A safety risk prediction system for gas turbine unit operation in a narrow space

Abstract

The application discloses a kind of safety risk prediction systems of gas unit narrow space operation, multi-source data acquisition module, dynamic risk prediction module, lightweight wearable early warning terminal, edge computing node and cloud linkage module;The multi-source data acquisition module is used to collect the multi-source heterogeneous data in the process of gas unit narrow space operation, and the safety risk prediction system of the gas unit narrow space operation adopts multi-source data global collection, and is combined with the improvement LSTM-GRU model and space topological feature correction, can be in advance to judge potential safety risk in narrow space;Edge-cloud collaborative architecture realizes local rapid early warning, avoids transmission delay;Explosion-proof lightweight wearable terminal graded sound and light early warning, response is timely;Cloud continuously optimizes model and outputs prevention and control suggestion, intelligent, high safety, adapts to gas unit narrow space operation safety control.

Description

Technical Field

[0001] This invention relates to the field of safety-related products for gas turbine units, specifically a safety risk prediction system for gas turbine unit operations in confined spaces. Background Technology

[0002] Working in confined spaces such as the furnace, flue, and pipe interlayer of gas turbine units is a crucial aspect of gas turbine unit operation and maintenance, and also a high-risk scenario for safety accidents. Confined spaces are characterized by limited space, poor ventilation, complex environment, and weak signal, making them prone to safety risks such as gas leaks, oxygen deficiency and asphyxiation, excessive levels of toxic and harmful gases, and equipment malfunctions. Once an accident occurs, rescue is difficult and the consequences are severe.

[0003] Currently, existing safety monitoring and risk prediction technologies for gas turbine unit operations in confined spaces have several shortcomings: First, traditional risk prediction relies heavily on single-sensor monitoring, collecting only limited environmental data such as gas concentration, temperature, and humidity, without integrating key data such as equipment operation, personnel operation, and spatial topology. This results in a single prediction dimension, failing to comprehensively reflect operational risks and leading to low prediction accuracy. Second, existing prediction models are mostly static early warning models, failing to consider the rapid spread and concentrated impact of risks in confined spaces. This makes dynamic, real-time risk prediction impossible, often only issuing warnings after a risk has occurred, resulting in significant delays and insufficient emergency response time for workers. Third, existing monitoring and early warning equipment is bulky and cumbersome, unsuitable for confined space operations. Large monitoring equipment struggles to enter confined spaces, while smaller equipment has limited functionality, failing to achieve multi-dimensional data collection and rapid early warning. Fourth, existing systems often suffer from a disconnect between "monitoring-early warning-prevention," only capable of risk monitoring and early warning, unable to generate targeted prevention and control recommendations based on risk conditions, thus hindering effective guidance for on-site personnel in risk prevention and control work. Summary of the Invention

[0004] The purpose of this invention is to provide a safety risk prediction system for gas turbine unit operations in confined spaces, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a safety risk prediction system for gas turbine unit operations in confined spaces, comprising: a multi-source data acquisition module, a dynamic risk prediction module, a lightweight wearable early warning terminal, an edge computing node, and a cloud-based linkage module; the multi-source data acquisition module is used to collect multi-source heterogeneous data during gas turbine unit operations in confined spaces, including operational environment data, equipment operation data, personnel operation data, and spatial topology data; the dynamic risk prediction module is connected to the multi-source data acquisition module and is used to preprocess and extract features from the multi-source heterogeneous data, and based on an improved LSTM-GRU hybrid model, combine spatial topology features to correct model weights, thereby realizing the safety risk prediction system for gas turbine unit operations in confined spaces. The system dynamically and in real-time predicts industrial safety risks and outputs risk classification results. The lightweight wearable early warning terminal, worn by workers, collects real-time status data, receives risk warning signals, and issues audible and visual alarms. Edge computing nodes, connected to the multi-source data acquisition module, dynamic risk prediction module, and lightweight wearable early warning terminal, are deployed near the entrance of confined spaces to process various collected data in real time, enabling rapid local early warning response and avoiding early warning delays caused by reliance on cloud transmission. The cloud-based linkage module, connected to the edge computing nodes, stores historical data, real-time risk data, and operational plans, iteratively optimizes the dynamic risk prediction model, and generates personalized risk prevention and control suggestions based on the risk prediction results.

[0006] As a preferred embodiment of the present invention, the multi-source data acquisition module includes an environmental sensor group, an equipment data acquisition unit, a personnel status acquisition unit, and a spatial topology acquisition unit. The environmental sensor group is deployed inside the confined space and around the work site to collect data on gas leak concentration, oxygen content, toxic and harmful gas concentration, space temperature and pressure, and ventilation volume. The equipment data acquisition unit is connected to the gas turbine control system to collect data on turbine vibration, pipeline pressure, valve status, and turbine operating load. The personnel status acquisition unit is integrated into a lightweight wearable early warning terminal to collect data on the location, standardization of actions, and wearing of protective equipment of the workers. The spatial topology acquisition unit is used to collect data on the geometric dimensions of the confined space, ventilation conditions, and distance between the work site and key equipment of the turbine.

[0007] As a preferred embodiment of the present invention, the working process of the dynamic risk prediction module includes:

[0008] S1. Preprocess the heterogeneous data from multiple sources collected by the multi-source data acquisition module, including data cleaning, normalization and outlier removal. Data augmentation is performed by adding noise and data mirroring to obtain a standardized dataset.

[0009] S2. Extract features from the standardized dataset, including environmental features, equipment features, personnel features, and spatial topology features, and construct a multi-dimensional feature vector.

[0010] S3. Input the multi-dimensional feature vector into the improved LSTM-GRU hybrid model. The improved LSTM-GRU hybrid model introduces a spatial topology feature weight correction factor on the basis of the traditional LSTM-GRU model. The weight ratio of each feature is dynamically adjusted according to the geometric size of the narrow space, ventilation conditions, and distance between the work point and the key equipment.

[0011] S4. The improved LSTM-GRU hybrid model after training outputs risk classification results, which include no risk, low risk, medium risk and high risk, corresponding to different warning levels.

[0012] As a preferred embodiment of the present invention, the training process of the improved LSTM-GRU hybrid model includes: collecting historical risk data and simulated risk data of gas turbine unit operations in confined spaces to construct a training dataset, wherein the training dataset contains multi-source heterogeneous data corresponding to different operating scenarios and different risk levels; dividing the training dataset into a training set and a test set, using the training set to train the improved LSTM-GRU hybrid model, adjusting the model parameters and spatial topology feature weight correction factors with the goal of minimizing prediction error; verifying the model performance through the test set, and if the prediction accuracy does not reach a preset threshold, continuing iterative optimization until the model prediction accuracy is ≥90%.

[0013] As a preferred embodiment of the present invention, the lightweight wearable early warning terminal adopts an explosion-proof design, is 60% smaller than traditional early warning devices, and integrates a miniature gas sensor, an inertial navigation module, an audible and visual alarm unit, and a wireless communication unit. The miniature gas sensor is used to collect real-time data on the gas concentration and oxygen content around the workers. The inertial navigation module is used to collect the workers' position and movement data. The audible and visual alarm unit emits audible and visual alarms of different intensities based on the received early warning signal. The wireless communication unit is used to realize bidirectional data transmission with edge computing nodes.

[0014] In a preferred embodiment of the present invention, the edge computing node includes a data processing unit, an early warning decision unit, and a communication unit. The data processing unit is used to process data collected by the multi-source data acquisition module and the lightweight wearable early warning terminal in real time, and to perform data fusion and feature extraction. The early warning decision unit is used to receive the risk classification results output by the dynamic risk prediction module, and, in conjunction with the locally processed data, to determine whether an early warning is triggered. If an early warning is triggered, an early warning signal is generated and sent to the lightweight wearable early warning terminal, with an early warning response time of ≤1 second. The communication unit is used to realize data interaction with the cloud linkage module, upload real-time data to the cloud, and receive model optimization parameters and prevention and control suggestions issued by the cloud.

[0015] In a preferred embodiment of the present invention, the cloud-based linkage module includes a data storage unit, a model optimization unit, and a prevention and control suggestion generation unit. The data storage unit is used to store multi-source heterogeneous historical data, real-time risk data, model training data, and work plan data. The model optimization unit is used to iteratively optimize the improved LSTM-GRU hybrid model in the dynamic risk prediction module based on historical and real-time data, and update the model parameters and spatial topology feature weight correction factors. The prevention and control suggestion generation unit is used to generate personalized risk prevention and control suggestions based on risk classification results, work site characteristics, and equipment operating status, including suggestions for upgrading protective equipment, ventilation optimization schemes, and work path adjustment suggestions.

[0016] As a preferred embodiment of the present invention, it also includes an interactive display module, which is connected to the edge computing node and the cloud linkage module respectively, and is used to display real-time data, risk classification results, early warning information and personalized prevention and control suggestions, supporting operators and managers to set parameters, query historical data and trace risks.

[0017] As a preferred embodiment of the present invention, the multi-source data acquisition module adopts a distributed deployment method. The environmental sensor group is evenly deployed in key locations according to the geometric dimensions of the narrow space and the distribution of work points to ensure the comprehensiveness and accuracy of data acquisition. The lightweight wearable early warning terminal supports multi-device linkage, and multiple terminals can achieve data sharing and early warning synchronization.

[0018] As a preferred embodiment of the present invention, the dynamic risk prediction module can predict potential risks 10-15 minutes in advance, including the risk of gas leakage and spread, the risk of oxygen deficiency and suffocation, the risk of equipment malfunction, and the risk of non-standard personnel operation. The risk classification result corresponds to the audible and visual alarm intensity of the lightweight wearable early warning terminal, with high risk corresponding to high-intensity audible and visual alarm and low risk corresponding to low-intensity audible and visual alarm.

[0019] Compared with the prior art, the beneficial effects of the present invention are:

[0020] This safety risk prediction system for gas turbine unit operations in confined spaces employs multi-source, full-domain data acquisition, combined with an improved LSTM-GRU model and spatial topology feature correction, enabling early prediction of potential safety risks in confined spaces. An edge-cloud collaborative architecture achieves rapid local early warning, avoiding transmission delays. Explosion-proof, lightweight wearable terminals provide tiered audible and visual warnings for timely response. The cloud continuously optimizes the model and outputs prevention and control suggestions, making it intelligent, highly safe, and suitable for the safety management of gas turbine unit operations in confined spaces. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of a safety risk prediction system for gas turbine unit operation in confined spaces according to the present invention;

[0022] Figure 2 This is a flowchart illustrating a safety risk prediction system for gas turbine unit operations in confined spaces according to the present invention.

[0023] Figure 3 This is a schematic diagram of the dynamic risk prediction module of a safety risk prediction system for gas turbine unit operations in confined spaces, according to the present invention.

[0024] In the diagram: 1. Multi-source data acquisition module; 2. Dynamic risk prediction module; 3. Lightweight wearable early warning terminal; 4. Edge computing node; 5. Cloud-based linkage module; 6. Interactive display module; 11. Environmental sensor group; 12. Equipment data acquisition unit; 13. Personnel status acquisition unit; 14. Spatial topology acquisition unit; 41. Data processing unit; 42. Early warning decision unit; 43. Communication unit; 51. Data storage unit; 52. Model optimization unit; 53. Prevention and control suggestion generation unit. Detailed Implementation

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0026] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," "front end," "rear end," "both ends," "one end," and "the other end," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0027] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installed," "equipped with," "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0028] Please see Figure 1-3This invention provides an embodiment of a safety risk prediction system for working in confined spaces within a gas turbine unit. The system includes a multi-source data acquisition module 1, a dynamic risk prediction module 2, a lightweight wearable early warning terminal 3, an edge computing node 4, a cloud-based linkage module 5, and an interactive display module 6 (the modules are connected wirelessly or via wired means). Specifically, the multi-source data acquisition module 1 includes an environmental sensor group 11, an equipment data acquisition unit 12, a personnel status acquisition unit 13, and a spatial topology acquisition unit 14. The environmental sensor group 11 is deployed in a distributed manner, with eight sensors evenly arranged inside the gas turbine unit furnace and around the work site. These sensors collect data on gas leakage concentration (detection range 0-100% LEL), oxygen content (detection range 0-25% VOL), concentration of toxic and harmful gases (such as carbon monoxide and hydrogen sulfide), and spatial temperature and pressure (temperature -2). The equipment data acquisition unit 12 is connected to the gas turbine control system via an interface to collect real-time data on unit vibration (detection range 0-10mm / s), pipeline pressure (0-2.0MPa), valve status (open / closed), and unit operating load (0-100%). The personnel status acquisition unit 13 is integrated into the lightweight wearable early warning terminal 3, which collects the position (accuracy ±0.5m) and action data of the workers through the inertial navigation module, and detects the wearing status of protective equipment (safety helmet, safety belt, gas mask) through the image recognition module. The spatial topology acquisition unit 14 collects the geometric dimensions of the furnace (length × width × height = 5m × 3m × 4m), ventilation conditions (ventilation volume 0-500m³ / h), and distance data between the work point and the gas pipeline through a laser rangefinder. Dynamic Risk Prediction Module 2: Preprocesses the data collected by Multi-Source Data Acquisition Module 1, using Z-score normalization to remove outliers and augmenting the data with noise and data mirroring. It extracts environmental, equipment, personnel, and spatial topology features using CNN to construct multi-dimensional feature vectors. These feature vectors are then input into an improved LSTM-GRU hybrid model, incorporating a spatial topology feature weight correction factor. The weights of each feature are dynamically adjusted based on the distance between the work site and the gas pipeline (20% increase in gas concentration feature weight when distance ≤ 1m; default weights when distance > 1m). Model training utilizes historical risk data and simulated data (covering 2000 sets of data from different scenarios). After training, the prediction accuracy reaches 92%, enabling the prediction of potential risks 12 minutes in advance and outputting four levels of risk classification: no risk, low risk, medium risk, and high risk.Lightweight wearable early warning terminal 3: Adopts explosion-proof design (explosion-proof rating Exd II CT6), with dimensions of 10cm×5cm×2cm and a weight of ≤100g, worn on the wrist of the operator; integrates a miniature gas sensor, oxygen content sensor, inertial navigation module, audible and visual alarm unit and Bluetooth communication unit 43; collects real-time data on gas concentration and oxygen content around the operator and uploads it to edge computing node 4; after receiving the early warning signal, it issues a high-frequency (1000Hz) high-brightness audible and visual alarm when there is a high risk, an intermediate-frequency (800Hz) intermediate-brightness audible and visual alarm when there is a medium risk, and a low-frequency (500Hz) low-brightness audible and visual alarm when there is a low risk. Edge computing node 4: Deployed near the furnace inlet, using an industrial-grade edge computing gateway (CPU is Intel Core i5, memory is 8GB); Data processing unit 41 uses a lightweight CNN algorithm to process various types of collected data in real time and filter key risk data; Early warning decision unit 42 receives the risk classification results from dynamic risk prediction module 2, and when medium risk or above is detected, it generates an early warning signal within 1 second and sends it to lightweight wearable early warning terminal 3; Communication unit 43 uses 5G+Bluetooth dual-mode communication to realize data interaction with cloud linkage module 5 and wearable terminal. Cloud-based linkage module 5: Utilizes a cloud server (Intel Xeon E5 CPU, 32GB RAM, 1TB storage); Data storage unit 51 stores historical data, real-time data, and operational plans; Model optimization unit 52 iteratively optimizes the improved LSTM-GRU hybrid model every 7 days based on historical and real-time data; Prevention and control suggestion generation unit 53 generates personalized prevention and control suggestions based on risk classification results—for example, in high-risk situations, it is recommended to wear portable gas masks, increase temporary ventilation equipment, and adjust work routes to avoid the vicinity of gas pipelines; in medium-risk situations, it is recommended to increase the frequency of environmental monitoring and standardize the operation of workers. Interactive display module 6: Deployed at the furnace inlet monitoring station, it uses a 15-inch industrial touchscreen to display real-time environmental data, equipment data, personnel location and actions, risk classification results, early warning information, and prevention and control suggestions, supporting parameter settings (such as adjusting early warning thresholds), historical data queries, and risk tracing.

[0029] The system operation process in this embodiment is as follows:

[0030] S1. Multi-source data acquisition module 1 collects environmental data, equipment data, personnel data and spatial topology data in real time during furnace operation and uploads them to edge computing node 4;

[0031] S2 and edge computing node 4 perform preliminary processing on the collected data and upload it to dynamic risk prediction module 2. At the same time, some key data are uploaded to cloud linkage module 5.

[0032] S3, Dynamic Risk Prediction Module 2 preprocesses the data and extracts features. It then uses an improved LSTM-GRU hybrid model to predict risks, outputs risk classification results, and sends them to the edge computing node 4.

[0033] S4, edge computing node 4 determines whether an early warning is triggered based on the risk classification results. If an early warning is triggered, it immediately sends an early warning signal to the lightweight wearable early warning terminal 3 and uploads the early warning information to the cloud.

[0034] S5, the lightweight wearable early warning terminal 3 receives the early warning signal and issues an audible and visual alarm of corresponding intensity to remind workers to pay attention to the risks;

[0035] S6 and cloud-based linkage module 5 generate personalized prevention and control suggestions based on risk classification results and real-time data, and send them to edge computing node 4 and interactive display module 6;

[0036] S7. Based on the audible and visual alarms and prevention and control recommendations, the operators shall take corresponding emergency measures. The managers shall monitor the operation in real time through the interactive display module 6 and guide the on-site operation.

[0037] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A safety risk prediction system for gas turbine unit operation in confined spaces, characterized in that: include: The system comprises a multi-source data acquisition module (1), a dynamic risk prediction module (2), a lightweight wearable early warning terminal (3), an edge computing node (4), and a cloud-based linkage module (5). The multi-source data acquisition module (1) is used to collect multi-source heterogeneous data during the operation of the gas turbine unit in a narrow space. The multi-source heterogeneous data includes operating environment data, equipment operation data, personnel operation data, and spatial topology data. The dynamic risk prediction module (2) is connected to the multi-source data acquisition module (1) and is used to preprocess and extract features from the multi-source heterogeneous data. Based on the improved LSTM-GRU hybrid model, the model weights are corrected by combining spatial topology features to realize dynamic real-time prediction of safety risks in the operation of the gas turbine unit in a narrow space and output risk classification results. The lightweight wearable early warning terminal (3) is worn by the operator to collect real-time status data of the operator and receive risk warning signals to issue audible and visual alarms. The edge computing node (4) is connected to the multi-source data acquisition module (1), the dynamic risk prediction module (2), and the lightweight wearable early warning terminal (3) respectively. It is deployed near the entrance of a narrow space to process various types of data collected in real time, realize local rapid early warning response, and avoid early warning delay caused by relying on cloud transmission. The cloud linkage module (5) is connected to the edge computing node (4) to store historical data, real-time risk data and operation plan, iteratively optimize the dynamic risk prediction model, and generate personalized risk prevention and control suggestions based on the risk prediction results.

2. The safety risk prediction system for gas turbine unit operation in confined spaces according to claim 1, characterized in that: The multi-source data acquisition module (1) includes an environmental sensor group (11), an equipment data acquisition unit (12), a personnel status acquisition unit (13), and a spatial topology acquisition unit (14). The environmental sensor group (11) is deployed inside the narrow space and around the work site to collect data on gas leakage concentration, oxygen content, toxic and harmful gas concentration, space temperature and pressure, and ventilation volume. The equipment data acquisition unit (12) is connected to the gas turbine control system to collect data on turbine vibration, pipeline pressure, valve status, and turbine operating load. The personnel status acquisition unit (13) is integrated into a lightweight wearable early warning terminal (3) to collect data on the location, standardization of actions, and wearing of protective equipment of the workers. The spatial topology acquisition unit (14) is used to collect data on the geometric dimensions of the narrow space, ventilation conditions, and distance between the work site and key equipment of the turbine.

3. The safety risk prediction system for gas turbine unit operation in confined spaces according to claim 1, characterized in that: The working process of the dynamic risk prediction module (2) includes: S1. Preprocess the heterogeneous data collected by the multi-source data acquisition module (1), including data cleaning, normalization and outlier removal. Data enhancement is performed by adding noise and data mirroring to obtain a standardized dataset. S2. Extract features from the standardized dataset, including environmental features, equipment features, personnel features, and spatial topology features, and construct a multi-dimensional feature vector. S3. Input the multi-dimensional feature vector into the improved LSTM-GRU hybrid model. The improved LSTM-GRU hybrid model introduces a spatial topology feature weight correction factor on the basis of the traditional LSTM-GRU model. The weight ratio of each feature is dynamically adjusted according to the geometric size of the narrow space, ventilation conditions, and distance between the work point and the key equipment. S4. The improved LSTM-GRU hybrid model after training outputs risk classification results, which include no risk, low risk, medium risk and high risk, corresponding to different warning levels.

4. The safety risk prediction system for gas turbine unit operation in confined spaces according to claim 1, characterized in that: The training process of the improved LSTM-GRU hybrid model includes: collecting historical risk data and simulated risk data of gas turbine unit operations in confined spaces to construct a training dataset, which contains multi-source heterogeneous data corresponding to different operating scenarios and risk levels; dividing the training dataset into a training set and a test set, using the training set to train the improved LSTM-GRU hybrid model, adjusting model parameters and spatial topology feature weight correction factors with the goal of minimizing prediction error; verifying model performance using the test set, and if the prediction accuracy does not reach a preset threshold, continuing iterative optimization until the model prediction accuracy is ≥90%.

5. A safety risk prediction system for gas turbine unit operation in confined spaces according to claim 1, characterized in that: The lightweight wearable early warning terminal (3) adopts an explosion-proof design and is 60% smaller than traditional early warning devices. It integrates a miniature gas sensor, an inertial navigation module, an audible and visual alarm unit, and a wireless communication unit (43). The miniature gas sensor is used to collect real-time data on the gas concentration and oxygen content around the workers. The inertial navigation module is used to collect data on the workers' location and actions. The audible and visual alarm unit issues audible and visual alarms of different intensities based on the received early warning signal. The wireless communication unit (43) is used to realize bidirectional data transmission with the edge computing node (4).

6. The safety risk prediction system for gas turbine unit operation in confined spaces according to claim 1, characterized in that: The edge computing node (4) includes a data processing unit (41), an early warning decision unit (42), and a communication unit (43); the data processing unit (41) is used to process the data collected by the multi-source data acquisition module (1) and the lightweight wearable early warning terminal (3) in real time, and to perform data fusion and feature extraction; The early warning decision unit (42) is used to receive the risk classification results output by the dynamic risk prediction module (2), and determine whether an early warning is triggered by combining the data processed locally. If an early warning is triggered, an early warning signal is generated and sent to the lightweight wearable early warning terminal (3). The early warning response time is ≤1 second. The communication unit (43) is used to realize data interaction with the cloud linkage module (5), upload real-time data to the cloud, and receive model optimization parameters and prevention and control suggestions issued by the cloud.

7. The safety risk prediction system for gas turbine unit operation in confined spaces according to claim 1, characterized in that: The cloud-based linkage module (5) includes a data storage unit (51), a model optimization unit (52), and a prevention and control suggestion generation unit (53). The data storage unit (51) is used to store multi-source heterogeneous historical data, real-time risk data, model training data, and operation plan data. The model optimization unit (52) is used to iteratively optimize the improved LSTM-GRU hybrid model in the dynamic risk prediction module (2) based on historical data and real-time data, and update the model parameters and spatial topology feature weight correction factors. The prevention and control suggestion generation unit (53) is used to generate personalized risk prevention and control suggestions based on the risk classification results, the characteristics of the work site and the equipment operating status, including suggestions for upgrading protective equipment, ventilation optimization schemes and work path adjustment suggestions.

8. A safety risk prediction system for gas turbine unit operation in confined spaces according to claim 1, characterized in that: It also includes an interactive display module (6), which is connected to the edge computing node (4) and the cloud linkage module (5) respectively. It is used to display real-time data, risk classification results, early warning information and personalized prevention and control suggestions, and supports operators and managers to set parameters, query historical data and trace risks.

9. A safety risk prediction system for gas turbine unit operation in confined spaces according to claim 1, characterized in that: The multi-source data acquisition module (1) adopts a distributed deployment method. The environmental sensor group (11) is evenly deployed in key positions according to the geometric size of the narrow space and the distribution of work points to ensure the comprehensiveness and accuracy of data acquisition. The lightweight wearable early warning terminal (3) supports multi-device linkage, and multiple terminals can realize data sharing and early warning synchronization.

10. A safety risk prediction system for gas turbine unit operation in confined spaces according to claim 1, characterized in that: The dynamic risk prediction module (2) can predict potential risks 10-15 minutes in advance, including the risk of gas leakage and diffusion, the risk of oxygen deficiency and suffocation, the risk of equipment malfunction and the risk of non-standard operation by personnel. The risk classification results correspond to the sound and light alarm intensity of the lightweight wearable early warning terminal (3). High risk corresponds to high-intensity sound and light alarm, and low risk corresponds to low-intensity sound and light alarm.

Images