A fault early warning and self-recovery control system for a distributed control system of a gas turbine

Abstract

The application relates to the technical field of gas turbine power generation control, in particular to a gas turbine distributed control system fault early warning and self-healing regulation and control system, which comprises a hardware architecture layer and a software algorithm layer; the hardware architecture layer comprises a data acquisition unit, an edge computing unit, a DCS core control unit, a self-healing execution unit and a man-machine interaction unit which are connected through PROFIBUS-DP / EtherCAT double-bus redundancy; the software algorithm layer comprises a data preprocessing module, an LSTM fault early warning model module, a fault grade division module, a self-healing regulation and control strategy library module and a parameter self-optimization module which are arranged in the edge computing unit and the DCS core control unit. The application realizes early warning of a gradual change type fault by 30-60 minutes through deep mining of implicit correlations of multi-source time series data by using LSTM, and a closed-loop self-healing mechanism of a fault probability-grade-strategy is constructed.

Description

Technical Field

[0001] This invention relates to the field of gas turbine power generation control, and more particularly to a fault early warning and self-healing control system for a gas turbine distributed control system. Background Technology

[0002] Distributed control systems (DCS) for gas turbines commonly employ a threshold-based judgment combined with rule-based reasoning for fault diagnosis. This approach has significant limitations: First, it lacks sensitivity to gradual faults such as blade wear and bearing aging, triggering alarms only when operating parameters exceed set thresholds, thus lacking early warning capabilities for impending faults. Second, once a fault is identified, the system typically relies on manual analysis and the development of control strategies, leading to response delays and potentially exacerbating the fault.

[0003] Furthermore, existing methods fail to fully exploit the deep correlations between multi-source time-series data such as temperature, pressure, and vibration, making diagnostic accuracy susceptible to random fluctuations in single parameters. Simultaneously, the control strategy module and fault diagnosis module of a power generation control system are typically designed independently, resulting in a fragmented architecture that prevents closed-loop intelligent linkage from fault warning to execution control, thus limiting the system's ability to autonomously handle faults. Summary of the Invention

[0004] This invention specifically relates to a fault early warning and self-healing control system for a gas turbine distributed control system. This invention utilizes LSTM deep mining to uncover implicit correlations in multi-source time-series data, achieving early warning of gradual-type faults 30-60 minutes in advance, and constructing a closed-loop self-healing mechanism based on "fault probability-level-strategy". To achieve the above objectives, the specific technical solution of the fault early warning and self-healing control system for a gas turbine distributed control system of this invention is as follows: A fault early warning and self-healing control system for a gas turbine distributed control system is characterized by comprising a hardware architecture layer and a software algorithm layer. The hardware architecture layer includes a data acquisition unit, an edge computing unit, a DCS core control unit, a self-healing execution unit, and a human-machine interaction unit, all redundantly connected via PROFIBUS-DP / EtherCAT dual buses. The data acquisition unit serves as the starting point, transmitting the acquired real-time operating parameters to the edge computing unit via a PROFIBUS-DP / EtherCAT dual-bus redundant network. After performing on-site analysis of the data, the edge computing unit uploads the generated fault level signal to the DCS core control unit through the same bus network. The DCS core control unit makes decisions based on the received signals, generates control commands, and sends the commands to the self-healing execution unit for execution via the control bus; The human-machine interaction unit is bidirectionally connected to the DCS core control unit and is used to monitor the system status and send manual intervention commands to the DCS core control unit. The software algorithm layer includes a data preprocessing module, an LSTM fault early warning model module, a fault level classification module, a self-healing control strategy library module, and a parameter self-optimization module deployed in the edge computing unit and the DCS core control unit. The data preprocessing module is connected to the LSTM fault early warning model module, and the LSTM fault early warning model module is connected to the fault level classification module. The fault level classification module outputs a fault level signal as a trigger condition, and the fault level signal is sent to the self-healing control strategy library module in the DCS core control unit. The self-healing control strategy library module maps and outputs specific control strategies based on the fault level signal; The parameter self-optimization module optimizes and updates the control parameters in the strategy library. Compared with the prior art, the present invention has the following beneficial effects: 1) Early warning capability: By utilizing LSTM neural networks to deeply mine the implicit correlations in multi-source time series data, early warning of gradual-type faults can be achieved 30-60 minutes in advance with an accuracy rate of no less than 95%; 2) Self-healing closed-loop: Construct a precise mapping mechanism of failure probability-level-strategy to achieve closed-loop processing from early warning to automatic execution, with a response time of ≤5 seconds, reducing the risk of manual intervention; 3) Strong compatibility: The hardware adopts a PROFIBUS-DP / EtherCAT dual-bus redundant design, which is compatible with existing gas turbine DCS systems and does not require large-scale hardware replacement. 4) Self-optimization capability: By driving policy iteration through reinforcement learning and combining it with data feedback to update the model, the system has the ability to adapt and improve performance over a long period of time. Furthermore, the data acquisition unit includes distributed smart sensors and a bus acquisition module connected to the smart sensor signals; the operating parameters are 24 types, and the sampling frequency is 100Hz.

[0005] The beneficial effect of adopting the previous step is that, through high-frequency, multi-dimensional data acquisition, it covers the key operating states of the unit, providing rich and real-time input features for the LSTM fault early warning model; the intelligent sensors include temperature sensors, pressure sensors, vibration sensors, and speed sensors; the data acquisition unit collects 24 types of operating parameters of key components such as gas turbine compressor, combustion chamber, turbine, and generator, with a sampling frequency of 100Hz and a data transmission delay of ≤10ms.

[0006] Furthermore, the edge computing unit adopts an embedded computing module with an FPGA+ARM architecture and has a pre-trained LSTM fault warning model built in.

[0007] Furthermore, the data preprocessing module is used to perform noise reduction, normalization, and time-series slicing on the operating parameters collected by the data acquisition unit, and to generate the input feature matrix of the LSTM fault early warning model module. The LSTM fault warning model module performs real-time inference based on the input feature matrix and outputs the fault probability value and fault type. The fault level classification module classifies faults into different levels based on the fault probability value; The self-healing control strategy library module is used to store control strategies corresponding to different fault levels; The parameter self-optimization module iteratively optimizes the control parameters based on a reinforcement learning algorithm.

[0008] Furthermore, in the LSTM fault warning model module, the input layer dimension of the pre-trained LSTM fault warning model is 24, the hidden layer has 3 layers and each layer contains 128 neurons; when the fault probability value output by real-time inference is ≥0.7, a warning is triggered, and when it is ≥0.9, a fault is determined to have occurred.

[0009] The benefits of adopting the previous step are that by deploying a dedicated computing architecture at the edge, the latency of data transmission and cloud processing is reduced, meeting the real-time requirements; the deep network structure that matches the data dimensions enhances the feature extraction capability; and the clear threshold setting ensures the accuracy of early warning and fault determination.

[0010] Furthermore, the fault level classification module divides faults into four levels: Level I warning, the judgment condition is 0.7 ≤ fault probability < 0.8; Level II minor fault, the judgment condition is 0.8 ≤ fault probability < 0.9; Level III moderate fault, the judgment condition is 0.9 ≤ fault probability < 0.95; Level IV severe fault is defined as a fault probability ≥ 0.95.

[0011] The beneficial effect of the previous step is that by dividing the fault probability into fine intervals, it is possible to achieve gradient quantitative identification of the degree of fault development, which provides a clear decision-making basis for subsequent targeted hierarchical control.

[0012] Furthermore, the self-healing control strategy library module stores control strategies in three categories: parameter adjustment, load switching, and standby startup, which are mapped to fault levels: For Level I warning, the strategy is to trigger the parameter self-optimization module to fine-tune the fuel supply and air intake. For a Level II minor fault, the strategy is to switch to redundant sensors and adjust the unit load to 80% of the rated value. For a Level III moderate fault, the strategy is to activate the bypass system and disconnect part of the load on the faulty component. For a Level IV severe fault, the strategy is to trigger an emergency shutdown procedure and start the standby unit.

[0013] The beneficial effect of adopting the previous step is that it matches differentiated physical control methods for faults of different severity, realizing a step-by-step response from parameter fine-tuning to system reconfiguration, taking into account both operating efficiency and unit safety.

[0014] Furthermore, the DCS core control unit is used to receive the fault level signal output by the fault level classification module, and retrieve the corresponding control strategy from the self-healing control strategy library module to generate control instructions.

[0015] The beneficial effect of adopting the previous step is that, through data exchange with the upper-level monitoring system, plant-wide information integration and collaborative control are achieved, thereby improving the overall level of intelligent operation and management.

[0016] Furthermore, the self-healing execution unit is connected to and controlled by the DCS core control unit, and includes a valve actuator, a frequency converter, a bypass switching device, and a standby unit start-up module, for performing flow regulation, speed control, loop switching, and standby equipment commissioning operations, respectively. These operations are issued by the DCS core control unit and correspond to the control strategy.

[0017] The advantage of adopting the previous step is that by configuring a complete array of actuators, it ensures that various control strategies (such as flow regulation, load switching, emergency shutdown, etc.) can be accurately and reliably implemented.

[0018] Furthermore, the human-machine interaction unit includes a host computer monitoring interface for displaying equipment operating parameters, fault warning information, and the execution status of control strategies.

[0019] The advantage of adopting the previous step is that it provides users with an intuitive interface for status monitoring and decision support, ensuring automated closed-loop operation while retaining the necessary capabilities for manual supervision and emergency intervention. Attached Figure Description

[0020] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the following description, in conjunction with preferred embodiments and appendices, provides for further clarification. Figure 1 This invention will be described in detail. This embodiment is only for explaining the invention and is not intended to limit the scope of protection of the invention.

[0022] Example 1 This embodiment provides a fault early warning and self-healing control system for a gas turbine distributed control system, characterized by comprising a hardware architecture layer and a software algorithm layer. The hardware architecture layer includes a data acquisition unit, an edge computing unit, a DCS core control unit, a self-healing execution unit, and a human-machine interface unit, all redundantly connected via a PROFIBUS-DP / EtherCAT dual-bus network. The data acquisition unit, acting as a starting point, transmits real-time operating parameters to the edge computing unit via the PROFIBUS-DP / EtherCAT dual-bus redundant network. The edge computing unit performs on-site analysis of the data and uploads the generated fault level signal to the DCS core control unit via the same bus network. The DCS core control unit makes decisions based on the received signals, generates control commands, and sends the commands to the self-healing execution unit for execution via the control bus. The human-machine interface unit is bidirectionally connected to the DCS core control unit and is used to monitor the system status and send manual intervention commands to the DCS core control unit. The software algorithm layer includes a data preprocessing module, an LSTM fault early warning model module, a fault level classification module, a self-healing control strategy library module, and a parameter self-optimization module deployed in the edge computing unit and the DCS core control unit. The data preprocessing module is connected to the LSTM fault early warning model module, which is connected to the fault level classification module. The fault level classification module outputs a fault level signal as a trigger condition, which is sent to the self-healing control strategy library module in the DCS core control unit. The self-healing control strategy library module maps and outputs specific control strategies based on the fault level signal. The parameter self-optimization module optimizes and updates the control parameters in the strategy library. The data acquisition unit includes distributed intelligent sensors and a bus acquisition module connected to the intelligent sensor signals. There are 24 types of operating parameters, with a sampling frequency of 100Hz. The edge computing unit adopts an embedded computing module with an FPGA+ARM architecture and has a pre-trained LSTM fault early warning model built in. The hardware architecture layer adopts a layered distributed design, including a data acquisition unit, an edge computing unit, a DCS core control unit, a self-healing execution unit, and a human-machine interaction unit. Each unit is redundantly connected through PROFIBUS-DP / EtherCAT dual buses.

[0023] The data acquisition unit consists of intelligent sensors and a bus acquisition module. It collects 24 types of operating parameters from key components such as the gas turbine compressor, combustion chamber, turbine, and generator, with a sampling frequency of 100Hz and a data transmission delay of ≤10ms. The intelligent sensors include temperature sensors, pressure sensors, vibration sensors, and speed sensors. The data acquisition unit, as the starting point, transmits the collected real-time operating parameters to the edge computing unit via a PROFIBUS-DP / EtherCAT dual-bus redundant network. The edge computing unit analyzes the data locally and uploads the generated fault level signal to the DCS core control unit via the same bus network. The DCS core control unit makes decisions based on the received signals, generates control commands, and sends the commands to the self-healing execution unit for execution via the control bus. The human-machine interface unit is bidirectionally connected to the DCS core control unit, used to monitor the system status and send manual intervention commands to the DCS core control unit. The edge computing unit deploys an embedded computing module (FPGA+ARM architecture) with a built-in pre-trained LSTM fault early warning model. It is responsible for receiving and collecting data in real time, performing feature analysis and fault early warning, thus avoiding the latency problem of cloud computing. The DCS core control unit, as the core of the system, receives early warning signals and fault levels from the edge computing unit, retrieves the preset self-healing control strategy library, and generates control commands. It also realizes data interaction with the power plant's SCADA system. The self-healing execution unit includes a valve actuator, frequency converter, bypass switching device, and standby unit start-up module, which executes the control commands issued by the DCS core control unit. The software algorithm layer includes a data preprocessing module, an LSTM fault warning model module, a fault level classification module, a self-healing control strategy library module, and a parameter self-optimization module deployed in the edge computing unit and the DCS core control unit. The data preprocessing module performs noise reduction, normalization, and time-series slicing on the operating parameters collected by the data acquisition unit, and generates the input feature matrix for the LSTM fault warning model module. The LSTM fault warning model module performs real-time inference based on the input feature matrix, outputting fault probability values ​​and fault types. The fault level classification module classifies faults into different levels based on the fault probability values. The self-healing control strategy library module stores control strategies corresponding to different fault levels. The parameter self-optimization module iteratively optimizes the control parameters based on a reinforcement learning algorithm.

[0024] The data preprocessing module is connected to the LSTM fault early warning model module, which is connected to the fault level classification module. The fault level classification module outputs a fault level signal as a trigger condition, which is then sent to the self-healing control strategy library module in the DCS core control unit. The self-healing control strategy library module maps and outputs a specific control strategy based on the fault level signal. The parameter self-optimization module optimizes and updates the control parameters in the strategy library.

[0025] The data preprocessing module performs noise reduction (using wavelet transform), normalization (Min-Max standardization), and time-series slicing (sliding window length set to 120s) on the collected multi-source time-series data, removes outlier values, and generates the input feature matrix of the LSTM fault warning model. In the LSTM fault warning model module, the pre-trained LSTM fault warning model has an input layer dimension of 24 and three hidden layers, each containing 128 neurons. A warning is triggered when the real-time inference output fault probability value is ≥0.7, and a fault is determined to have occurred when it is ≥0.9. The fault level classification module divides faults into four levels: Level I warning (0.7 ≤ fault probability < 0.8); Level II minor fault (0.8 ≤ fault probability < 0.9); Level III moderate fault (0.9 ≤ fault probability < 0.95); and Level IV severe fault (fault probability ≥ 0.95).

[0026] LSTM fault prediction model module: Model training: The training set is historical fault data (including normal operation data, gradual fault data, and sudden fault data). The input layer dimension is 24 (corresponding to 24 types of operating parameters), the hidden layer is set to 3 layers, each with 128 neurons, and the output layer is the fault probability value (0~1) and fault type (such as blade wear, bearing failure, gas leakage, and 10 other common faults). Real-time inference: The edge computing unit inputs the pre-processed real-time data into the pre-trained model and outputs the fault probability value. When the probability value is ≥0.7, an early warning is triggered, and when it is ≥0.9, a fault is determined to have occurred.

[0027] Fault level classification module: Based on the fault probability value and the scope of impact, it is divided into 4 levels.

[0028]

[0029] The self-healing control strategy library module stores control strategies in three categories: parameter adjustment, load switching, and standby startup, which are mapped to fault levels: for Level I warning, the strategy is to trigger the parameter self-optimization module to fine-tune the fuel supply and air intake; for Level II minor fault, the strategy is to switch to redundant sensors and adjust the unit load to 80% of the rated value; for Level III moderate fault, the strategy is to start the bypass system and disconnect part of the load of the faulty component; for Level IV severe fault, the strategy is to trigger the emergency shutdown procedure and start the standby unit.

[0030] The self-healing control strategy library module establishes a mapping relationship between fault levels and control strategies, which include three types: parameter adjustment, load switching, and standby startup.

[0031] Level I warning: Triggers the parameter self-optimization module to fine-tune the fuel supply and air intake to suppress the parameter deterioration trend; Level II minor fault: Switch to redundant sensors to collect data, adjust the unit load to 80% of the rated value, and reduce the speed of fault development; Level III Medium Fault: Activate the bypass system, disconnect part of the load on the faulty component, and push the fault information to the operation and maintenance terminal; Level IV severe fault: Triggers emergency shutdown procedure, starts standby units, and ensures power grid continuity; Parameter self-optimization module: Based on reinforcement learning algorithm, with the goal of "reducing failure probability + optimizing unit efficiency", it iteratively optimizes the control parameters and feeds the optimized parameters back to the LSTM model training set to realize the self-updating of the model.

[0032] The DCS core control unit receives the fault level signal output by the fault level classification module and retrieves the corresponding control strategy from the self-healing control strategy library module to generate control commands. The self-healing execution unit includes a valve actuator, a frequency converter, a bypass switching device, and a standby unit startup module. The human-machine interface includes a host computer monitoring interface for displaying equipment operating parameters, fault warning information, and control strategy execution status; the human-machine interface includes a host computer monitoring interface that displays equipment operating parameters, fault warning information, and control strategy execution status, and supports manual intervention permission settings.

[0033] Example 2 Taking a 100MW gas turbine combined cycle power plant as an example, the system of this invention is deployed as follows: 1. Install 24 sets of intelligent sensors at key locations such as the gas turbine compressor outlet, combustion chamber, and turbine inlet, and connect them to the edge computing unit via EtherCAT bus; 2. The edge computing unit adopts an FPGA+ARM embedded architecture and a pre-trained LSTM fault warning model (the training set contains 100,000 sets of historical running data, of which fault data accounts for 30%). 3. The DCS core control unit is equipped with a self-healing control strategy library and interfaces with the power plant's existing SCADA system; 4. The self-healing actuator is connected to the fuel regulating valve, bypass valve, and standby generator set controller, and the priority of the control commands is set (the emergency shutdown command has the highest priority).

[0034] The operation flow of this invention: Data acquisition: The sensors collect 24 types of parameters in real time, including temperature (T1~T8), pressure (P1~P6), vibration (V1~V4), and rotational speed (N1~N2). After noise reduction by wavelet transform, the data is transmitted to the edge computing unit. Fault warning: The edge computing unit inputs preprocessed data into the LSTM fault warning model and outputs a fault probability value. When the probability of blade wear failure reaches 0.75, a Level I warning is triggered. Self-healing regulation: The DCS core control unit retrieves the Level I strategy, and the parameter self-optimization module fine-tunes the fuel supply by -5% and the air intake by +3%, suppressing the trend of blade wear and deterioration. Model update: After the adjusted data is preprocessed, it is added to the LSTM fault early warning model training set to retrain the model and improve the accuracy of subsequent early warnings.

[0035] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, the above-described features have similar functions to (but are not limited to) those disclosed in this application.

Claims

1. A fault early warning and self-healing control system for a gas turbine distributed control system, characterized in that, Includes the hardware architecture layer and the software algorithm layer; The hardware architecture layer includes a data acquisition unit, an edge computing unit, a DCS core control unit, a self-healing execution unit, and a human-machine interaction unit, all redundantly connected via PROFIBUS-DP / EtherCAT dual buses. The data acquisition unit serves as the starting point, transmitting the acquired real-time operating parameters to the edge computing unit via a PROFIBUS-DP / EtherCAT dual-bus redundant network. After performing on-site analysis of the data, the edge computing unit uploads the generated fault level signal to the DCS core control unit through the same bus network. The DCS core control unit makes decisions based on the received signals, generates control commands, and sends the commands to the self-healing execution unit for execution via the control bus; The human-machine interaction unit is bidirectionally connected to the DCS core control unit and is used to monitor the system status and send manual intervention commands to the DCS core control unit. The software algorithm layer includes a data preprocessing module, an LSTM fault early warning model module, a fault level classification module, a self-healing control strategy library module, and a parameter self-optimization module deployed in the edge computing unit and the DCS core control unit. The data preprocessing module is connected to the LSTM fault early warning model module, and the LSTM fault early warning model module is connected to the fault level classification module. The fault level classification module outputs a fault level signal as a trigger condition, and the fault level signal is sent to the self-healing control strategy library module in the DCS core control unit. The self-healing control strategy library module maps and outputs specific control strategies based on the fault level signal; The parameter self-optimization module optimizes and updates the control parameters in the strategy library.

2. The fault early warning and self-healing control system for a gas turbine distributed control system according to claim 1, characterized in that, The data acquisition unit includes distributed intelligent sensors and a bus acquisition module connected to the signals of the intelligent sensors; there are 24 types of operating parameters, and the sampling frequency is 100Hz.

3. The fault early warning and self-healing control system for a gas turbine distributed control system according to claim 1, characterized in that, The edge computing unit adopts an embedded computing module with an FPGA+ARM architecture and has a pre-trained LSTM fault warning model built in.

4. The fault early warning and self-healing control system for a gas turbine distributed control system according to claim 1, characterized in that, The data preprocessing module is used to perform noise reduction, normalization and time-series slicing on the operating parameters collected by the data acquisition unit, and to generate the input feature matrix of the LSTM fault early warning model module. The LSTM fault warning model module performs real-time inference based on the input feature matrix and outputs the fault probability value and fault type. The fault level classification module classifies faults into different levels based on the fault probability value; The self-healing control strategy library module is used to store control strategies corresponding to different fault levels; The parameter self-optimization module iteratively optimizes the control parameters based on a reinforcement learning algorithm.

5. The gas turbine distributed control system fault early warning and self-healing regulation system according to claim 4, characterized in that, In the LSTM fault warning model module, the pre-trained LSTM fault warning model has an input layer dimension of 24 and a hidden layer of 3 layers, each containing 128 neurons; when the fault probability value output by real-time inference is ≥0.7, a warning is triggered, and when it is ≥0.9, a fault is determined to have occurred.

6. The fault early warning and self-healing control system for a gas turbine distributed control system according to claim 4, characterized in that, The fault level classification module classifies faults into four levels: Level I warning, the judgment condition is 0.7 ≤ fault probability < 0.8; Level II minor fault, the judgment condition is 0.8 ≤ fault probability < 0.9; Level III moderate fault, the judgment condition is 0.9 ≤ fault probability < 0.95; Level IV severe fault is defined as a fault probability ≥ 0.

95.

7. The fault early warning and self-healing control system for a gas turbine distributed control system according to claim 6, characterized in that, The self-healing control strategy library module stores control strategies in three categories: parameter adjustment, load switching, and standby startup, and maps them to fault levels: For Level I warning, the strategy is to trigger the parameter self-optimization module to fine-tune the fuel supply and air intake. For a Level II minor fault, the strategy is to switch to redundant sensors and adjust the unit load to 80% of the rated value. For a Level III moderate fault, the strategy is to activate the bypass system and disconnect part of the load on the faulty component. For a Level IV severe fault, the strategy is to trigger an emergency shutdown procedure and start the standby unit.

8. The fault early warning and self-healing control system for a gas turbine distributed control system according to claim 1, characterized in that, The DCS core control unit is used to receive the fault level signal output by the fault level classification module and retrieve the corresponding control strategy from the self-healing control strategy library module to generate control instructions.

9. The fault early warning and self-healing control system for a gas turbine distributed control system according to claim 1, characterized in that, The self-healing execution unit is connected to and controlled by the DCS core control unit. It includes a valve actuator, a frequency converter, a bypass switching device, and a standby unit start-up module, which are used to perform flow regulation, speed control, loop switching, and standby equipment commissioning operations, respectively. The operations are issued by the DCS core control unit and correspond to the control strategy.

10. The fault early warning and self-healing control system for a gas turbine distributed control system according to claim 1, characterized in that, The human-computer interaction unit includes a host computer monitoring interface for displaying equipment operating parameters, fault warning information, and the execution status of control strategies.

Images