement points is 800, covering key operating parameters of major equipment such as boilers, steam turbines, and generators, including types such as temperature, pressure, flow rate, liquid level, current, and voltage.

[0053] Multimodal data coordination processing is performed according to the following steps:

[0054] Raw measurement point data reception: The subsystem periodically collects raw measurement point data from the distributed control system via the OPCUA client, forming a raw measurement point dataset. ,in For the first The original measurement values ​​of each measuring point, for example The original measured value of the outlet water temperature corresponding to the economizer of the boiler. This corresponds to the original measured value of the steam inlet pressure of the high-pressure cylinder of the steam turbine. The original measurement point data of the distributed control system are collected, with a specific sampling period of 1 second.

[0055] Unit operating condition identification: The preset unit operating condition identification rules are based on unit load division, specifically divided into three ranges: low load (120MW~180MW), medium load (180MW~240MW), and high load (240MW~300MW). The subsystem collects real-time generator output power (measurement point number). Determine the current operating condition, for example, when At that time, it was determined to be a medium load condition.

[0056] Target coordination strategy selection: Preset data coordination strategies include coordination strategies based on mechanistic constraints and coordination strategies based on neural network residual prediction. In this embodiment, the matching rule between operating conditions and coordination strategies is as follows: under low-load operating conditions, the coordination strategy based on neural network residual prediction is selected; under medium-to-high-load operating conditions, the coordination strategy based on mechanistic constraints is selected. This matching rule is preset through a configuration file and can be modified by the user through a graphical interface.

[0057] The coordination strategy is applied in the following ways:

[0058] Implementation of coordination strategy based on mechanism constraints: Under medium-to-high load conditions, this strategy is applied to the original dataset. Processing is then performed. First, an observation matrix is ​​constructed. , It is an 800×800 sparse matrix, and the values ​​of its elements are determined according to the physical relationships between the measuring points. For example, for a boiler feedwater system, the feedwater flow measuring points ( Economizer inlet flow measurement point ( ), steam drum makeup water flow measurement point ( There is a flow balance relationship, therefore The row and column elements corresponding to these measurement points are set to 1, and the elements corresponding to other unrelated measurement points are set to 0, so as to reflect the observation correlation between measurement points.

[0059] covariance matrix This is an 800×800 diagonal matrix, where the diagonal elements represent the variance of the measurement error at each measuring point, determined according to the technical parameters of the measuring instrument. For example, a pressure sensor with an accuracy class of 0.2 has a measurement range of 0~10MPa, and its standard deviation of measurement error is... The corresponding variance is This value is used as The diagonal elements corresponding to the pressure measurement points.

[0060] Equality constraint set It consists of the mass conservation equation and the energy conservation equation. Taking a boiler system as an example, the mass conservation equation is that the total feedwater flow rate equals the sum of the steam flow rate and the blowdown flow rate, i.e. ,in, The value after coordination of the main steam flow measurement points. This is the value after coordination of the boiler blowdown flow measurement points; the energy conservation equation is that the heat input of fuel equals the sum of the heat output of steam and all heat losses, i.e. ,in This is the value after coordination of fuel consumption measurement points. For the lower heating value of fuel, The main steam enthalpy value, For the enthalpy value of the feedwater, This is the sum of all heat losses in the boiler.

[0061] Constrained optimization problem Solving using the Lagrange multiplier method, constructing the Lagrange function. ,in This is the Lagrange multiplier vector. Through... and Taking the partial derivatives and setting them to zero yields a system of equations. The Newton-Raphson method is used to iteratively solve this system. The convergence condition for the iteration is that the equations converge after two iterations... The maximum value of the difference between the elements of the vector is less than The number of iterations does not exceed 20, and the final reconciled dataset is obtained. .

[0062] Implementation of a coordination strategy based on neural network residual prediction: This strategy is applied under low-load conditions. The neural network used is a backpropagation (BP) neural network with 800 input layer nodes, corresponding to the original measurements from 800 measuring points. Three hidden layers are set, with 256, 128, and 64 nodes per layer, respectively. The output layer has 800 nodes, corresponding to the coordinated data. The training samples for the neural network are derived from the unit's historical operating data. 100,000 sets of valid data under low-load conditions are selected, specifically, data integrity ≥95% and no obvious outliers. 80,000 sets are used as the training set, and 20,000 sets as the validation set. During training, the goal is to minimize the sum of squared residuals. The Adam optimizer is used with a learning rate of 0.001, and the training iterations are 1000. Training stops after the validation set error converges. The original dataset... The input is a trained backpropagation (BP) neural network, and the output is the reconciled dataset. .

[0063] Residual vector calculation: according to the formula Calculate the residual vector ,in Dimensions and Consistent , Reflecting the The deviation between the original measured value and the coordinated value at each measuring point is used for subsequent uncertainty calculation.

[0064] Uncertainty interval calculation and standard data stream generation: for each frame of coordinated data The uncertainty interval is calculated and bound as follows:

[0065] Calculation of Type A standard uncertainty components: Type A standard uncertainty components Through residual vector The statistical characteristics are determined, and the calculation formula is as follows: The sliding time window is set to 30 seconds. Since the data sampling period is 1 second, the number of valid samples within the sliding time window is... ; For the first The residuals at each measuring point The standard deviation within the sliding time window is calculated as follows: ,in For the first Time of the first The residuals at each measuring point For the first time within the sliding time window The average value of the residuals at each measuring point. For example, the residuals of the 10th measuring point within the sliding time window are 0.2℃, 0.1℃, -0.3℃... a total of 30 data points, calculated as follows: , ,but .

[0066] Calculation of Type B standard uncertainty components: Type B standard uncertainty components The maximum permissible error is determined based on the instrument's calibration certificate. This certificate specifies the instrument's maximum permissible error; for example, the maximum permissible error for a temperature sensor might be ±0.5℃, assuming a uniform distribution. Therefore, the temperature sensor corresponds to The maximum permissible error of a certain pressure sensor is ±0.02 MPa, corresponding to... .

[0067] Calculation of combined standard uncertainty: Combined standard uncertainty Through formula Calculation. Continuing with the example of the 10th measuring point above, ≈0.046℃, ≈0.289℃, then ≈0.293℃.

[0068] Calculation of expanded uncertainty and uncertainty interval: Preset coverage factor Corresponding to a 95% confidence level, the expanded uncertainty is... Taking the 10th measuring point as an example, ≈0.586℃, corresponding to an uncertainty range of... .

[0069] Standard data stream generation: This involves generating coordinated data. and its corresponding expanded uncertainty The data is encapsulated in JSON format and includes fields such as measurement point number, timestamp, reconciled value, expanded uncertainty, and confidence level. Based on the magnitude of uncertainty, it is divided into three levels: high, medium, and low, forming a standard data stream with confidence labels. For example:

[0070] {"pointId":10,"timestamp":"2024-05-2010:00:00","value":280.5,"uncertainty":0.586,"confidenceLevel":"High"}.

[0071] In this embodiment, the dynamically reconfigurable computing subsystem is deployed on another industrial server on the edge computing node, using Docker as the container runtime environment and Kubernetes as the container orchestration tool to achieve lifecycle management of computing microservices.

[0072] Design and Deployment of Computing Microservices: This embodiment designs six core computing microservices, each encapsulating specific performance computing functions, and all are deployed in independent Docker containers, as detailed below:

[0073] Boiler Efficiency Calculation Microservice: The function is to calculate the boiler thermal efficiency. The input data includes 15 measurement points with confidence labels, such as fuel consumption, fuel lower heating value, main steam flow rate, main steam enthalpy, feedwater flow rate, feedwater enthalpy, flue gas temperature, and flue gas oxygen content. The output data is the boiler thermal efficiency and the corresponding expanded uncertainty.

[0074] Steam Turbine Heat Consumption Calculation Microservice: The function is to calculate the steam turbine heat rate. The input data includes data with confidence labels from 20 measurement points, such as main steam flow rate, main steam pressure, main steam temperature, reheat steam flow rate, reheat steam pressure, reheat steam temperature, exhaust steam pressure, and generator power. The output data is the steam turbine heat rate and the corresponding expanded uncertainty.

[0075] The microservice for calculating plant power consumption rate calculates the plant power consumption rate. The input data includes 30 measurement points with confidence labels, such as generator output power and power consumption of each auxiliary machine. The output data is the plant power consumption rate and the corresponding expanded uncertainty.

[0076] Flue gas heat loss calculation microservice: The function is to calculate the boiler flue gas heat loss. The input data includes data with confidence labels from 10 measuring points such as flue gas temperature, flue gas oxygen content, and fuel element analysis data. The output data is the flue gas heat loss and the corresponding expanded uncertainty.

[0077] Microservice for calculating heat loss of fly ash combustibles: The function is to calculate the heat loss of fly ash combustibles. The input data includes data with confidence labels from 5 measurement points, including fly ash carbon content, fuel lower heating value, and fuel ash content. The output data is the heat loss of fly ash combustibles and the corresponding expanded uncertainty.

[0078] The integrated power supply coal consumption calculation microservice is used to calculate the integrated power supply coal consumption of a computer group. The input data includes data with confidence labels from eight measurement points, such as boiler efficiency, turbine heat rate, plant power consumption rate, and fuel lower heating value. It also includes output data from other microservices. The output data is the integrated power supply coal consumption and the corresponding expanded uncertainty.

[0079] Each computing microservice's service contract is defined using the OpenAPI 3.0 specification, clearly defining the field names, data types, units, and input / output data reliability requirements. The input data reliability requirements specifically specify the maximum allowed expanded uncertainty threshold for each input data item. For example, in the boiler efficiency calculation microservice, the main steam flow data Smoke exhaust temperature data ℃, if the expanded uncertainty of the input data exceeds the corresponding If the data is deemed unreliable, the microservice will refuse to process it and will report a data quality anomaly.

[0080] The service orchestration and hot deployment engine is implemented as follows:

[0081] The service orchestration and hot deployment engine is implemented based on Kubernetes Deployment and Service resources, supporting both blue-green deployment and canary deployment strategies. The specific implementation process is as follows:

[0082] Container image management: Container images for compute microservices are stored in a private image repository (Harbor) integrated with the GitLab version control system. When an algorithm codebase in GitLab is updated and triggers the Jenkins build pipeline, the code is automatically compiled, a new Docker image is built, and pushed to the Harbor private image repository. Image tags are named in version number-timestamp format, such as v1.2-202405201530.

[0083] Blue-green deployment and implementation: Taking the upgrade of the boiler efficiency calculation microservice v1.2 as an example, the blue-green deployment process is as follows:

[0084] The engine first creates a new version of the microservice instance in the Kubernetes cluster, with the deployment image being v1.2-202405201530. This instance runs in parallel with the old version, and the two environments share the same configuration and storage resources.

[0085] The engine routes the standard data stream with a trustworthiness label to both the old and new versions of the microservice instance simultaneously, continuously collecting the output results of the two instances, boiler efficiency, and scalability uncertainty.

[0086] During the verification process, the overlap range of the uncertainty intervals of the output results from the new and old instances is first determined, requiring an overlap rate of no less than 90%. Secondly, it is determined whether the uncertainty of the output results themselves meets the preset convergence condition, with the expanded uncertainty ≤ 0.3%. If both conditions are met within 5 consecutive minutes, the new version service is deemed to have passed verification.

[0087] After successful verification, the engine switches traffic from the old version instance to the new version instance by modifying the Kubernetes Service routing rules. The switching process takes ≤1 second and the unit does not need to be shut down.

[0088] After the traffic switch is complete, continuously monitor the running status of the new version instance for 30 minutes. If it is running normally, destroy the old version instance and complete the deployment. If an anomaly occurs, immediately roll back the traffic to the old version instance to ensure stable system operation.

[0089] Canary Deployment Implementation: When adding a new smoke exhaust heat loss calculation microservice v1.0, a canary deployment strategy is adopted. First, deploy one canary instance, routing 5% of traffic to this instance, while the remaining 95% of traffic is still handled by the existing alternative functional module. Continuously monitor the accuracy of the canary instance's output results, response time, and resource utilization, ensuring CPU utilization ≤30% and memory utilization ≤40%. If it runs normally for one hour, gradually increase the traffic ratio to 20%, 50%, and 100%, with each traffic adjustment interval of 30 minutes. After all traffic switching is complete, destroy the existing alternative functional module, completing the canary deployment.

[0090] The data quality monitoring module is implemented as follows:

[0091] The data quality monitoring module is implemented using Prometheus + Grafana. It continuously monitors the changing trend of the uncertainty interval of each measurement point data output by the real-time data credibility perception and labeling subsystem. The specific implementation is as follows:

[0092] Threshold settings: Two uncertainty thresholds are preset. The first threshold is the expanded uncertainty at each measurement point. The normal allowable upper limit is set according to the importance of the measuring point. For example, the first threshold of the main steam pressure measuring point is 0.1 MPa, the first threshold of the secondary measuring point such as the ambient temperature measuring point is 1℃, and the second threshold is the emergency threshold. The second threshold of the main steam pressure measuring point is 0.2 MPa, and the second threshold of the ambient temperature measuring point is 2℃.

[0093] Monitoring and Alarm: The data quality monitoring module collects the expanded uncertainty of each measuring point every 10 seconds. When a certain measuring point If the threshold is exceeded for 3 consecutive minutes, an audible and visual alarm is issued via Grafana, and alarm information, including the measurement point number, current uncertainty value, and threshold range, is pushed to the power plant operation management platform; when When the second threshold is exceeded, the module immediately sends a data quality degradation event to the service orchestration and hot deployment engine.

[0094] Alternate calculation model switching: Taking the steam turbine heat consumption calculation microservice as an example, it relies on the main steam pressure measurement point ( The data. When of When the pressure exceeds the second threshold of 0.2 MPa, the service orchestration and hot deployment engine receives a data quality degradation event and immediately switches the microservice instance to a pre-configured standby computing model. The standby computing model reduces its reliance on main steam pressure data and only uses other highly reliable data such as main steam temperature and reheat steam parameters for calculation. Although the calculation accuracy is slightly reduced, it can still ensure that a valid heat rate reference value is output when the data quality is poor.

[0095] In this embodiment, the graphical process orchestration and credibility visualization subsystem is deployed on an industrial server on an edge computing node, presented via a web interface, and supports access from mainstream browsers such as Chrome and Firefox. The specific implementation is as follows:

[0096] Implementation of graphical workflow orchestration: The subsystem provides components including a data source component, a credibility filtering component, a computation microservice component, and a result output component. Users can customize and orchestrate computation workflows by dragging and dropping components and connecting them. The specific operation process is as follows:

[0097] Component Selection and Configuration: After logging into the web interface, users can drag and drop the DCS data source component from the left-hand component library to the workflow editing area. Configure the component's connection parameters, including the OPCUA server address, port number, and authentication information, and select the measurement point data to be collected. Drag and drop the credibility filtering component to the editing area, connect it to the DCS data source component, and set filtering logic rules, such as extended uncertainty ≤ 0.5℃ and a high credibility level, to filter data that meets credibility requirements. Drag and drop the boiler efficiency calculation microservice component, turbine heat consumption calculation microservice component, and comprehensive power supply coal consumption calculation microservice component to the editing area, connecting them sequentially according to the data flow (DCS data source component → credibility filtering component → boiler efficiency calculation microservice component → comprehensive power supply coal consumption calculation microservice component, DCS data source component → credibility filtering component → turbine heat consumption calculation microservice component → comprehensive power supply coal consumption calculation microservice component). Finally, drag and drop the result output component to the editing area, connect it to the comprehensive power supply coal consumption calculation microservice component, and configure the output method: real-time display, historical storage, and report export.

[0098] Workflow saving and running: After completing the workflow orchestration, the user clicks the save button to store the workflow configuration information in the local database (MySQL 8.0). The configuration information includes component types, connection relationships, parameter settings, etc.; when the user clicks the run button, the subsystem parses the workflow configuration information into an executable process script, calls relevant components and microservices through the trusted data-driven bus, and starts the performance calculation process.

[0099] Implementation of Credibility Visualization Function: The credibility visualization subsystem provides end-to-end credibility tracing visualization functionality, presented in the following manner:

[0100] Real-time performance indicator display: In the main display area of ​​the web interface, the calculation results of key performance indicators such as comprehensive power supply coal consumption, boiler efficiency, and turbine heat consumption rate are displayed in real time in the form of line charts. Each data point corresponds to an uncertainty interval, which is represented by a vertical error line centered on the data point. The length of the error line is twice the expanded uncertainty. Users can intuitively view the numerical changes and confidence range of the performance indicators.

[0101] End-to-end reliability tracing: When a user clicks on a performance indicator data point, the system pops up a tracing window, displaying the entire calculation chain of that indicator in flowchart form (data source → data coordination → reliability filtering → various calculation microservices → result output), and annotating the changes in data uncertainty at each stage. For example, clicking on the comprehensive power supply coal consumption data point allows users to view the uncertainty of its input data, as well as the uncertainty propagation process of each input data in its own calculation chain, helping users to locate the key links affecting the reliability of performance indicators.

[0102] Historical data query and analysis: Users can query performance index data and corresponding uncertainty ranges for a specified time period through the time selector. It supports exporting historical data reports in Excel format. The reports include information such as timestamps, performance index values, expanded uncertainty, and confidence level for subsequent offline analysis and optimization.

[0103] In this embodiment, the trusted data-driven bus is implemented based on the EMQX message server and uses the MQTTv3.1.1 protocol for data transmission. The specific implementation is as follows:

[0104] Data transmission function: The standard data stream with credibility labels generated by the real-time data credibility perception and labeling subsystem is published to a specified topic through an MQTT client. Each computing microservice of the dynamically reconfigurable computing subsystem subscribes to the topic as a subscriber to obtain input data. The output data of the computing microservice is published to the topic powerplant / data / result. The graphical process orchestration and credibility visualization subsystem subscribes to the topic to obtain data for visualization.

[0105] Adaptive Scheduling Strategy: The trusted data-driven bus has a built-in data quality event monitoring mechanism. When it receives a data quality degradation event from the data quality monitoring module, it immediately triggers adaptive scheduling of the computing microservices. For example, when the data quality of a certain measurement point degrades, the bus automatically adjusts the data routing rules, marks the data of that measurement point as low-trust and routes it to the corresponding backup computing microservice instance. At the same time, it suspends the transmission of the data to the original computing microservice instance to ensure the stability of the computing process.

[0106] Data transmission reliability assurance: A QoS level 2 message transmission mechanism is adopted to ensure that data is not lost or duplicated; the bus supports message caching function. When the computing microservice instance is temporarily offline, the message server caches the data and resends the data after the instance comes back online. The caching time is configurable, and in this embodiment it is set to 1 hour.

[0107] In this implementation, the cross-platform integration and intelligent application ecosystem subsystem is deployed in the cloud, as detailed below:

[0108] Algorithm Model Library: The algorithm model library is implemented using MinIO object storage service, storing container images of various computing microservices and their corresponding metadata. The metadata is configured according to third-party intelligent application encapsulation specifications, including a unique identifier for the microservice (e.g., boiler-efficiency-calculation-v1.2), a functional description (calculating the thermal efficiency of a 300MW coal-fired power plant boiler), input / output data pattern definitions (input field names, data types, and units; output field names, data types, and units), and input data credibility requirements (for each input field). Resource requirements description (CPU requirement: 2 cores, memory requirement: 4GB, storage requirement: 10GB).

[0109] The Federated Learning Coordination Server is as follows:

[0110] The federated learning coordination server is implemented using the PyTorch Federated Learning framework, coordinating the collaborative training of computational microservices from 10 similar edge nodes of a thermal power plant. Taking the model optimization of the boiler efficiency calculation microservice as an example, the collaborative training process is as follows:

[0111] The cloud-based coordination server initializes global model parameters and sends these initial parameters to the boiler efficiency calculation microservice instances on each edge node.

[0112] Each edge node uses local historical running data (in this embodiment, each node provides 50,000 sets of valid data) to train the local model and calculate the gradient of the model parameters without uploading the original data, thus ensuring data privacy.

[0113] Each edge node encrypts its local parameter gradients and uploads them to the cloud coordination server. The server then uses a federated averaging algorithm to aggregate the gradients of all nodes and update the global model parameters.

[0114] The cloud server distributes the updated global model parameters to each edge node. Each node receives the updated parameters and updates its local model, completing one collaborative training iteration.

[0115] Repeat the above process, iterating 50 times with a 24-hour interval between each iteration, to finally generate a globally optimized model. After receiving the globally optimized model, the service orchestration and hot deployment engine automatically builds a new container image and updates the boiler efficiency calculation microservice instances on each edge node through blue-green deployment, improving calculation accuracy.

[0116] Third-party intelligent application integration: Third-party developers develop a microservice for calculating coal mill pulverizing efficiency according to the platform's defined third-party intelligent application encapsulation specifications, build a container image, and upload it to the cloud algorithm model library, while providing complete metadata information. Authorized users of the power plant can query the metadata of this microservice through the interface of the graphical process orchestration and trustworthiness visualization subsystem. After confirming that its functions match their requirements, they can subscribe to the container image. The system automatically downloads the image to the power plant's edge computing nodes and instantiates and deploys it. Users can directly drag and drop this microservice component into the computing workflow to achieve online calculation of coal mill pulverizing efficiency without modifying the existing system architecture.

[0117] In this embodiment, the internal algorithm of the integrated power supply coal consumption calculation microservice uses the uncertainty interval of the input data to perform probability calculations and realize the propagation of uncertainty. The specific implementation is as follows:

[0118] Comprehensive coal consumption for power supply The calculation function is ,in For boiler efficiency, This refers to the turbine heat rate. The power consumption rate of the plant is expressed by the following function:

[0119]

[0120] in, In this embodiment, the lower heating value of the fuel is used. , is a known constant, and its uncertainty is negligible.

[0121] Output Combined standard uncertainty Calculated using the following formula:

[0122]

[0123] Partial derivative calculation:

[0124]

[0125]

[0126]

[0127] Parameter values: Assuming at a certain moment , ; , ; , Correlation coefficient Boiler efficiency and turbine heat rate show a weak positive correlation. Boiler efficiency has a weak positive correlation with plant power consumption rate. The turbine heat consumption rate is weakly positively correlated with the plant power consumption rate.

[0128] Uncertainty propagation is calculated as follows:

[0129] Calculate the partial derivative values: , , ;

[0130] Calculate the squares of each term: , , ;

[0131] Calculate the cross term: The other two cross terms are calculated similarly;

[0132] Substituting into the formula, we can calculate the result. And then according to ( Calculate the expanded uncertainty The final output is the comprehensive coal consumption for power supply. and its expanded uncertainty .

[0133] In this embodiment, after the platform is deployed, it runs continuously for 30 days, and the running effect is as follows:

[0134] Real-time data processing: The average time from raw data acquisition to the generation of a standard data stream with credibility labels is 80ms, and the average response time of each computing microservice is 45ms, meeting the requirements for real-time performance calculation of thermal power units. Calculation accuracy: The deviation between the calculated boiler efficiency value and the offline test value is ≤0.5%, the deviation between the calculated turbine heat rate value and the design value is ≤1%, and the deviation between the calculated comprehensive power supply coal consumption value and the statistical value is ≤2g / (kW·h), meeting the reference requirements for power plant operation optimization. System flexibility: When upgrading computing microservices through blue-green deployment, the unit does not need to be shut down, the upgrade process does not affect normal operation, and the upgrade time is ≤5 minutes; when adding new computing microservices, they can be quickly integrated through graphical workflow orchestration, with an integration cycle of ≤2 hours. Credibility traceability: The visualization function clearly displays the uncertainty range of each performance indicator and the entire link credibility propagation process, helping operators accurately judge data reliability and providing a reliable basis for operational adjustments.

[0135] This embodiment, through the specific implementation described above, achieves modular, high-precision online performance calculation of thermal power unit operation, and compared with existing technologies, has the following advantages:

[0136] By adopting a microservices and containerized architecture, modular encapsulation and dynamic reconfigurability of computing functions are achieved. Upgrades, replacements, and additions of computing microservices can be completed without system downtime, reducing system maintenance and functional expansion costs and enhancing the platform's long-term viability. Through multimodal data coordination and uncertainty calculation, quantified uncertainty intervals are bound to each data frame, enabling end-to-end reliability traceability of performance indicators, improving the reliability of calculation results, and providing a more credible reference for operational optimization. Employing a cloud-edge collaborative architecture and federated learning technology, collaborative model optimization across multiple power plants is achieved while ensuring data privacy, improving the versatility and accuracy of the computing algorithms; it also supports convenient access for third-party intelligent applications, enriching the platform's functional ecosystem. The implementation of data quality monitoring and adaptive scheduling strategies ensures stable system operation even when data quality at some measurement points is poor, enhancing the platform's robustness.

[0137] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A modular, high-precision online performance calculation platform for the operation of thermal power units, characterized in that, The platform consists of: The real-time data credibility perception and labeling subsystem is used to obtain raw measurement point data from power plant data sources, perform multimodal data coordination processing, and calculate and bind a quantified uncertainty range for each frame of processed data based on data coordination residuals and measurement instrument accuracy, generating a standard data stream with credibility labels. The dynamic reconfigurable computing subsystem based on microservices and containers includes multiple computing microservices that encapsulate specific performance computing functions. Each computing microservice is deployed in an independent container, and its service contract clearly defines the format of input and output data and the required data trustworthiness requirements. The dynamic reconfigurable computing subsystem also includes a service orchestration and hot deployment engine, which is used to dynamically manage the lifecycle of the computing microservices without downtime of the machine group, including blue-green deployment or canary release of services based on container images, and realize online loading, replacement and unloading of computing microservices. The graphical workflow orchestration and credibility visualization subsystem provides a graphical interface for users to customize the orchestration of computing workflows, including data sources, data processing filters, and one or more computing microservices, by dragging and dropping components and connecting them. It also provides full-link credibility traceability visualization of the final performance indicators based on the uncertainty range. The trusted data-driven bus connects the real-time data trustworthiness perception and labeling subsystem, the dynamic reconfigurable computing subsystem, and the graphical process orchestration and visualization subsystem. It is responsible for transmitting the standard data stream with trustworthiness labels and triggering the adaptive scheduling strategy of computing microservices based on data quality events.

2. The modular high-precision online performance calculation platform for the operation process of a thermal power unit according to claim 1, characterized in that, The multimodal data coordination processing in the real-time data credibility perception and labeling subsystem specifically performs the following steps: Receive raw measurement point datasets from the distributed control system. ,in Indicates the first The original measured values ​​of each measuring point This represents the total number of measurement points. According to the preset unit operating condition identification rules, a target coordination strategy is selected from multiple preset data coordination strategies. The data coordination strategy includes at least a coordination strategy based on mechanism constraints and a coordination strategy based on neural network residual prediction. Apply the target coordination strategy to the original measurement point dataset The data is processed to obtain the reconciled dataset. and the corresponding residual vector ,in The observation matrix is ​​constructed based on the correlation between measurement points, where the coordination strategy based on mechanism constraints is solved by solving a constraint optimization problem. Implementation, in which The optimization process must satisfy a set of equality constraints consisting of mass conservation and energy conservation equations, given the covariance matrix of the measurement error. .

3. The modular high-precision online performance calculation platform for the operation process of a thermal power unit according to claim 1, characterized in that, For each frame of processed data, a quantized uncertainty range is calculated and bound in real time, which is achieved through the following process: For the coordinated data Calculate its combined standard uncertainty. The calculation formula is: ,in For the Type B standard uncertainty component determined according to the measuring instrument calibration certificate, Based on the residual vector The Type A standard uncertainty components determined by statistical characteristics are obtained through... Calculation, where For the residual vector corresponding to the first The residuals at each measuring point The standard deviation within the sliding time window. This represents the number of valid samples within the sliding time window. Based on the combined standard uncertainty and the preset inclusion factor Calculate the expanded uncertainty The expanded uncertainty The uncertainty interval is defined. ; The coordinated data and its corresponding expanded uncertainty The data is encapsulated into a unified format data packet to form the standard data stream with the credibility label.

4. The modular high-precision online performance calculation platform for the operation process of a thermal power unit according to claim 1, characterized in that, The input data reliability requirements defined in the service contract of the computational microservice include the maximum allowed expanded uncertainty threshold for the input data item. ; When executing the blue-green deployment or canary release, the service orchestration and hot deployment engine simultaneously routes the standard data stream with the trustworthiness label to both the old and new versions of the computing microservice instance. It then verifies the new version of the service and makes traffic switching decisions by comparing the consistency of the output results of the old and new instances within the overlapping range of uncertainty intervals and determining whether the uncertainty of the output results themselves meets the preset convergence conditions.

5. The modular high-precision online performance calculation platform for the operation process of a thermal power unit according to claim 1, characterized in that, The dynamic reconfigurable computing subsystem also includes a data quality monitoring module, which continuously monitors the uncertainty range of the data at each measurement point output by the real-time data credibility perception and labeling subsystem. An alarm is triggered when the uncertainty of data at a specific measurement point continuously exceeds a preset first threshold. When the uncertainty exceeds a higher second threshold, the data quality monitoring module sends a data quality degradation event to the service orchestration and hot deployment engine, triggering the service orchestration and hot deployment engine to switch the computing microservice instance that depends on the measurement point data to a pre-configured backup computing model with lower requirements for input data quality.

6. The modular high-precision online performance calculation platform for the operation process of a thermal power unit according to claim 1, characterized in that, The platform adopts a cloud-edge collaborative architecture, and the core components of the real-time data credibility perception and labeling subsystem, the dynamic reconfigurable computing subsystem, and the graphical process orchestration and credibility visualization subsystem are deployed on the local edge computing nodes of the power plant. The platform also includes a cloud-based cross-platform integration and intelligent application ecosystem subsystem, which includes an algorithm model library, a federated learning coordination server, and third-party intelligent application encapsulation specifications. The computing microservices on the edge computing nodes interact with the model parameter gradients of similar computing microservices on other power plant edge nodes based on the federated learning coordination server, so as to collaboratively train and generate a global optimization model. The service orchestration and hot deployment engine dynamically updates local computing microservice instances based on the global optimization model.

7. The modular high-precision online performance calculation platform for the operation process of a thermal power unit according to claim 1, characterized in that, The components provided by the graphical process orchestration and credibility visualization subsystem include a data source component, a credibility filtering component, a computational microservice component, and a result output component. The credibility filtering component allows users to set logical rules to route data streams to different downstream computing microservice component branches based on the size of the uncertainty range of the input data.

8. The modular high-precision online performance calculation platform for the operation process of a thermal power unit according to claim 1, characterized in that, The platform includes at least one internal algorithm of the aforementioned computational microservice, which performs probability calculations using the uncertainty interval carried by the input data to achieve uncertainty propagation, for functions with the following relationships: The output of the computational microservices Combined standard uncertainty Propagation calculations are performed using the following formula: in, For function For input quantity The partial derivatives, Input quantity The combined standard uncertainty, Input quantity and The estimated correlation coefficient between them; The computational microservice ultimately outputs the computation result. and its expanded uncertainty .

9. The modular high-precision online performance calculation platform for the operation process of a thermal power unit according to claim 1, characterized in that, When the service orchestration and hot deployment engine performs dynamic management of computing microservices, the container images it uses are stored in a private image repository integrated with a version control system. When the algorithm code library in the version control system is updated and the build pipeline is triggered, a new container image is automatically generated and pushed to the private image repository; After the service orchestration and hot deployment engine detects a new image event, it automatically starts the blue-green deployment or canary release process.

10. A modular high-precision online performance calculation platform for the operation process of a thermal power unit according to claim 6, characterized in that, The third-party intelligent application encapsulation specification specifically stipulates the metadata that the computing microservices accessing the platform must provide. The metadata includes at least: a unique identifier for the microservice, a functional description, an input and output data pattern definition, input data credibility requirements, and a resource requirement description. The intelligent application algorithms encapsulated based on this specification are built into container images and uploaded to the algorithm model library of the cross-platform integration and intelligent application ecosystem subsystem. The graphical process orchestration and credibility visualization subsystem provides an interface for authorized users to subscribe to specified container images from the algorithm model library and instantiate and deploy them to specified edge computing nodes for operation.

Images