A method for integrated centralized monitoring and collaborative control of auxiliary control systems of a thermal power plant

Abstract

The application belongs to the technical field of thermal power plant control, and discloses a kind of integrated centralized monitoring and collaborative control method for auxiliary control system of thermal power plant, and multiple protocol adaptive analysis is realized by wide temperature salt spray resistant embedded edge collection unit, compatible with various manufacturer equipment communication protocol, complete data standardization packaging with IEC 61850 standard, with dynamic calibration and signal isolation, redundant transmission double anti-interference technology, guarantee data stable transmission. Build four-level architecture integrated centralized monitoring platform from field device layer to management decision layer, realize whole auxiliary control subsystem data unified integration, storage and display through double ring fiber redundancy ring network, combined with digital twin virtual-real linkage complete working condition simulation and strategy pre-play, optimize plant-level monitoring information system, management information system interface realizes main and auxiliary system data intercommunication, layered permission and encrypted transmission strengthen data security, change scattered control to integrated centralized monitoring, improve data utilization.

Description

Technical Field

[0001] This invention belongs to the field of thermal power plant control technology, specifically a method for integrated centralized monitoring and collaborative control of auxiliary control systems in thermal power plants. Background Technology

[0002] The auxiliary control system of a thermal power plant is a core component ensuring the safe and stable operation of the unit, achieving environmental emission standards, and optimizing energy efficiency. It comprises multiple independent subsystems, including desulfurization, denitrification, ash removal, coal conveying, circulating water, and chemical water treatment. Its operational status directly impacts the safety, economy, and environmental friendliness of the thermal power plant. Currently, most domestic thermal power plants still employ a traditional decentralized control model for their auxiliary control systems. Each subsystem is designed and deployed independently by different manufacturers, leading to the following technical problems:

[0003] Traditional auxiliary control systems generally adopt the classic control strategy based on PID regulation. This relies heavily on preset fixed parameters and simple feedback loops, making it difficult to adapt to complex operating conditions such as frequent fluctuations in unit load and changes in coal quality. This often leads to response lag and overshoot, resulting in decreased system stability. Furthermore, unplanned outages due to mismatched control parameters account for approximately 30% of all outages. Additionally, the use of different communication protocols by various subsystems and the lack of a unified data standard create data silos and insufficient data utilization.

[0004] While some existing centralized monitoring solutions attempt to integrate various auxiliary control subsystems, they mostly only achieve data acquisition and display functions, without forming an effective collaborative control mechanism. Furthermore, they suffer from problems such as poor protocol compatibility, weak anti-interference capability of data transmission, and low accuracy of anomaly diagnosis.

[0005] Traditional auxiliary control systems lack data privacy protection mechanisms, centralized learning models pose data security risks, and control strategies lack self-optimization capabilities, making it difficult to dynamically adapt to complex operating conditions. Operation and maintenance still rely on human experience, resulting in high operation and maintenance costs and untimely fault handling. Summary of the Invention

[0006] The purpose of this invention is to provide an integrated centralized monitoring and collaborative control method for auxiliary control systems in thermal power plants, so as to solve one or more problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an integrated centralized monitoring and collaborative control method for auxiliary control systems of thermal power plants, comprising the following specific steps; In the data acquisition and preprocessing stage, an embedded edge acquisition unit adapted to the auxiliary control subsystem of the thermal power plant is deployed to preprocess, predict anomalies, and standardize the equipment operating parameters, status signals, and environmental parameters of the auxiliary control system to obtain a standardized dataset. A dataset calibration mechanism and anti-interference transmission guarantee are set up, and a communication interface compatible with industrial scenarios is reserved. During the platform construction phase, an integrated centralized monitoring platform with a layered architecture was built, a redundant communication network was established, data storage and display and a closed-loop mechanism for virtual and real linkage of digital twins were integrated, the interface with the power plant's existing plant-level monitoring information system and management information system was optimized, and a layered permission control and data encryption transmission mechanism was set up. In the strategy modeling stage, a collaborative control framework integrating model predictive control and deep reinforcement learning is constructed, the coupling relationship between auxiliary control subsystems is quantified and control weights are dynamically allocated, a working condition prediction model is constructed, and the reward and punishment function of the control algorithm is designed in combination with multi-dimensional operating indicators. During the diagnostic execution phase, based on the coupling relationship between the collaborative control framework and the subsystem, a hierarchical anomaly diagnosis model integrating CNN and LSTM is constructed to identify the anomaly level of the equipment and match the corresponding self-healing control strategy, link the operation and maintenance system and set up a fault tracing module. In the strategy self-optimization phase, based on the abnormal diagnosis results and fault tracing information, a hierarchical federated learning architecture is constructed. A differentiated data security strategy is adopted through the privacy desensitization and grading module. The collaborative control strategy is optimized by iteratively optimizing the data of each subsystem, and a strategy self-optimization feedback mechanism is set up. During the collaborative scheduling phase, a global collaborative scheduling mechanism is constructed based on the optimized collaborative control strategy and model parameters. A dynamic scheduling priority module is set to realize the linkage operation, data interaction and operation and maintenance assistance between the auxiliary control system, the host system and the plant-level monitoring information system. During the integration and debugging phase, based on the system's coordinated operation data, the functional modules are modularly integrated and phased debugging is carried out. Phased verification indicators are set, full lifecycle operation verification is conducted, and an OTA remote upgrade and compatibility verification mechanism is established.

[0008] Furthermore, the acquisition and preprocessing stage deploys an embedded edge acquisition unit based on a wide temperature range and salt spray resistance, covering all auxiliary control subsystems of the thermal power plant, and integrates a multi-protocol adaptive parsing module. The multi-protocol adaptive parsing module can automatically identify the communication protocols of equipment from different manufacturers and dynamically adapt to protocol version updates.

[0009] The embedded edge acquisition unit collects the operating parameters, status signals and environmental parameters of each auxiliary control device in real time. The edge computing module completes data noise reduction and anomaly removal preprocessing. The edge-side lightweight AI prediction module is based on a small sample training model to identify minor parameter anomalies in advance. Then, the data is standardized and packaged according to the IEC 61850 standard, and the data naming specifications and formats are unified to generate a structured dataset.

[0010] The embedded edge acquisition unit adopts an integrated embedded hardware architecture, with an industrial-grade ARM Cortex-A9 processor at its core. It has a built-in power isolation module, a multi-channel analog signal acquisition module, a digital input / output module, a dedicated protocol conversion chip, and a local cache unit. A single device can simultaneously connect to 16 channels of 4-20mA analog signals and 8 channels of digital switch signals. It supports DIN rail mounting and hot-swappable online replacement, and can be directly deployed in various auxiliary control field cabinets in thermal power plants, such as desulfurization, denitrification, coal conveying, and circulating water systems.

[0011] The dataset dynamic calibration mechanism combines historical data and equipment nameplate parameters to calibrate the accuracy of the collected data in real time. At the same time, it adopts dual anti-interference technology of signal isolation and redundant transmission to improve the anti-interference capability of data transmission. It also reserves a 5G / industrial Ethernet dual-mode communication interface to adapt to the subsequent network upgrade needs of thermal power plants.

[0012] Furthermore, during the platform construction phase, an integrated centralized monitoring platform with a four-level architecture consisting of a field equipment layer, a control execution layer, a monitoring and scheduling layer, and a management decision-making layer is constructed. Redundant Ethernet is used as the communication backbone, and a dual-ring fiber optic redundant ring network is built.

[0013] The dual-ring fiber optic redundant ring network adopts independent physical cabling for the main transmission ring and the backup transmission ring. The two ring networks do not interfere with each other and operate independently. All nodes in the ring network support the ERPS Ethernet ring network protection protocol. When abnormal situations such as fiber breakage or node failure occur in the ring network, the system completes automatic link switching and route reconstruction within 50ms, and the transmission rate stably reaches 1000Mbps, meeting the synchronous transmission requirements of real-time data, video data, and control commands.

[0014] The integrated centralized monitoring platform integrates a real-time database and a historical data archiving module, and connects to the standardized calibration dataset output in the data acquisition and preprocessing stage. It realizes the unified integration, storage and display of data from all auxiliary control subsystems. It constructs a full life cycle virtual simulation model of each auxiliary control subsystem using digital twin virtual-real linkage closed-loop technology. The platform can perform operating condition simulation and control strategy pre-play through the virtual simulation model, and use the pre-play results to correct the control parameters of physical equipment.

[0015] The real-time database uses in-memory storage mode, with millisecond-level response for real-time data read and write, and a storage period of 7 days; the historical data archiving module uses disk array storage, and archives data according to device number, time node, and parameter type, with a storage period of no less than 3 years, and supports fast retrieval by time, device, and parameter type.

[0016] Meanwhile, the interface design with the plant-level monitoring information system and management information system is optimized, and a standardized data interaction protocol is adopted to achieve data interoperability with the power plant's main system. The hierarchical permission control module assigns different data viewing and control operation permissions according to the job responsibilities of operation and maintenance personnel, dispatchers, and managers, and strengthens data security by combining data encryption transmission technology.

[0017] Furthermore, the strategy modeling stage is based on the integrated centralized monitoring platform built in the platform construction stage and the data collected from various auxiliary control subsystems. It designs a collaborative control framework that integrates model predictive control and deep reinforcement learning. The auxiliary control subsystem coupling factor dynamic allocation module is based on the standardized historical data collected and preprocessed in the acquisition stage. It constructs dynamic mathematical models of each auxiliary control subsystem, quantifies the coupling relationship between subsystems, and dynamically allocates the control weights of model predictive control and deep reinforcement learning under different coupling scenarios.

[0018] Coupling relationship quantification includes four core dimensions: water temperature correlation coupling, load matching coupling, energy consumption linkage coupling, and environmental protection synergy coupling. By statistically analyzing the correlation degree of parameters of each subsystem through massive historical operation data, scenarios with a correlation degree ≥ 0.8 are identified as strongly coupled scenarios, those with a correlation degree between 0.5 and 0.8 are identified as medium coupled scenarios, and those with a correlation degree below 0.5 are identified as weakly coupled scenarios. The control weights of model predictive control and deep reinforcement learning are dynamically adjusted according to the coupling level.

[0019] The working condition prediction model uses the LSTM algorithm to mine the variation patterns of historical load, coal quality, and environmental parameters, and predict the trend of working condition changes. This enables the model predictive control algorithm to plan control objectives in advance, and the deep reinforcement learning algorithm to complete parameter optimization preparation in advance.

[0020] The deep reinforcement learning algorithm adopts a lightweight network design and incorporates reward and penalty functions based on multiple dimensions such as energy consumption, environmental compliance rate, and equipment wear and tear.

[0021] Furthermore, the diagnostic execution phase, based on the collaborative control framework and quantified subsystem coupling relationships established in the strategy modeling phase, integrates CNN and LSTM deep learning algorithms to construct a hierarchical anomaly diagnosis model. Based on real-time data from the centralized monitoring platform, it identifies problems such as auxiliary control equipment failures, parameter anomalies, and pipeline leaks, and classifies them into three fault levels: minor anomalies, general faults, and serious faults.

[0022] The fusion structure is: CNN → LSTM → softmax classification layer. CNN is used to extract local features, and LSTM is used to extract temporal features. The specific training steps are as follows: ≥1000 labeled samples are normalized; Adam optimizer is used with a learning rate of 0.001, a batch size of 32, and 150 iterations; training is completed with an accuracy of ≥98%; the input is real-time parameters from multiple devices, and the output is the anomaly level and type.

[0023] For anomalies of different levels, an adaptive and self-healing control strategy is designed. Minor anomalies are self-healed by automatically adjusting control parameters, while fluctuation patterns are recorded and fed back into the strategy self-optimization module. General faults trigger automatic switching of backup equipment, synchronously adjusting the operating status of related subsystems and adopting dynamic load distribution technology for related subsystems. Severe faults immediately issue alarm signals, synchronously push the optimal emergency response plan, link the operation and maintenance system to generate maintenance work orders, and locate the root cause of the fault.

[0024] Furthermore, in the strategy self-optimization phase, based on the abnormal diagnosis results, fault tracing information, and strategy feedback requirements of the diagnosis execution phase, the hierarchical federated learning framework constructs a three-level hierarchical federated learning architecture including edge nodes, regional nodes, and central nodes. The edge nodes are used for local data training, the regional nodes are used for parameter aggregation of similar subsystems, and the central nodes are used for global model optimization. Each layer adopts a different encryption mechanism to achieve the sharing and collaborative optimization of model parameters of multiple subsystems.

[0025] After completing local training, edge nodes encrypt and upload model parameters. Regional nodes perform weighted aggregation of parameters from subsystems of the same type. The central node integrates parameters from the entire system to optimize the global model. After optimization, the latest model parameters are encrypted and sent back to each edge node. Edge nodes automatically load and update their local models.

[0026] The local model training at edge nodes is set with termination criteria. When the model's diagnostic accuracy, control response precision, and parameter prediction error reach the preset qualified indicators for multiple consecutive iterations, and the model loss value stabilizes and no longer decreases, the local training process is automatically terminated. If the preset indicators are not reached, the system automatically expands the local running sample data and continues iterative training until the indicators are met before performing the encrypted upload of model parameters.

[0027] The privacy desensitization and grading module adopts differentiated desensitization strategies for different types of auxiliary control data. Sensitive data is protected by both encryption desensitization and access control, while ordinary data is desensitized by lightweight methods. Based on the collaborative control framework in the strategy modeling stage, the module utilizes historical and real-time operating data from each auxiliary control subsystem and continuously optimizes the collaborative control strategy through local training at edge nodes and parameter aggregation in the cloud. This allows for dynamic adaptation to complex operating conditions such as coal quality fluctuations, environmental changes, and load adjustments.

[0028] Simultaneously, an energy consumption analysis module is integrated to mine energy efficiency optimization space based on historical data, automatically adjust the operating parameters of auxiliary control equipment, and the strategy self-optimization feedback mechanism inputs the self-healing results and fault tracing information of the diagnosis execution stage back into the federated learning model.

[0029] Furthermore, the coordinated scheduling phase constructs a global coordinated scheduling mechanism based on the coordinated control strategy optimized in the strategy self-optimization phase and the model parameters of each subsystem, and a dynamic scheduling priority module dynamically allocates the scheduling priority of each auxiliary control subsystem according to the power plant's operating objectives.

[0030] When the load on the main system is adjusted, the centralized monitoring platform issues optimization instructions to each auxiliary control subsystem in real time through the collaborative control framework to adjust the operating status of the auxiliary control equipment. When an abnormality occurs in the auxiliary control system, it synchronously feeds back to the main system to adjust the main system's operating parameters. At the same time, it connects to the plant-level monitoring information system to push the auxiliary control system's operating data, energy consumption data, and fault data synchronously, generating energy efficiency benchmarking reports and equipment health scores. The management decision support module provides suggestions for operation and maintenance optimization, equipment replacement, and energy consumption control based on data mining.

[0031] The host and auxiliary control system are linked to set an abnormal feedback level trigger threshold. Only when the auxiliary control system experiences an abnormal problem of general fault or serious fault level, it immediately sends a real-time fault feedback signal to the host system, triggering the host system parameter adjustment and operation protection logic. Minor abnormalities are only self-healed by adjusting parameters locally and logging, without sending a feedback signal to the host system, so as to avoid frequent minor abnormalities interfering with the stable operation of the host system.

[0032] Furthermore, the integrated debugging phase, based on the global collaborative scheduling mechanism and the operating data of each module built in the collaborative scheduling phase, adopts a modular integration approach to integrate the edge acquisition unit, centralized monitoring platform, collaborative control framework, self-healing control module, and strategy self-optimization module of the aforementioned phase;

[0033] On-site debugging is carried out in stages. First, the debugging and optimization of a single subsystem are completed, then the collaborative debugging between subsystems is carried out, and finally the linkage debugging with the host system and management information system is achieved, and the indicators are verified in stages.

[0034] After debugging, full lifecycle verification is performed, system operation data is collected, control strategies and diagnostic accuracy are optimized through federated learning models, and a system upgrade mechanism is established to support OTA remote upgrades and function expansion.

[0035] The beneficial effects of this invention are as follows:

[0036] 1. This invention achieves multi-protocol adaptive parsing through a wide-temperature, salt-spray resistant embedded edge acquisition unit, compatible with communication protocols of various manufacturers' equipment. Data is standardized and encapsulated according to the IEC 61850 standard. Combined with dynamic calibration and signal isolation, and redundant transmission dual anti-interference technologies, it ensures stable data transmission. It constructs an integrated centralized monitoring platform with a four-level architecture from the field equipment layer to the management decision-making layer. Through a dual-ring fiber optic redundant ring network, it achieves unified integration, storage, and display of data from all auxiliary control subsystems. Combined with digital twin virtual-physical linkage, it completes operational condition simulation and strategy pre-playing. It optimizes the interfaces of the plant-level monitoring information system and management information system to achieve data interoperability between the main and auxiliary systems. Layered permissions and encrypted transmission enhance data security, transforming decentralized control into integrated centralized monitoring and improving data utilization.

[0037] 2. This invention constructs a collaborative control framework that integrates model predictive control and deep reinforcement learning. It quantifies the coupling relationship of auxiliary control subsystems such as desulfurization and circulating water and dynamically allocates control weights. It combines the LSTM algorithm to mine the patterns of load, coal quality, and environmental parameters to achieve advance prediction of operating conditions, enabling the control strategy to adapt to operational fluctuations in advance. It integrates CNN and LSTM to build a hierarchical anomaly diagnosis model to identify equipment failures, parameter anomalies, and other problems. It classifies three levels of faults and matches adaptive self-healing strategies. Minor anomalies are adjusted and self-healed, general faults are automatically switched to standby machines, and serious faults are linked to maintenance and work orders are generated. The fault tracing module locates the root cause.

[0038] 3. This invention establishes a three-tiered federated learning architecture at the edge, region, and center levels, employing differentiated encryption and privacy desensitization strategies. Sensitive data is dual-protected, while ordinary data undergoes lightweight processing. This achieves collaborative optimization of multi-subsystem models while ensuring data privacy. Through a strategy self-optimization feedback mechanism, combined with fault tracing and self-healing data-driven continuous iterative control strategies, it dynamically adapts to complex changes in coal quality, load, and environment. An integrated energy consumption analysis module automatically optimizes equipment parameters, uncovering potential for energy efficiency improvement. A global collaborative scheduling mechanism enables deep linkage between auxiliary control, host, and management information systems. Supported by OTA remote upgrades and full lifecycle verification, the system continuously iterates and expands, reducing manual maintenance costs and improving power plant energy efficiency and environmental compliance rates. Attached Figure Description

[0039] Figure 1 This is the overall flowchart of the integrated centralized monitoring and collaborative control system for auxiliary control systems in thermal power plants according to the present invention;

[0040] Figure 2 This is a flowchart of the graded abnormality diagnosis and self-healing control sub-process of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] like Figures 1 to 2 As shown in the figure, this embodiment of the invention provides an integrated centralized monitoring and collaborative control method for auxiliary control systems in thermal power plants, including the following specific steps;

[0043] In this embodiment of the invention, the acquisition and preprocessing stage deploys an embedded edge acquisition unit based on a wide temperature range (-40℃~85℃) and salt spray resistance, covering all auxiliary control subsystems of the thermal power plant. The auxiliary control subsystems include integrated multi-protocol adaptive parsing modules for desulfurization, denitrification, ash removal, coal conveying, circulating water, and chemical water treatment. The multi-protocol adaptive parsing module can automatically identify communication protocols of different manufacturers' equipment, including ModbusRTU / TCP, Profibus-DP, and CANopen, and dynamically adapt to protocol version updates.

[0044] The multi-protocol adaptive parsing module has a built-in protocol parsing library and an automatic matching engine. After power-on, it actively sends protocol probe frames and quickly identifies mainstream industrial protocols such as Modbus RTU / TCP, Profibus-DP, and CANopen through response characteristics. It automatically loads the corresponding parsing rules and automatically reads the version information and updates the parsing rules online when the protocol version is updated. It can complete the adaptation of new protocols without manual configuration. When multiple protocols are connected at the same time, a priority scheduling mechanism is adopted, prioritizing the parsing of control command protocols and then processing data acquisition protocols. When protocol frame conflicts occur, low-priority data is automatically cached and transmitted later when the channel is idle.

[0045] The embedded edge acquisition unit collects the operating parameters, status signals, and environmental parameters of each auxiliary control device in real time, such as flow, pressure, temperature, and current. The edge computing module completes data noise reduction and anomaly removal preprocessing. The lightweight AI prediction module on the edge side is based on a small sample training model to identify minor parameter anomalies in advance, achieving dual protection of preprocessing and prediction. Then, the data is standardized and packaged according to the IEC 61850 standard, and the data naming specifications and formats are unified to generate a structured dataset.

[0046] The data standardization and encapsulation follows the IEC 61850 standard specification and adopts an object-oriented data modeling approach. The operating parameters and status signals of auxiliary control equipment are divided into a three-level structure of logical node class, data object class, and data attribute class. The camelCase naming rule is uniformly adopted, and the data format is uniformly converted into XML structured text. Private protocol fields of different equipment manufacturers are removed, and a standardized dataset compatible with the entire plant system is generated to ensure that the data can be directly recognized and called by the monitoring platform.

[0047] The multi-protocol adaptive parsing module adopts a lightweight one-dimensional CNN network with the following layers: input layer (single-dimensional temporal parameters) → convolutional layer → pooling layer → fully connected layer. The specific training steps are as follows: collect a small sample set (≥50 samples), 8:2 training set and validation set; learning rate is 0.001, batch size is 8, and iteration is 50 rounds; training is completed when the validation accuracy is ≥95%; the input is single-device temporal parameters, and the output is the normal / abnormal prediction result.

[0048] The dataset dynamic calibration mechanism combines historical data and equipment nameplate parameters to calibrate the accuracy of the collected data in real time, laying a high-precision data foundation for subsequent centralized monitoring and collaborative control. At the same time, it adopts dual anti-interference technology of signal isolation and redundant transmission to improve the anti-interference capability of data transmission and reserves a 5G / industrial Ethernet dual-mode communication interface.

[0049] Dynamic calibration employs a dual-mode approach: real-time comparison and periodic calibration. Real-time calibration uses the rated parameters on the device's factory nameplate as the baseline, comparing the collected data with the baseline in real time and automatically correcting any deviations exceeding the threshold. Periodic calibration uses historical stable operating data of the device as a reference, automatically performing full-parameter calibration every 24 hours. The calibration process does not affect the device's normal data acquisition and transmission, and the calibration results are automatically recorded in the data log.

[0050] Signal isolation employs a dual isolation scheme of magnetic coupling isolation and optoelectronic isolation to block electromagnetic interference, surge impact, and ground potential difference signals generated by on-site frequency converters and power cables from entering the acquisition unit, thus avoiding data distortion and equipment damage. Redundant transmission adopts a dual-channel independent transmission architecture with the main channel responsible for regular data transmission and the backup channel in real-time hot standby. When the main channel is interrupted, loses packets, or experiences excessive interference, the system automatically switches to the backup channel within 10ms.

[0051] The platform construction phase establishes an integrated centralized monitoring platform with a four-level architecture: field equipment layer, control execution layer, monitoring and scheduling layer, and management decision-making layer. Redundant Ethernet is used as the communication backbone, and a dual-ring fiber optic redundant ring network is built.

[0052] In this embodiment of the invention, the integrated centralized monitoring platform integrates a real-time database and a historical data archiving module, and interfaces with the standardized calibration dataset output from the data acquisition and preprocessing stage. This enables unified integration, storage, and display of data from all auxiliary control subsystems. A virtual simulation model of the entire lifecycle of each auxiliary control subsystem is constructed using digital twin virtual-physical linkage closed-loop technology. The platform can perform operating condition simulation and control strategy pre-playing through the virtual simulation model, and use the pre-playing results to correct the control parameters of the physical equipment. This forms a closed-loop control system of virtual pre-playing, physical execution, data feedback, and model optimization, intuitively presenting the equipment operating status, pipeline flow direction, and parameter change trends.

[0053] After completing the strategy pre-play, the virtual model generates the optimal combination of control parameters and pushes it to the physical device controller through a standardized interface. The controller automatically receives and replaces the original control parameters. The parameter distribution process is encrypted and verified throughout to ensure that the parameters are executed accurately. After reverse adjustment, the virtual model synchronously tracks the operating status of the physical device.

[0054] The synchronization of virtual and physical data between the digital twin virtual model and the physical device adopts a dual mechanism of timed synchronization and event-triggered synchronization. Under normal operating conditions, timed synchronization performs a full parameter synchronization refresh every 500 milliseconds. When the device experiences a sudden change in state, a control command is issued, a fault is triggered, or a parameter exceeds the limit, the real-time synchronization process is immediately triggered to update the virtual model state in the first time, ensuring that the virtual scene is consistent with the operating data, device status, and parameter values ​​of the physical device.

[0055] The virtual simulation model uses 3D CAD drawings of field equipment, real-time collected operating parameters, equipment history, and pipeline layout drawings as core data sources. It builds a full-scene virtual model of the auxiliary control system at a 1:1 scale, which fully restores the equipment structure, pipeline flow direction, operating status, and parameter changes. The control strategy pre-simulation results are synchronously refreshed to the physical equipment controller every 500ms. The virtual model and the physical equipment interact bidirectionally and synchronously in real time, intuitively displaying changes in operating conditions, parameter fluctuations, and control effects.

[0056] Meanwhile, the interface design with the plant-level monitoring information system and management information system is optimized, and a standardized data interaction protocol is adopted to achieve data interoperability with the power plant's main system. The hierarchical permission control module assigns different data viewing and control operation permissions according to the job responsibilities of operation and maintenance personnel, dispatchers, and managers, and strengthens data security by combining data encryption transmission technology.

[0057] The data encryption transmission adopts a dual encryption method of link layer encryption and application layer encryption. The link layer uses a symmetric encryption algorithm to ensure the security of the transmission channel, while the application layer uses an asymmetric encryption algorithm to encrypt sensitive operational data. The entire process of data uploading, downloading, and interaction is encrypted.

[0058] The management information system divides operation permissions into a three-tier management system. Maintenance personnel only have the authority to view equipment status, confirm on-site parameters, and report fault information, but no authority to issue control commands. Dispatch personnel have the authority to issue control commands, adjust operating strategies, and control subsystem linkage, but no authority to configure the system or allocate permissions. Management personnel have full authority to configure system parameters, allocate permissions, manage strategies, and query logs. All operations use dual authentication of account password and dynamic password, and operation records are fully traceable.

[0059] In this embodiment of the invention, the strategy modeling stage is based on the integrated centralized monitoring platform built in the platform construction stage and the data collected from various auxiliary control subsystems. A collaborative control framework integrating model predictive control and deep reinforcement learning is designed. The auxiliary control subsystem coupling factor dynamic allocation module is based on the standardized historical data collected and preprocessed in the acquisition stage to construct dynamic mathematical models of each auxiliary control subsystem and quantify the coupling relationship between subsystems. The coupling relationship includes water temperature coordination between the desulfurization system and the circulating water system, and load matching between the coal conveying system and the boiler pulverizing system. The control weights of MPC model predictive control and DRL deep reinforcement learning are dynamically allocated under different coupling scenarios. For example, when the load fluctuates drastically, the DRL weight is increased to achieve rapid adaptive adjustment; when running in steady state, the MPC weight is increased to ensure control accuracy.

[0060] In the strongly coupled scenario, the weights of deep reinforcement learning are increased, and the control output is adaptively adjusted based on real-time operating conditions. In the medium coupled scenario, the weights of the two algorithms are evenly distributed to balance response speed and control accuracy. In the weakly coupled scenario, the weights of model prediction control are increased, and stable control is achieved through prediction, resulting in a smooth transition of control output.

[0061] Based on the control weights assigned to different coupling scenarios, the parameters of the model-predicted control output are weighted and fused with the adaptive adjustment parameters output by deep reinforcement learning to generate the final control command sent to the equipment. During the fusion process, a built-in rate limiting and gradual adjustment module controls the amplitude and rate of change of the control output to avoid mechanical shock or operational fluctuations to the field auxiliary control equipment caused by sudden changes in control parameters, ensuring that the equipment is always in a stable operating state.

[0062] DRL uses the DDPG algorithm: Actor network and Critic network. The specific training steps are as follows: experience replay pool capacity 10000; reward function r = 0.6 × energy efficiency + 0.3 × environmental protection - 0.1 × loss; Actor learning rate 0.0005, Critic 0.001, γ = 0.95, 200 iterations; MPC prediction step size 5, control step size 1; DRL weight 0.7 when load fluctuation ≥ 10%, and steady-state MPC weight 0.8.

[0063] The operating condition prediction model uses the LSTM algorithm to mine the changing patterns of historical load, coal quality, and environmental parameters, and predicts the changing trends of operating conditions. This enables the Model Predictive Control (MPC) algorithm to plan control objectives in advance, and the deep reinforcement learning algorithm to complete parameter optimization preparation in advance, thus achieving advanced coordination between the auxiliary control system and the main generator system.

[0064] Model hierarchy: Input layer (3-dimensional time series: load, coal quality, temperature) → LSTM layer (dropout=0.2) → fully connected layer; Training steps: Data normalization to 0-1; Adam optimizer, learning rate 0.002, batch size 16, 100 iterations; MSE loss function, convergence with prediction error ≤3%; Input is historical multi-dimensional parameters, output is the predicted load value for the next 1 hour.

[0065] The deep reinforcement learning algorithm adopts a lightweight network design to reduce the difficulty of deployment at the edge. At the same time, it combines multi-dimensional indicators such as energy consumption, environmental compliance rate, and equipment wear and tear to design reward and punishment functions, thereby improving the global optimization capability of the control strategy.

[0066] In this embodiment of the invention, the diagnostic execution stage is based on the collaborative control framework and quantified subsystem coupling relationship established in the strategy modeling stage, integrates CNN and LSTM deep learning algorithms, constructs a hierarchical anomaly diagnosis model, and identifies problems such as auxiliary control equipment failure, parameter anomaly, and pipeline leakage based on real-time data from the centralized monitoring platform, classifying them into three fault levels: minor anomaly, general fault, and serious fault.

[0067] In the hierarchical anomaly diagnosis model, the CNN module is used to extract local mutation features, waveform distortion features, and numerical jump features of equipment operating parameters, while the LSTM module is used to extract long-term time-series change trends, periodic fluctuation patterns, and linkage change features of related parameters. The feature data extracted by the two modules are deeply fused and then input into the classification layer to complete the anomaly type identification and fault level determination. The comprehensiveness of anomaly diagnosis is improved through the complementarity of the two features, avoiding the problems of missed judgment and misjudgment.

[0068] Minor anomalies are slight deviations in parameters, with no risk of equipment downtime and no impact on continuous system operation; general faults are abnormal shutdowns of a single device, which can be restored by switching to a backup device and do not affect the operation of the host; serious faults are multi-device linkage faults, which pose safety risks and may lead to reduced load or shutdown of the host. The severity level is determined based on a comprehensive assessment of real-time parameters, equipment status, and associated impacts.

[0069] For anomalies of different levels, an adaptive and self-healing control strategy is designed. Minor anomalies achieve self-healing by automatically adjusting control parameters, while recording fluctuation patterns and inputting them back into the strategy self-optimization module. General faults such as single device shutdowns trigger automatic switching of backup equipment, synchronously adjusting the operating status of related subsystems, and employing dynamic load distribution technology for related subsystems to reduce the impact on the main system. Severe faults immediately issue alarm signals, synchronously push the optimal emergency response plan, and link the operation and maintenance system to generate maintenance work orders. The fault tracing module locates the root cause of the fault based on fault data and historical cases. This achieves full-process control of anomaly diagnosis, automatic response, self-healing recovery, and fault tracing, reducing the risk of unplanned downtime.

[0070] The fault tracing module uses real-time fault data, equipment operation history, historical fault cases, and related subsystem operation data as the basis for analysis. It adopts a combination of forward tracing and reverse deduction to first locate the time point of the fault and related equipment, then investigate the inducing factors such as parameter fluctuations, equipment start-up and shutdown, and external interference, and finally locate the root cause of the fault. The tracing results are automatically generated into a report and synchronized to the operation and maintenance system.

[0071] The criteria for minor anomalies are that the operating parameters deviate from the rated value within ±5%, which does not affect the normal operation of the equipment. The system can quickly self-heal by automatically fine-tuning the control parameters, and the parameter adjustment range is controlled within ±10% of the rated value to avoid large fluctuations. General faults are identified as problems such as the shutdown of a single auxiliary control device or operation interruption. The backup equipment completes automatic start-up and commissioning switching within 3 seconds. At the same time, the load of related subsystems is smoothly distributed and dynamically adjusted according to the system operation requirements.

[0072] In this embodiment of the invention, the strategy self-optimization stage is based on the abnormal diagnosis results, fault tracing information, and strategy feedback requirements of the diagnosis execution stage. A hierarchical federated learning framework is constructed to create a three-level hierarchical federated learning architecture comprising edge nodes, regional nodes, and a central node. The edge nodes are used for local data training, the regional nodes are used for parameter aggregation of similar subsystems, and the central node is used for global model optimization. Each layer employs a different encryption mechanism to achieve the sharing and collaborative optimization of model parameters across multiple subsystems. Edge nodes refer to each auxiliary control subsystem, and the encryption mechanisms include lightweight symmetric encryption at the edge and homomorphic encryption at the central.

[0073] The privacy desensitization and grading module adopts differentiated desensitization strategies for different types of auxiliary control data. Sensitive data includes desulfurization agent formulas and core equipment parameters, while ordinary data includes ambient temperature and routine operating parameters. Sensitive data is protected by both encryption desensitization and access control, while ordinary data is desensitized using lightweight methods. The privacy desensitization and grading module takes into account both privacy protection and data utilization efficiency. Based on the collaborative control framework in the strategy modeling stage, it utilizes historical and real-time operating data from each auxiliary control subsystem and continuously optimizes the collaborative control strategy through local training at edge nodes and parameter aggregation in the cloud. This allows for dynamic adaptation to complex operating conditions such as coal quality fluctuations, environmental changes, and load adjustments.

[0074] The privacy desensitization and grading module first classifies and labels the auxiliary control data, marking core equipment parameters, reagent ratios, control strategies, and other data as sensitive data, and employing triple desensitization protection through data encryption, field masking, and permission binding; while marking ambient temperature, normal operating status, and other data as ordinary data, and employing lightweight desensitization processing through data normalization and feature extraction, thereby maximizing the preservation of data value while ensuring data privacy and security.

[0075] Simultaneously, an energy consumption analysis module is integrated to mine energy efficiency optimization space based on historical data, automatically adjust the operating parameters of auxiliary control equipment, and input the self-healing results and fault tracing information of the diagnostic execution stage back into the federated learning model to achieve self-optimization of strategy optimization, execution feedback, and model iteration.

[0076] The energy consumption analysis module collects and statistically analyzes energy consumption data from all auxiliary control equipment in real time, including power consumption, water consumption, and chemical consumption. It establishes an equipment energy consumption benchmark library and a historical best energy consumption comparison model. It automatically identifies equipment and subsystems with high energy consumption and low operating efficiency. It also adaptively adjusts key operating parameters such as equipment operating frequency, valve opening, and start-stop sequence based on real-time operating conditions to continuously optimize equipment operating status and reduce energy consumption deviation per unit of power generation.

[0077] In this embodiment of the invention, the collaborative scheduling phase is based on the collaborative control strategy optimized in the strategy self-optimization phase and the model parameters of each subsystem to construct a global collaborative scheduling mechanism, realizing deep linkage between auxiliary control, main control, and management. The dynamic scheduling priority module dynamically allocates the scheduling priority of each auxiliary control subsystem according to the power plant's operating objectives, such as environmental protection priority, energy efficiency priority, and safety priority. This includes increasing the scheduling priority of desulfurization and denitrification systems when there is high pressure to meet environmental standards, and increasing the scheduling priority of circulating water and coal conveying systems during deep peak shaving.

[0078] The scheduling priority is sorted according to the core logic of safe operation, environmental compliance, optimal energy efficiency, and convenient operation and maintenance. When the power plant triggers different operational objectives such as environmental control, deep peak shaving, and emergency response, the system automatically switches the corresponding scheduling priority, giving priority to responding to the control commands and operational requirements of higher-level subsystems. Lower-level commands are executed with a delay of no more than 1 second. When commands conflict, lower-priority commands are automatically blocked to ensure that the core operational objectives of the power plant are achieved first.

[0079] When the load on the main control system is adjusted, the centralized monitoring platform issues optimization instructions to each auxiliary control subsystem in real time through the collaborative control framework to adjust the operating status of the auxiliary control equipment. The operating status of the auxiliary control equipment includes increasing the circulating water flow and optimizing the desulfurization and denitrification parameters in advance when the load increases. When the auxiliary control system malfunctions, it is synchronously fed back to the main control system to adjust the main control operating parameters and prevent the impact of the fault from expanding. At the same time, it connects to the plant-level monitoring information system to push the operating data, energy consumption data and fault data of the auxiliary control system in a synchronous manner, generate energy efficiency benchmarking reports and equipment health scores, and the management decision support module provides suggestions for operation and maintenance optimization, equipment replacement and energy consumption control based on data mining.

[0080] The platform first verifies the legality and security of the instructions, then splits the instructions according to the subsystem priority and sends them out synchronously through redundant communication links. After receiving the instructions, the subsystem immediately sends back confirmation. The execution process transmits the status back in real time, and the results are reported after the instructions are completed. Instructions that are not executed normally are automatically resent once.

[0081] The host and auxiliary control system interact in a two-way real-time communication mode. The auxiliary control system pushes the operating status, fault information, and energy consumption data to the host system in real time, and the host system sends load commands and operating targets to the auxiliary control system synchronously. The data exchange frequency between the two parties is once per second, and the command response delay does not exceed 500ms, so as to avoid the operational risks caused by mismatch of operating conditions.

[0082] In this embodiment of the invention, the integration and debugging phase, based on the global collaborative scheduling mechanism and the operational data of each module built in the collaborative scheduling phase, adopts a modular integration approach to integrate the edge acquisition unit, centralized monitoring platform, collaborative control framework, self-healing control module, and strategy self-optimization module from the aforementioned phases, thereby reducing transformation costs.

[0083] On-site debugging is carried out in stages. First, the debugging and optimization of a single subsystem are completed. Then, the collaborative debugging between subsystems is carried out. Finally, the linkage debugging with the host system and management information system is achieved. Stage-by-stage verification indicators are set. For example, the debugging of a single subsystem must meet the standards for data acquisition accuracy and control response delay. The collaborative debugging must meet the standards for linkage response delay of subsystems.

[0084] The single subsystem debugging and verification indicators include a data acquisition accuracy rate of no less than 99.5% and a control command response delay of no more than 1 second; the inter-subsystem collaborative debugging and verification indicators include a linkage response time of no more than 2 seconds, no packet loss in data interaction, and no conflict in control logic; the linkage debugging and verification indicators with the host and management system include normal data communication across the entire system, error-free command execution, and accurate handling of abnormal linkages. Debugging and operation can only be completed after all indicators meet the standards.

[0085] After commissioning, long-term full lifecycle verification is carried out, continuously collecting system operation data, and continuously optimizing control strategies and diagnostic accuracy through federated learning models. At the same time, a system upgrade mechanism is established to support OTA remote upgrades and function expansion. The upgrade compatibility verification module ensures the compatibility of the upgraded system, thereby improving the safety, economy and environmental protection of power plant operation.

[0086] The full lifecycle verification adopts a long-term online operation mode, continuously collecting data on equipment operation, fault handling, and strategy execution, iteratively updating the federated learning model monthly, and conducting system performance retests every quarter to compare and verify control accuracy, diagnostic accuracy, and energy consumption optimization effects.

[0087] The OTA remote upgrade adopts a stable transmission mode with packet-encrypted transmission and breakpoint resume. Before the upgrade, the system configuration, control strategy, historical data and other core information are automatically and completely backed up. The upgrade process monitors the progress and status in real time. If the upgrade fails or is interrupted, the system can automatically roll back to the original stable version within 1 minute. The compatibility verification covers the communication protocols, hardware models and software versions of all connected auxiliary control devices.

[0088] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0089] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for integrated centralized monitoring and collaborative control of auxiliary control systems in thermal power plants, characterized in that, The specific steps include the following: In the data acquisition and preprocessing stage, an embedded edge acquisition unit adapted to the auxiliary control subsystem of the thermal power plant is deployed to preprocess, predict anomalies, and standardize the equipment operating parameters, status signals, and environmental parameters of the auxiliary control system to obtain a standardized dataset. A dataset calibration mechanism and anti-interference transmission guarantee are set up, and a communication interface compatible with industrial scenarios is reserved. During the platform construction phase, an integrated centralized monitoring platform with a layered architecture was built, a redundant communication network was established, data storage and display and a closed-loop mechanism for virtual and real linkage of digital twins were integrated, the interface with the power plant's existing plant-level monitoring information system and management information system was optimized, and a layered permission control and data encryption transmission mechanism was set up. In the strategy modeling stage, a collaborative control framework integrating model predictive control and deep reinforcement learning is constructed, the coupling relationship between auxiliary control subsystems is quantified and control weights are dynamically allocated, a working condition prediction model is constructed, and the reward and punishment function of the control algorithm is designed in combination with multi-dimensional operating indicators. During the diagnostic execution phase, based on the coupling relationship between the collaborative control framework and the subsystem, a hierarchical anomaly diagnosis model integrating CNN and LSTM is constructed to identify the anomaly level of the equipment and match the corresponding self-healing control strategy, link the operation and maintenance system and set up a fault tracing module. In the strategy self-optimization phase, based on the abnormal diagnosis results and fault tracing information, a hierarchical federated learning architecture is constructed. A differentiated data security strategy is adopted through the privacy desensitization and grading module. The collaborative control strategy is optimized by iteratively optimizing the data of each subsystem, and a strategy self-optimization feedback mechanism is set up. During the collaborative scheduling phase, a global collaborative scheduling mechanism is constructed based on the optimized collaborative control strategy and model parameters. A dynamic scheduling priority module is set to realize the linkage operation, data interaction and operation and maintenance assistance between the auxiliary control system, the host system and the plant-level monitoring information system. During the integration and debugging phase, based on the system's coordinated operation data, the functional modules are modularly integrated and phased debugging is carried out. Phased verification indicators are set, full lifecycle operation verification is conducted, and an OTA remote upgrade and compatibility verification mechanism is established.

2. The integrated centralized monitoring and collaborative control method for auxiliary control systems of thermal power plants according to claim 1, characterized in that, The embedded edge acquisition unit in the acquisition and preprocessing stage adopts a wide-temperature, salt spray-resistant industrial-grade design, which can fully cover the auxiliary control subsystem of thermal power plants. It integrates a multi-protocol adaptive parsing module and an edge-side lightweight AI prediction module. The multi-protocol adaptive parsing module can automatically identify the communication protocols of different manufacturers' equipment and dynamically adapt to protocol version updates. The edge-side lightweight AI prediction module identifies minor parameter anomalies in advance based on a small-sample trained model. The preprocessing operations include data noise reduction and anomaly removal. The standardized packaging follows the IEC 61850 standard, unifies data naming specifications and formats, and generates a structured dataset.

3. The integrated centralized monitoring and collaborative control method for auxiliary control systems of thermal power plants according to claim 2, characterized in that, The dataset calibration mechanism is a dynamic calibration, which combines historical equipment operating data and nameplate parameters to calibrate the accuracy of the collected data in real time; the anti-interference transmission guarantee adopts dual technologies of signal isolation and redundant transmission; the communication interface is a 5G / industrial Ethernet dual-mode communication interface, which is adapted to the subsequent network upgrade needs of thermal power plants.

4. The integrated centralized monitoring and collaborative control method for auxiliary control systems of thermal power plants according to claim 3, characterized in that, The integrated centralized monitoring platform in the platform construction phase adopts a four-level hierarchical architecture consisting of a field equipment layer, a control execution layer, a monitoring and scheduling layer, and a management decision-making layer. The communication backbone network adopts a dual-ring fiber optic redundant ring network. The platform integrates a real-time database and a historical data archiving module, and interfaces with standardized datasets to achieve unified integration, storage, and display of data from all auxiliary control subsystems. The digital twin virtual-real linkage closed-loop mechanism can construct a virtual simulation model of the entire life cycle of the auxiliary control subsystem, support operating condition simulation and control strategy pre-playing, and use the pre-playing results to guide the adjustment of physical equipment control parameters.

5. The integrated centralized monitoring and collaborative control method for auxiliary control systems of thermal power plants according to claim 4, characterized in that, The platform's interfaces with the plant-level monitoring information system and management information system employ a standardized data interaction protocol to achieve data interoperability. The hierarchical access control module assigns corresponding data viewing permissions and control operation permissions based on the job responsibilities of operation, maintenance, scheduling, and management personnel, and strengthens end-to-end data security by combining data encryption transmission technology.

6. The integrated centralized monitoring and collaborative control method for auxiliary control systems of thermal power plants according to claim 5, characterized in that, In the collaborative control framework of the strategy modeling stage, the auxiliary control subsystem coupling factor dynamic allocation module constructs dynamic mathematical models of each auxiliary control subsystem based on standardized historical data, quantifies the coupling relationship between subsystems, and dynamically allocates the control weights of model predictive control and deep reinforcement learning under different coupling scenarios; the operating condition prediction model mines the variation patterns of historical load, coal quality, and environmental parameters based on the LSTM algorithm to predict the trend of operating condition changes. The deep reinforcement learning algorithm adopts a lightweight network design, and the reward and punishment function is designed in combination with multi-dimensional operation indicators such as energy consumption, environmental compliance rate, and equipment wear.

7. The integrated centralized monitoring and collaborative control method for auxiliary control systems of thermal power plants according to claim 6, characterized in that, The hierarchical anomaly diagnosis model in the diagnostic execution phase is constructed by fusing CNN and LSTM. Based on real-time data from the platform, it identifies auxiliary control equipment faults, parameter anomalies, and pipeline leaks, and classifies them into three levels: minor anomalies, general faults, and severe faults. The adaptive self-healing control strategy matches corresponding handling logic to different anomaly levels. Minor anomalies achieve self-healing by automatically adjusting control parameters. General faults trigger automatic switching of backup equipment and achieve dynamic load distribution of related subsystems. Severe faults immediately trigger alarms and push emergency response plans, synchronously linking with the operation and maintenance system to generate maintenance work orders, and locating the root cause of the fault through the fault tracing module.

8. The integrated centralized monitoring and collaborative control method for auxiliary control systems of thermal power plants according to claim 7, characterized in that, The hierarchical federated learning architecture in the strategy self-optimization phase is a three-tier architecture comprising edge nodes, regional nodes, and a central node. Edge nodes are used for local data training, regional nodes are used for parameter aggregation of similar subsystems, and the central node is used for global model optimization. Each level is configured with a differentiated encryption mechanism. The privacy desensitization and grading module implements differentiated data security strategies, employing differentiated desensitization strategies for different types of auxiliary control data. Sensitive data is protected by both encryption desensitization and access control, while ordinary data is processed with lightweight desensitization. The strategy self-optimization feedback mechanism feeds back self-healing results and fault tracing information into the federated learning model to continuously iterate and optimize the collaborative control strategy.

9. The integrated centralized monitoring and collaborative control method for auxiliary control systems of thermal power plants according to claim 8, characterized in that, The dynamic scheduling priority module in the collaborative scheduling phase dynamically allocates the scheduling priority of each auxiliary control subsystem according to the power plant's operating objectives. When the load of the main system is adjusted, the platform issues optimization instructions to each auxiliary control subsystem through the collaborative control framework. When an abnormality occurs in the auxiliary control system, it synchronously feeds back to the main system and adjusts the main system's operating parameters. At the same time, it connects to the plant-level monitoring information system to synchronously push auxiliary control system operation, energy consumption, and fault data, generate energy efficiency benchmarking reports and equipment health scores, and provide auxiliary decision-making suggestions for operation and maintenance optimization, equipment replacement, and energy consumption management.

10. The integrated centralized monitoring and collaborative control method for auxiliary control systems of thermal power plants according to claim 9, characterized in that, The integration and debugging phase adopts a modular integration approach to integrate the edge acquisition unit, centralized monitoring platform, collaborative control framework, self-healing control module, and strategy self-optimization module. The phased debugging sequentially carries out single subsystem debugging and optimization, inter-subsystem collaborative debugging, and linkage debugging with the host and management information system, and sets corresponding phased verification indicators. In the full life cycle verification phase, by collecting system operation data, the control strategy and diagnostic accuracy are continuously optimized using a federated learning model. OTA remote upgrades and function expansion are supported.

Images