Edge layer data acquisition and analysis method and system for gas turbine power plant management and control

Abstract

The invention discloses an edge layer data acquisition and analysis method and system for gas turbine power plant management and control, and relates to the technical field of intelligent power plant management and control. A gas turbine power plant management and control server analyzes key characteristic values of multi-source sensor data, and adaptively generates a lightweight model according to the key characteristic values of the multi-source sensor data; communicating with the edge acquisition nodes to deploy a lightweight model; the edge acquisition nodes monitor core equipment partitions of the gas turbine power plant based on the lightweight model, and the lightweight model processes monitoring data and then transmits a monitoring result to the gas turbine power plant management and control server. According to the marginal layer data collection and analysis method and system for gas turbine power plant management and control, the marginal collection nodes are deployed close to the core equipment partition, sensor data can be directly collected in a short distance, and the length of a data transmission path is reduced; meanwhile, data cleaning, feature extraction and lightweight model processing are all completed on the edge side, and centralized processing delay after the whole original data is uploaded to a gas turbine power plant management and control server is avoided.

Description

Technical Field

[0001] This invention relates to the field of smart power plant management and control technology, and in particular to an edge layer data acquisition and analysis method and system for gas turbine power plant management and control. Background Technology

[0002] Currently, the management and control of gas turbine power plants mainly relies on a combination of traditional industrial control systems and information systems. The core technologies include:

[0003] Data acquisition technology: Based on DCS (Distributed Control System) and SCADA (Supervisory and Data Acquisition System), key parameters (such as gas turbine speed, temperature, pressure, vibration, etc.) are collected, and sensors and PLCs are used to complete equipment status monitoring.

[0004] The system adopts a hierarchical structure of industrial Ethernet using fieldbus. The field layer connects devices through bus protocols such as Profibus and Modbus, while the control layer aggregates data to the central control room server through industrial Ethernet (such as EtherNet / IP).

[0005] The communication method is mainly wired (fiber optic, twisted pair), and in some scenarios, WiFi or 4G is used to enable mobile terminal access. Data transmission is mostly one-way aggregation and lacks real-time two-way interaction.

[0006] The data analysis model adopts centralized storage plus post-event analysis. After the data is uploaded to the plant-level data center, performance evaluation and fault tracing are carried out through offline modeling. Some power plants have introduced basic AI models to achieve simple early warning (such as over-temperature alarm).

[0007] The main drawback of existing technologies is insufficient real-time performance. Data needs to be forwarded through multiple layers to the central control room or cloud for analysis, and the latency is usually between hundreds of milliseconds and seconds, which makes it difficult to meet the real-time control requirements of rapid dynamic processes in gas turbines (such as combustion instability and surge).

[0008] Gas turbine power plants generate gigabytes of data per second (such as high-frequency vibration signals and infrared thermal imaging data). Centralized transmission leads to backbone network congestion and a surge in data center storage costs.

[0009] Equipment from different manufacturers (such as GE gas turbines and Siemens generators) uses proprietary protocols, resulting in inconsistent data formats and creating "information silos," making cross-system collaborative analysis difficult.

[0010] Analysis lag: Traditional analysis relies on manual triggering or timed tasks, and fault warnings are mostly "post-event responses", unable to proactively intervene in the early stages of faults (such as early wear of bearings).

[0011] Lack of edge capabilities: Field devices only have data acquisition functions and do not have local computing units deployed, making it impossible to complete real-time processing (such as noise filtering and feature extraction) near the data source, resulting in invalid data occupying transmission resources. Summary of the Invention

[0012] To address the shortcomings of the aforementioned medical device inventory management technologies, this invention provides a method and system for edge-layer data acquisition and analysis in gas turbine power plant management. The technical solution adopted is as follows:

[0013] The edge-layer data acquisition and analysis method for gas turbine power plant management includes the following steps:

[0014] Step 1: Divide the gas turbine power plant into zones according to the core equipment, and install edge acquisition nodes in each zone;

[0015] Step 2: The edge acquisition node collects multi-source sensor data in real time and performs edge-end data fusion;

[0016] Step 3: The edge acquisition nodes complete data cleaning and feature extraction, and only upload the key feature values ​​of the multi-source sensor data to the gas turbine power plant control server;

[0017] Step 4: The gas turbine power plant control server analyzes the key feature values ​​of multi-source sensor data, adaptively generates a lightweight model based on the key feature values ​​of multi-source sensor data, and communicates with the edge acquisition nodes to deploy the lightweight model.

[0018] Step 5: The edge acquisition nodes monitor the core equipment zones of the gas turbine power plant based on the lightweight model. After processing the monitoring data, the lightweight model transmits the monitoring results to the gas turbine power plant management and control server.

[0019] Step 6: The gas turbine power plant management server integrates and displays the global monitoring results based on the monitoring results.

[0020] Optionally, the edge acquisition node has a built-in protocol conversion engine that automatically identifies the communication protocol of the access device.

[0021] Optionally, the protocol conversion engine can automatically identify the communication protocol using the following methods:

[0022] The edge acquisition node first identifies the baud rate and signal speed, signal level and encoding method, transmission medium and interface type by recognizing signal characteristics, and then performs preliminary screening and matching based on a pre-set communication protocol database to obtain multiple preliminary matching communication protocols;

[0023] Then, by analyzing the structural features of the data frame, including frame boundaries and fixed identifiers, verification methods and length features, data field encoding and semantic features, the exact matching communication protocol is obtained by performing structural feature matching with multiple preliminary matching communication protocols based on the structural features.

[0024] Finally, protocol confirmation based on interaction logic is achieved through a communication handshake process.

[0025] Optionally, in step 4, the adaptive generation of the lightweight model includes the following steps:

[0026] Step 41, let the key feature values ​​uploaded by the edge acquisition nodes be set X = {x1, x2, x3, ..., x...} n}, where n is the feature dimension. The gas turbine power plant control server calculates the feature contribution rate of multiple key feature values ​​in X through principal component analysis, selects the top m principal components whose cumulative contribution rate is greater than the set contribution rate threshold, and compresses the feature dimension to m. The value of m and the type of key feature value are used as the core constraints for model structure screening.

[0027] Step 42: Based on the device type, call the basic model library. First, exclude models whose input layer dimension cannot be adjusted to m. Then, calculate the matching degree between the key feature value type and the input items of the remaining basic models. Select the basic model with the highest matching degree as the target model.

[0028] Step 43: Based on the resource parameters of the edge acquisition nodes, the target model is lightly compressed to obtain a lightweight model;

[0029] Step 44: The lightweight model is packaged and transmitted to the edge acquisition node via the MQTT protocol. After receiving the model, the edge acquisition node first verifies the file integrity, then executes the decompression command, and finally loads and runs the model, returning a deployment success signal.

[0030] Optionally, in step 42, the formula for calculating the matching degree between the key feature value type and the remaining base model inputs is:

[0031]

[0032] Where Acc is the model's accuracy on the historical feature set matching the m-dimensional dimension, FLPs is the model's computational cost, and d opt Let α be the optimal input dimension for the model, γ be the precision weight, and γ be the dimension fit weight. The relationship between m and d is strengthened using the exponential function exp. opt Proximity has an impact.

[0033] Optionally, in step 43, the resource constraint of the edge acquisition node is set as: maximum memory M max Maximum computing power C max Maximum real-time memory usage (M)used And the maximum real-time computing power occupied by C used Calculate the minimum free memory M free M free =M max -M used ; Calculate the minimum idle computing power C free C free =C max -C used ; with minimum free memory M free and minimum idle computing power C free Due to resource constraints.

[0034] Optionally, in step 43, model pruning, model quantization, and knowledge distillation are used to perform lightweight compression on the target model to obtain a lightweight model, making the volume of the lightweight model smaller than M. free The computing power requirement is less than C. free .

[0035] Optionally, after the edge acquisition node loads the lightweight model, it performs localized inference on the real-time acquired feature data to obtain monitoring results, and transmits the localized inference results to the gas turbine power plant control server. The gas turbine power plant control server performs cross-domain fusion on the monitoring results of the edge acquisition nodes in each zone, and displays the original monitoring results and cross-domain fusion results as a digital twin.

[0036] Optionally, the gas turbine power plant management and control server inputs the original monitoring results and cross-domain fusion results into the gas turbine power plant operation AI early warning model. The gas turbine power plant operation AI early warning model outputs the early warning type and early warning result. The gas turbine power plant management and control server controls the early warning unit to execute early warning actions based on the early warning type and early warning result.

[0037] An edge-layer data acquisition and analysis system for gas turbine power plant management and control is used to implement edge-layer data acquisition and analysis methods for gas turbine power plant management and control. The edge-layer data acquisition and analysis system includes multiple edge acquisition nodes, a gas turbine power plant management and control server, and an early warning unit. The multiple edge acquisition nodes are respectively installed in multiple partition physical centers of the core equipment of the gas turbine power plant and are respectively connected to the gas turbine power plant management and control server. The multiple edge acquisition nodes perform cross-domain fusion of monitoring data based on lightweight models to obtain monitoring results, and transmit the monitoring results to the gas turbine power plant management and control server. The gas turbine power plant management and control server performs cross-domain fusion of monitoring results, inputs the original monitoring results and cross-domain fusion results into the gas turbine power plant to run the AI ​​early warning model, and controls the early warning unit to execute early warning actions based on the early warning results.

[0038] In summary, the present invention has at least one of the following beneficial technical effects:

[0039] This invention provides an edge-layer data acquisition and analysis method and system for gas turbine power plant management and control. Edge acquisition nodes are deployed close to core equipment in partitions, enabling direct and close-range acquisition of sensor data and reducing data transmission path length. At the same time, data cleaning, feature extraction, and lightweight model processing are all completed at the edge, avoiding the centralized processing delay after the raw data is fully uploaded to the gas turbine power plant management and control server.

[0040] The gas turbine power plant management server adaptively generates a lightweight model based on key feature values ​​collected in actual operation. This means that the model can dynamically match the real-time operating status of the equipment, avoid the limitations of fixed models, and improve the model's adaptability to real-world scenarios.

[0041] By employing a layered architecture that combines edge-side localized processing with global server collaboration, synergistic advantages are achieved in terms of real-time performance, resource efficiency, monitoring accuracy, and system adaptability, which can significantly improve the intelligence level and operational reliability of gas turbine power plant management. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating the edge layer data acquisition and analysis method for the management and control of gas turbine power plants according to the present invention. Detailed Implementation

[0043] The present invention will be further described in detail below with reference to the accompanying drawings.

[0044] This invention discloses a method and system for edge layer data acquisition and analysis for the management and control of gas turbine power plants.

[0045] Reference Figure 1 Example 1, an edge layer data acquisition and analysis method for gas turbine power plant management and control, includes the following steps:

[0046] Step 1: Divide the gas turbine power plant into zones according to the core equipment, and install edge acquisition nodes in each zone;

[0047] Step 2: The edge acquisition node collects multi-source sensor data in real time and performs edge-end data fusion;

[0048] Step 3: The edge acquisition nodes complete data cleaning and feature extraction, and only upload the key feature values ​​of the multi-source sensor data to the gas turbine power plant control server;

[0049] Step 4: The gas turbine power plant control server analyzes the key feature values ​​of multi-source sensor data, adaptively generates a lightweight model based on the key feature values ​​of multi-source sensor data, and communicates with the edge acquisition nodes to deploy the lightweight model.

[0050] Step 5: The edge acquisition nodes monitor the core equipment zones of the gas turbine power plant based on the lightweight model. After processing the monitoring data, the lightweight model transmits the monitoring results to the gas turbine power plant management and control server.

[0051] Step 6: The gas turbine power plant management server integrates and displays the global monitoring results based on the monitoring results.

[0052] By adopting the above technical solution, edge acquisition nodes are deployed close to core equipment in partitions, enabling direct and close-range acquisition of sensor data and reducing data transmission path length. At the same time, data cleaning, feature extraction, and lightweight model processing are all completed at the edge, avoiding the centralized processing delay after the raw data is fully uploaded to the gas turbine power plant's control server.

[0053] The lightweight model runs locally at the edge, enabling real-time analysis of device status changes and rapid output of results. Compared to the mode of only collecting data at the edge and processing it fully on the server, the response time can be significantly shortened, providing more timely basis for device anomaly warning and real-time control.

[0054] Edge acquisition nodes first complete data cleaning and feature extraction, and only upload key feature values ​​instead of raw data, which can significantly reduce the amount of data transmission, reduce the occupation of network bandwidth, and avoid transmission congestion.

[0055] The gas turbine power plant management server does not need to process massive amounts of raw data, but only analyzes key feature values, reducing the server's computational load; at the same time, the lightweight model is adapted to the computing power limitations of edge acquisition nodes, avoiding processing bottlenecks caused by insufficient computing power at the edge.

[0056] The gas turbine power plant management server adaptively generates a lightweight model based on the key feature values ​​collected in actual operation. This means that the model can dynamically match the real-time operating status of the equipment (such as feature differences under different loads and operating conditions), avoiding the limitations of fixed models and improving the model's adaptability to real-world scenarios.

[0057] The management server integrates the monitoring results of each edge node globally, and can associate the status of each zone from the overall perspective of the power plant, avoiding the one-sidedness of local monitoring, providing managers with a complete and global picture of equipment operation, and assisting in the formulation of systematic management and control strategies.

[0058] The method utilizes a layered architecture that combines edge-side localized processing with global server collaboration, achieving synergistic advantages in real-time performance, resource efficiency, monitoring accuracy, and system adaptability. This significantly improves the intelligence level and operational reliability of gas turbine power plant management.

[0059] Example 2: The edge acquisition node has a built-in protocol conversion engine that automatically identifies the communication protocol of the access device.

[0060] Example 3: The method for the protocol conversion engine to automatically identify the communication protocol is as follows:

[0061] The edge acquisition node first identifies the baud rate and signal speed, signal level and encoding method, transmission medium and interface type by recognizing signal characteristics, and then performs preliminary screening and matching based on a pre-set communication protocol database to obtain multiple preliminary matching communication protocols;

[0062] Then, by analyzing the structural features of the data frame, including frame boundaries and fixed identifiers, verification methods and length features, data field encoding and semantic features, the exact matching communication protocol is obtained by performing structural feature matching with multiple preliminary matching communication protocols based on the structural features.

[0063] Finally, protocol confirmation based on interaction logic is achieved through a communication handshake process.

[0064] By adopting the above technical solution, the edge acquisition node determines the device's communication protocol type by identifying signal characteristics, data format, and communication handshake process information, and completes protocol conversion based on the protocol type. The specific method can be divided into the following three core identification steps and subsequent protocol conversion process, which are explained in detail below:

[0065] Signal Feature Recognition: Preliminary Classification Based on Physical Layer Parameters

[0066] Signal feature identification focuses on the physical layer attributes of communication signals. By collecting the electrical characteristics and transmission parameters of the signals, preliminary screening of protocols is performed. The core identification parameters and methods are as follows:

[0067] Baud rate and signal rate: Edge acquisition nodes have built-in signal sampling modules to monitor the signal transmission rate in real time (common baud rates are 9600, 19200, 38400, etc., while Ethernet protocols are 10Mbps / 100Mbps / 1Gbps). For example, the Modbus RTU protocol often uses a 9600 baud rate, while the Profinet protocol is based on 100Mbps Ethernet by default. Rate matching can initially eliminate incompatible protocol types (e.g., excluding signals with a baud rate of 38400 from the Profinet protocol).

[0068] Signal Level and Encoding Method: Monitor the signal level standard (e.g., RS485 is a differential signal, with a high-low level difference typically greater than 200mV; TTL is a single-ended signal, with a high level of 35V and a low level of 0.8V) and encoding method (e.g., Manchester encoding, NRZ encoding). For example, the Ethernet protocol uses Manchester encoding (signal transitions represent logic 1 / 0), while Modbus RTU mostly uses NRZ encoding (fixed levels represent logic states). Encoding characteristics can further narrow down the protocol range.

[0069] Transmission medium and interface type: Determine the characteristics of the protocol transmission medium by combining the access interface (such as RS232 / RS485 interface, Ethernet RJ45 interface, wireless LoRa module). For example, RS485 interface usually corresponds to bus protocols such as Modbus RTU and Profibus DP; RJ45 interface corresponds to Ethernet protocols such as TCP / IP, Modbus TCP, and Profinet. The association between interface type and signal characteristics can be used for quick classification.

[0070] Data format recognition: precise matching based on frame structure features;

[0071] Data format recognition accurately distinguishes protocols by parsing the structural features of data frames (such as frame header, frame trailer, checksum, and data field format). The core recognition logic is as follows:

[0072] Frame boundaries and fixed identifiers:

[0073] Monitor whether the data frame contains a fixed start / end character (e.g., Modbus RTU frames have no fixed frame header, but frames are separated by idle time; Modbus TCP's MBAP header is fixed at 7 bytes, with the first byte being 0x00; OPCUA protocol frames use a specific GUID as an identifier). By detecting the start / end character of the frame, the complete range of the frame can be located and its features extracted.

[0074] Verification method and length characteristics:

[0075] The parsing mechanism identifies the checksum field type (e.g., Modbus RTU uses CRC16, Modbus ASCII uses LRC, and TCP / IP uses checksum) and data field length rules (e.g., Profibus DP data frames have variable lengths but are limited by the Maximum Transmission Unit (MTU); BACnet protocol frames contain an explicit length field). For example, if a 2-byte CRC checksum is detected at the end of the frame and the data field length is ≤256 bytes, it can be preliminarily identified as the Modbus RTU protocol.

[0076] Data domain encoding and semantic features:

[0077] Analyze the encoding format (binary, ASCII, JSON, etc.) and semantic identifiers (such as function codes and register addresses) of the data fields. For example, the Modbus protocol data field contains a function code (0x03 indicates reading a holding register, 0x06 indicates writing to a single register), and the register address is a 16-bit integer; the Profinet protocol data field contains IO data and alarm information, and carries a unique device identifier (DeviceID). These semantic features can be used to accurately match the protocol type.

[0078] The communication handshake process identification identifies the protocol type by monitoring the message exchange sequence, handshake rules, and state machine during device connection establishment. The core identification dimensions are as follows:

[0079] Connection establishment process:

[0080] Record the initial interaction steps of device communication:

[0081] If a "request and response" interaction exists, it may be a master-slave bus protocol such as ModbusRTU / ASCII;

[0082] If a "three-way handshake" (SYN, SYN+ACK, ACK) and a "four-way handshake" process are present, then it is the TCP protocol;

[0083] If there are device discovery, parameterization, and cyclic communication steps (such as Profinet's DCP protocol for device discovery and IRT real-time channel establishment), then it is an Industrial Ethernet Real-Time Protocol.

[0084] Timeout and retransmission mechanism: Monitor message retransmission rules (e.g., dynamically adjust TCP retransmission timeout, fix Modbus RTU retransmission interval at 3.5 character intervals, and LoRa protocol retransmission based on ACK confirmation). For example, if no acknowledgment is received after message transmission, retransmission is performed at intervals of 100ms ± 20ms, and the number of retransmissions is ≤ 3, which can be considered as conforming to the characteristics of industrial wireless protocols (e.g., WirelessHART).

[0085] Session layer identifier interaction: For higher-order protocols (such as OPCUA and MQTT), the handshake process includes session establishment information (e.g., the CONNECT message in MQTT contains the client ID and protocol version; the Hello message in OPCUA contains the endpoint URL and security policy). By parsing these session layer identifiers, the protocol type can be directly matched.

[0086] Once the protocol type is confirmed through the above three steps, the protocol conversion engine performs the following operations:

[0087] Protocol mapping: Calls the built-in protocol mapping library (which contains mapping relationships between mainstream industrial protocols and standard protocols inside edge nodes).

[0088] Unified format: The timing, units, and data types (such as floating-point and integer) of different protocols are uniformly converted into an internal format that the edge nodes can process (such as timestamps accurate to milliseconds, temperature units uniformly set to ℃, and pressure units uniformly set to MPa), laying the foundation for subsequent edge data fusion (step 2).

[0089] Dynamic adaptation: If a new protocol (not a type in the built-in library) is accessed, the protocol conversion engine records its signal characteristics, data format and handshake rules, generates a temporary mapping table and feeds it back to the management server. The server updates the protocol library and then synchronizes it to the edge nodes to achieve self-iteration of protocol recognition.

[0090] Edge acquisition nodes can automatically identify the communication protocols of access devices (such as Modbus, Profinet, OPCUA, etc.) and complete protocol conversion to ensure that multi-source sensor data can be uniformly acquired, fused and processed, providing standardized data input for subsequent data cleaning, feature extraction and lightweight modeling.

[0091] Example 4, step 4, the adaptive generation of the lightweight model includes the following steps:

[0092] Step 41, let the key feature values ​​uploaded by the edge acquisition nodes be set X = {x1, x2, x3, ..., x...} n}, where n is the feature dimension. The gas turbine power plant control server calculates the feature contribution rate of multiple key feature values ​​in X through principal component analysis, selects the top m principal components whose cumulative contribution rate is greater than the set contribution rate threshold, and compresses the feature dimension to m. The value of m and the type of key feature value are used as the core constraints for model structure screening.

[0093] Step 42: Based on the device type, call the basic model library. First, exclude models whose input layer dimension cannot be adjusted to m. Then, calculate the matching degree between the key feature value type and the input items of the remaining basic models. Select the basic model with the highest matching degree as the target model.

[0094] Step 43: Based on the resource parameters of the edge acquisition nodes, the target model is lightly compressed to obtain a lightweight model;

[0095] Step 44: The lightweight model is packaged and transmitted to the edge acquisition node via the MQTT protocol. After receiving the model, the edge acquisition node first verifies the file integrity, then executes the decompression command, and finally loads and runs the model, returning a deployment success signal.

[0096] In Example 5, step 42, the formula for calculating the matching degree between the key feature value type and the remaining basic model input items is:

[0097]

[0098] Where Acc is the model's accuracy on the historical feature set matching the m-dimensional dimension, FLPs is the model's computational cost, and d opt Let α be the optimal input dimension for the model, γ be the precision weight, and γ be the dimension fit weight. The relationship between m and d is strengthened using the exponential function exp. opt Proximity has an impact.

[0099] In Example 6, step 43, the resource constraint of the edge acquisition node is set as: maximum memory M max Maximum computing power C max Maximum real-time memory usage (M) used And the maximum real-time computing power occupied by C used Calculate the minimum free memory M free M free =M max -M used ; Calculate the minimum idle computing power C free C free =C max -C used ; with minimum free memory M free and minimum idle computing power C free Due to resource constraints.

[0100] In Example 7, step 43 involves using model pruning, model quantization, and knowledge distillation to lightweight compress the target model, resulting in a lightweight model with a volume smaller than M. free The computing power requirement is less than C. free .

[0101] By adopting the above technical solution, feature dimension compression (from n to m) based on principal component analysis (PCA) has a clear quantitative objective: to retain only the first m principal components whose cumulative contribution rate exceeds the threshold.

[0102] Principal component analysis extracts the most representative features (principal components with high contribution rates) from the data, compressing dimensionality while avoiding the loss of key information, and ensuring that the feature set input to the model can effectively reflect the equipment status (such as the correlation between core indicators such as temperature and vibration).

[0103] After the dimension is compressed from n to m, the computational cost of subsequent model training and inference decreases linearly or exponentially with the reduction of dimension. This reduces the burden on the server to generate the model and lays the foundation for lightweight operation of edge nodes.

[0104] Step 42 filters target models through dimensional matching and matching metric calculations, and combines this with a specific matching degree formula (considering accuracy, computational cost, optimal input dimension, and weights) to achieve accurate model-scene adaptation.

[0105] First, exclude models whose input layer dimension cannot be adjusted to m to avoid performance waste caused by forcing small dimensions into large models or accuracy loss caused by "forced dimension expansion of small models", thus reducing adaptation errors from the source.

[0106] Quantitative balance between accuracy and efficiency: In the matching degree formula, α·Acc ensures the basic monitoring accuracy of the model. Suppressing computationally intensive models (to avoid insufficient computing power at edge nodes), γ·exp(-|md opt |) Enhance the optimal input dimension d of the model through an exponential function. opt The closeness to the actual dimension m. This quantization mechanism ensures that the selected target model can meet the current feature dimension requirements while maintaining high accuracy under edge computing power constraints, avoiding extreme cases of sacrificing accuracy for lightweighting or ignoring computing power for accuracy.

[0107] Step 43 achieves tailored model and edge node hardware through resource-constrained computation (minimum free memory, minimum free computing power) and targeted compression techniques (pruning, quantization, knowledge distillation):

[0108] Model pruning removes redundant neurons (such as removing connections that have minimal impact on the output) to reduce model parameters; model quantization converts 32-bit floating-point parameters into 8-bit integers to reduce memory usage and computation (typically compressing the size by 4 times); knowledge distillation transfers knowledge from the teacher model (complex server model) to the student model (lightweight edge model) while compressing the data and preserving core reasoning capabilities.

[0109] The combination of these three elements enables the lightweight model to reduce its size and computing power requirements while keeping the loss of key monitoring performance to a very small extent.

[0110] Step 44 ensures the reliability of the entire model deployment process through MQTT protocol transmission, integrity verification, and feedback mechanisms: low-power and high-efficiency transmission, suitable for long-distance or weak network environments between edge nodes and servers, avoiding transmission delays from affecting the timeliness of model updates.

[0111] After receiving the model, the edge node first verifies the integrity of the file to avoid model running errors caused by data corruption during transmission; after decompression and loading, it returns a deployment success signal, enabling the server to perceive the deployment status in real time and trigger a retry mechanism for failed nodes to ensure model consistency across all partition edge nodes.

[0112] In Example 8, after the edge acquisition node loads the lightweight model, it performs localized inference on the real-time acquired feature data to obtain the monitoring results, and transmits the localized inference results to the gas turbine power plant control server. The gas turbine power plant control server performs cross-domain fusion on the monitoring results of the edge acquisition nodes in each zone, and displays the original monitoring results and the cross-domain fusion results in a digital twin.

[0113] By adopting the above technical solutions, core equipment in gas turbine power plants, such as gas turbines, waste heat boilers, and steam turbines, exhibit strong coupling relationships. For example, gas turbine exhaust temperature affects waste heat boiler efficiency, and boiler pressure affects steam turbine operation. Individual zone edge nodes can only monitor the status of local equipment, making it difficult to identify cross-zone related faults. Cross-domain fusion, by integrating monitoring results from various zones (such as correlation features like temperature, pressure, and vibration), can construct a global equipment status correlation model, identifying systemic risks that cannot be perceived by a single zone.

[0114] Improving the accuracy of monitoring results: Cross-domain fusion can reduce local misjudgments through multi-source result verification. For example, if a node at the edge of a zone falsely reports an abnormal temperature due to temporary sensor drift, the server can determine that the abnormality is an isolated error rather than a real fault by fusing temperature, flow, and other related data from adjacent zones, thus reducing the false alarm rate. Conversely, if monitoring results from multiple zones point to the same potential problem (such as abnormal vibrations detected in multiple zones near the gas turbine), cross-domain fusion can strengthen the confidence of the fault and avoid missed reports.

[0115] Cross-domain integration results can reflect the overall operating status of equipment throughout the plant (such as load distribution and energy efficiency levels in each zone), providing data support for servers to formulate global optimization strategies (such as dynamic load allocation and overall maintenance planning), and avoiding the problem of local optima but global inefficiency caused by formulating strategies based on data from a single zone.

[0116] In Example 9, the gas turbine power plant management and control server inputs the original monitoring results and cross-domain fusion results into the gas turbine power plant operation AI early warning model. The gas turbine power plant operation AI early warning model outputs the early warning type and early warning result. Based on the early warning type and early warning result, the gas turbine power plant management and control server controls the early warning unit to execute early warning actions.

[0117] By adopting the above technical solutions and combining AI early warning with cross-domain fusion results, the limitations of monitoring single devices or zones can be overcome, and multi-zone collaborative anomalies can be identified (such as the correlation between gas supply system and turbine unit failures), achieving global risk perception of the overall operating status of the power plant. Simultaneously, the early warning results, combined with digital twin displays, can intuitively present the location, impact range, and development trend of anomalies, providing managers with panoramic decision support and improving the overall safety management level of the power plant.

[0118] Example 10: An edge-layer data acquisition and analysis system for gas turbine power plant management and control. This system implements an edge-layer data acquisition and analysis method for gas turbine power plant management and control. The system includes multiple edge acquisition nodes, a gas turbine power plant management and control server, and an early warning unit. The multiple edge acquisition nodes are installed in multiple partitioned physical centers of the core equipment of the gas turbine power plant and are communicatively connected to the gas turbine power plant management and control server. Each edge acquisition node performs inference on the monitoring data at the edge based on a lightweight model to obtain monitoring results, and then transmits these results to the gas turbine power plant management and control server. The gas turbine power plant management and control server performs cross-domain fusion of the monitoring results and inputs the original monitoring results and the cross-domain fusion results into the gas turbine power plant's AI early warning model. Based on the early warning results, the server controls the early warning unit to execute early warning actions.

[0119] The following specific embodiments illustrate the implementation principle of the present invention:

[0120] A gas turbine power plant includes a gas turbine (GT25000 type), a waste heat boiler (dual-pressure horizontal type), a steam turbine (single-shaft combined cycle), a generator, and auxiliary systems (fuel supply, cooling system, etc.). The core equipment is distributed in 5 functional zones (gas turbine zone, waste heat boiler zone, steam turbine zone, generator zone, and auxiliary system zone). Each zone is equipped with multiple types of sensors (temperature, vibration, pressure, flow, current, etc.). The sensor communication protocols cover Modbus RTU, Profinet, OPCUA, WirelessHART, etc. It is necessary to realize real-time monitoring of equipment operating status, anomaly early warning, and global control.

[0121] I. Partitioning and Edge Node Deployment:

[0122] 1. Partitioning: The core equipment is divided into 5 partitions (numbered Z1 and Z5) according to its functions. Each partition has one edge acquisition node (model ECU800, computing power 8 TOPS, memory 8GB, supports multi-protocol access) deployed at its physical center, for a total of 5 edge nodes (E1 and E5), corresponding to Z1 and Z5 respectively.

[0123] 2. Node Deployment: Edge nodes are connected to sensors / smart meters within the zone via industrial switches, using distributed power supply (DC24V), and are designed to be dustproof and vibration resistant (suitable for power plant workshop environments). They also establish communication with the gas turbine power plant management server (deployed in the central control room, with a computing power of 200 TOPS and 128GB of memory) via a 5G industrial gateway.

[0124] II. Protocol Conversion and Multi-Source Data Acquisition:

[0125] Edge node E1 (gas turbine area) needs to access the following equipment data:

[0126] Gas turbine shaft vibration sensors (2 units, Modbus RTU communication protocol, RS485 interface);

[0127] Combustion chamber temperature sensors (4 units, Profinet protocol, RJ45 interface);

[0128] Intelligent fuel flow meter (1 unit, OPCUA protocol, Ethernet interface);

[0129] Cooling system pressure sensors (2 units, WirelessHART protocol, wireless transmission).

[0130] Protocol conversion process:

[0131] 1. Signal Feature Recognition:

[0132] The E1's built-in protocol conversion engine detects the following through the signal sampling module: Modbus RTU sensor baud rate 9600, RS485 differential signal (level difference 250mV), NRZ encoding; Profinet device signal rate 100Mbps, Manchester encoding, RJ45 interface; based on the pre-built protocol database (containing 12 types of industrial protocol features), it initially matches 4 candidate protocols (Modbus RTU, Profinet, OPCUA, WirelessHART).

[0133] 2. Data frame structure matching:

[0134] Parsing Modbus RTU frames: There is no fixed frame header (separated by 3.5 characters of idle time), the frame tail contains a 2-byte CRC check, and the data field contains the function code (0x03) and register address (0x00010x0008), which completely matches the preset Modbus RTU structure characteristics;

[0135] Parsing the Profinet frame reveals that it contains a DCP device discovery message (DeviceID = 0x1A3B), and the data field is divided into IO data (16 bytes) and alarm information (8 bytes), matching the Profinet structure characteristics.

[0136] 3. Communication handshake confirmation:

[0137] When establishing a connection with the OPCUA instrument, the interaction logic of "Hello message (endpoint URL = opc.tcp: / / 192.168.1.10:4840) + security policy negotiation" was detected, confirming that the protocol is OPCUA;

[0138] When interacting with WirelessHART sensors, an "ACK confirmation retransmission (timeout 100ms±10ms, retransmission 3 times)" mechanism was detected, which matches the WirelessHART protocol.

[0139] 4. Protocol Conversion Results: The engine converts the above four types of protocol data into the internal standard format of the edge node (timestamp accurate to milliseconds, temperature unit ℃, pressure unit MPa, flow unit kg / s), laying the foundation for data fusion.

[0140] III. Data Acquisition and Feature Extraction:

[0141] 1. Multi-source data acquisition:

[0142] Edge nodes E1 and E5 collect data in real time at a frequency of 10Hz, for example:

[0143] Gas turbine section (E1): Shaft vibration (X / Y direction, 0.50 mm / s), combustion chamber outlet temperature (650-850℃), natural gas flow rate (2040 kg / s), cooling water pressure (0.8-1.2 MPa);

[0144] Waste heat boiler area (E2): Economizer outlet water temperature (280-320℃), steam pressure (4.5-5.5MPa), flue gas oxygen content (35%).

[0145] 2. Edge data fusion:

[0146] E1 performs spatiotemporal alignment on vibration, temperature, and flow data at the same timestamp (based on GPS synchronized clock, error ≤1ms), and removes isolated values ​​caused by sensor drift (e.g., if a temperature sensor instantly jumps to 1000℃, it is judged as abnormal and replaced with a moving average).

[0147] 3. Feature extraction and uploading:

[0148] Feature extraction is performed on the raw data (approximately 720,000 data points per hour), retaining only the key feature values:

[0149] Time-domain characteristics: peak vibration (daily maximum / minimum), mean temperature (5-minute moving average), and variance of flow fluctuation;

[0150] Frequency domain characteristics: 1X / 2X vibration frequencies of the gas turbine rotor (extracted via FFT conversion);

[0151] Correlation characteristics: Pearson correlation coefficient between combustion chamber temperature and fuel flow rate.

[0152] Ultimately, only 120 key feature values ​​are uploaded per hour (a 99.98% reduction in data volume), which are then sent to the management server via the MQTT protocol.

[0153] IV. Lightweight Model Generation and Deployment:

[0154] Taking the gas turbine area (E1) as an example, the server generates a lightweight "equipment anomaly monitoring model":

[0155] 1. Feature Dimension Compression:

[0156] The server receives 15 key features (vibration, temperature, flow rate, etc.) uploaded by E1 and calculates the feature contribution rate through principal component analysis (PCA): the cumulative contribution rate of the first 5 principal components (vibration peak value, temperature mean, 1X frequency, flow rate variance, temperature-flow rate correlation coefficient) reaches 92% (the threshold is set to 85%), so the dimension is compressed to m=5.

[0157] 2. Target model selection:

[0158] The server's basic model library (including Random Forest, LSTM, and CNN) is called. CNNs with input layer dimensions that cannot be adjusted to 5 (fixed input dimension 16) are excluded. The matching degree of the remaining models (Random Forest and LSTM) is calculated:

[0159] Random Forest: 91% historical accuracy, 2.3 TOPS computation, optimal input dimension 5 (consistent with m=5);

[0160] LSTM: 89% historical accuracy, 3.8 TOPS computation, optimal input dimension 8;

[0161] Substituting the matching degree formula (precision weight α = 0.6, dimension adaptation weight β = 0.4), the matching degree of Random Forest (0.87) is higher than that of LSTM (0.72), so Random Forest is selected as the target model.

[0162] 3. Lightweight model compression:

[0163] E1 resource constraints: Maximum memory 4GB, maximum computing power 5TOPS; Minimum free memory = 4GB, currently occupied 1.2GB = 2.8GB, minimum free computing power = 5TOPS, currently loaded 1.5TOPS = 3.5TOPS.

[0164] Lightweighting of Random Forest (original size 256MB, computational cost 2.3 TOPS):

[0165] Pruning: Remove 15% of redundant decision trees (impact on accuracy <2%);

[0166] Quantization: Converts 32-bit floating-point parameters to 8-bit integers (compressing size by 4 times);

[0167] Knowledge distillation: Using the server complex model (XGBoost) as the "teacher", optimize the output layer weights of the "student model".

[0168] The final lightweight model has a size of 32MB (<2.8GB) and a computational cost of 1.8TOPS (<3.5TOPS), satisfying the E1 resource constraint.

[0169] 4. Deployment process:

[0170] The server sends the model package (with CRC check) to E1 via the MQTT protocol; E1 checks the file integrity (CRC match), decompresses it, loads the model, and returns a "deployment successful" signal, all within 12 seconds.

[0171] V. Edge Monitoring and Cross-Domain Integration:

[0172] 1. Edge localization monitoring:

[0173] E1 uses a lightweight model to infer real-time feature data (updated every 100ms): when "vibration peak value > 45mm / s and 1X frequency fluctuation > 5Hz" is detected, it is determined as "rotor imbalance warning" and the result (including timestamp and 96% confidence level) is uploaded to the server.

[0174] 2. Cross-domain server integration:

[0175] The server receives the "rotor imbalance warning" from E1 (gas turbine area) and simultaneously retrieves the "steam pressure fluctuation" data from E2 (waste heat boiler area) and the "bearing temperature rise" data from E3 (steam turbine area). Through correlation analysis, it is found that: gas turbine rotor imbalance → waste heat boiler steam flow pulsation → uneven steam turbine bearing load, the three form a causal chain, and are judged as "systematic mechanical vibration abnormality" (excluding false alarms from a single sensor).

[0176] VI. AI Early Warning and Execution:

[0177] 1. Early warning model output:

[0178] The server inputs the cross-domain fusion results into the "Gas Turbine Power Plant Operation AI Early Warning Model" and outputs an early warning type of "Level 2 Mechanical Fault". The warning result is: the fault location is the No. 3 bearing of the gas turbine, and the affected area involves the waste heat boiler superheater. It is recommended to shut down the plant for maintenance within 1 hour.

[0179] 2. Early warning execution:

[0180] Server control and early warning unit (audio-visual alarm, central control room large screen, management personnel mobile APP):

[0181] The factory's audible and visual alarm is activated (red flashing + buzzer).

[0182] The large-screen digital twin interface highlights the fault area and related devices;

[0183] Send an SMS and app push notification to the Operations Director (including fault details and handling suggestions).

[0184] Edge-side inference response time is reduced from 2.3 seconds in traditional centralized server processing to 80ms, and anomaly warnings are issued 15 minutes earlier;

[0185] Data transfer volume decreased by 92% (from 1.8TB per day to 144GB), and server computing load decreased by 65%.

[0186] Through cross-domain integration, the false alarm rate was reduced from 8% to 1.2%, successfully avoiding two unit trips caused by gas turbine rotor imbalance;

[0187] Adaptability: The lightweight model is dynamically updated according to operating conditions (such as automatically adjusting the temperature threshold during high loads in summer), and the model adaptation accuracy reaches 94%.

[0188] Through the above implementation, the gas turbine power plant has achieved intelligent management and control through edge localization processing and global server collaboration, which has greatly improved operational reliability and significantly reduced annual downtime losses.

[0189] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for edge-layer data acquisition and analysis for the management and control of gas turbine power plants, characterized in that, Includes the following steps: Step 1: Divide the gas turbine power plant into zones according to the core equipment, and install edge acquisition nodes in each zone; Step 2: The edge acquisition node collects multi-source sensor data in real time and performs edge-end data fusion; Step 3: The edge acquisition nodes complete data cleaning and feature extraction, and only upload the key feature values ​​of the multi-source sensor data to the gas turbine power plant control server; Step 4: The gas turbine power plant control server analyzes the key feature values ​​of multi-source sensor data, adaptively generates a lightweight model based on the key feature values ​​of multi-source sensor data, and communicates with the edge acquisition nodes to deploy the lightweight model. Step 5: The edge acquisition nodes monitor the core equipment zones of the gas turbine power plant based on the lightweight model. After processing the monitoring data, the lightweight model transmits the monitoring results to the gas turbine power plant management and control server. Step 6: The gas turbine power plant management server integrates and displays the global monitoring results based on the monitoring results.

2. The edge layer data acquisition and analysis method for gas turbine power plant management and control according to claim 1, characterized in that, The edge acquisition node has a built-in protocol conversion engine that automatically identifies the communication protocol of the access device.

3. The edge layer data acquisition and analysis method for gas turbine power plant management and control according to claim 1, characterized in that, The protocol conversion engine automatically identifies the communication protocol using the following method: The edge acquisition node first identifies the baud rate and signal speed, signal level and encoding method, transmission medium and interface type by recognizing signal characteristics, and then performs preliminary screening and matching based on a pre-set communication protocol database to obtain multiple preliminary matching communication protocols; Then, by analyzing the structural features of the data frame, including frame boundaries and fixed identifiers, verification methods and length features, data field encoding and semantic features, the exact matching communication protocol is obtained by performing structural feature matching with multiple preliminary matching communication protocols based on the structural features. Finally, protocol confirmation based on interaction logic is achieved through a communication handshake process.

4. The edge layer data acquisition and analysis method for gas turbine power plant management and control according to claim 1, characterized in that, In step 4, the adaptive generation of the lightweight model includes the following steps: Step 41, let the key feature values ​​uploaded by the edge acquisition nodes be set as X = {x1, x2, x3, ..., x...} n }, where n is the feature dimension. The gas turbine power plant control server calculates the feature contribution rate of multiple key feature values ​​in X through principal component analysis, selects the top m principal components whose cumulative contribution rate is greater than the set contribution rate threshold, and compresses the feature dimension to m. The value of m and the type of key feature value are used as the core constraints for model structure screening. Step 42: Based on the device type, call the basic model library. First, exclude models whose input layer dimension cannot be adjusted to m. Then, calculate the matching degree between the key feature value type and the input items of the remaining basic models. Select the basic model with the highest matching degree as the target model. Step 43: Based on the resource parameters of the edge acquisition nodes, the target model is lightly compressed to obtain a lightweight model; Step 44: The lightweight model is packaged and transmitted to the edge acquisition node via the MQTT protocol. After receiving the model, the edge acquisition node first verifies the file integrity, then executes the decompression command, and finally loads and runs the model, returning a deployment success signal.

5. The edge layer data acquisition and analysis method for gas turbine power plant management and control according to claim 4, characterized in that, In step 42, the formula for calculating the matching degree between the key feature value type and the remaining base model inputs is: Where Acc is the model's accuracy on the historical feature set matching the m-dimensional dimension, FLPs is the model's computational cost, and d opt Let α be the optimal input dimension for the model, γ be the precision weight, and γ be the dimension fit weight. The relationship between m and d is strengthened using the exponential function exp. opt Proximity has an impact.

6. The edge layer data acquisition and analysis method for gas turbine power plant management and control according to claim 5, characterized in that, In step 43, the resource constraint of the edge acquisition node is set as: maximum memory M max Maximum computing power C max Maximum real-time memory usage (M) used And the maximum real-time computing power occupied by C used Calculate the minimum free memory M free M free =M max -M used ; Calculate the minimum idle computing power C free C free =C max -C used ; with minimum free memory M free and minimum idle computing power C free Due to resource constraints.

7. The edge layer data acquisition and analysis method for gas turbine power plant management and control according to claim 6, characterized in that, In step 43, model pruning, model quantization, and knowledge distillation are used to lightweight compress the target model to obtain a lightweight model, making the size of the lightweight model smaller than M. free The computing power requirement is less than C. free .

8. The edge layer data acquisition and analysis method for gas turbine power plant management and control according to claim 7, characterized in that, After the edge acquisition node loads the lightweight model, it performs localized inference on the real-time acquired feature data to obtain monitoring results, and transmits the localized inference results to the gas turbine power plant control server. The gas turbine power plant control server performs cross-domain fusion on the monitoring results of the edge acquisition nodes in each zone, and displays the original monitoring results and cross-domain fusion results as a digital twin.

9. The edge layer data acquisition and analysis method for gas turbine power plant management and control according to claim 8, characterized in that, The gas turbine power plant management and control server inputs the original monitoring results and cross-domain fusion results into the gas turbine power plant operation AI early warning model. The gas turbine power plant operation AI early warning model outputs the early warning type and early warning result. Based on the early warning type and early warning result, the gas turbine power plant management and control server controls the early warning unit to execute early warning actions.

10. An edge-layer data acquisition and analysis system for the management and control of gas turbine power plants, characterized in that: To implement the edge layer data acquisition and analysis method for gas turbine power plant management and control as described in claim 9, the edge layer data acquisition and analysis system includes multiple edge acquisition nodes, a gas turbine power plant management and control server, and an early warning unit. The multiple edge acquisition nodes are respectively installed in multiple partition physical centers of the core equipment of the gas turbine power plant and are respectively connected to the gas turbine power plant management and control server. The multiple edge acquisition nodes respectively perform inference on the monitoring data at the edge based on a lightweight model to obtain monitoring results, and transmit the monitoring results to the gas turbine power plant management and control server. The gas turbine power plant management and control server performs cross-domain fusion of the monitoring results, inputs the original monitoring results and cross-domain fusion results into the gas turbine power plant to run the AI ​​early warning model, and controls the early warning unit to execute early warning actions based on the early warning results.

Images