A remote data monitoring system for a combustion engine generator set

Abstract

The application discloses a kind of remote data monitoring systems of gas turbine generator set, specifically relates to gas turbine generator set data monitoring field, including working condition perception module, monitoring task allocation module, edge self-adaptive cooperation module and feedback module.A kind of remote data monitoring systems of gas turbine generator set working condition perception module is based on the dynamic resource demand coefficient of condition early warning model generation of multi-source heterogeneous data fusion, realized the prospective adjustment of resource allocation, from source reduced the processing pressure caused by sudden data volume;Through monitoring task allocation module, combine dynamic resource demand coefficient with segmented PID control algorithm, effectively alleviate the defect that fixed resource allocation mechanism in prior art cannot cope with sudden data processing demand when unit condition mutates suddenly;Through edge self-adaptive cooperation module, deploy lightweight data processing container, multi-level cache and priority marking, and optimize communication parameters, realize the adaptive cooperative optimization of data processing and communication transmission.

Description

Technical Field

[0001] This invention relates to the field of data monitoring technology for gas turbine generator sets, and more specifically, to a remote data monitoring system for gas turbine generator sets. Background Technology

[0002] With the increasing demand for intelligent operation and maintenance of gas turbine generator sets, the real-time and reliability requirements of remote data monitoring systems are becoming increasingly prominent. Traditional monitoring adopts a wired sensor network and local industrial control computer architecture, realizes data aggregation through RS485 bus, and relies on threshold comparison algorithms for status warning. However, due to technical bottlenecks such as high wiring complexity, poor scalability, and large data transmission delay fluctuations in actual operation, the monitoring coverage of remote units is low.

[0003] To address the aforementioned issues, existing technologies propose replacing wired devices by deploying low-power wireless sensor nodes and adding an edge computing gateway on the site side to achieve data preprocessing. Compared to traditional monitoring, this reduces wiring length and stabilizes end-to-end latency through an adaptive frequency hopping mechanism.

[0004] However, in actual use, it still has some shortcomings. For example, the fixed computing resource allocation mechanism of the edge computing gateway is difficult to cope with the sudden data processing needs when the unit's operating conditions change. When multiple units trigger high-frequency sampling simultaneously, the overflow rate of the processing queue of the edge node increases, resulting in the loss of key data frames. This leads to low data integrity rate at peak load, which directly affects the decision accuracy of the remote diagnostic system. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a remote data monitoring system for gas turbine generator sets, which solves the problems mentioned in the background art through the following solutions.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A remote data monitoring system for gas turbine generator sets includes a system operation database, a system central processing module, and a user information terminal, and further includes:

[0008] Operating condition perception module: used to collect equipment operation data of the target gas turbine generator set in real time, build an operating condition early warning model, and generate dynamic resource demand coefficients;

[0009] Monitoring task allocation module: Used to dynamically adjust and allocate monitoring tasks to the target gas turbine generator set based on the dynamic resource demand coefficient using a programmable logic architecture;

[0010] Edge adaptive collaboration module: used to receive the trigger signal from the monitoring task allocation module and perform edge adaptive collaboration operation, which includes data processing, communication optimization and knowledge transfer;

[0011] Feedback module: used to parse the feedback indicators stored in the database through the PID controller, generate a closed-loop feedback of resource allocation correction coefficients, and feed them back to the monitoring task allocation module;

[0012] The database includes all data text of a gas turbine generator set remote data monitoring system, and collects information text output by each module in real time. The central processing module is used for information text instructions output by each module in the central control system. The user terminal is a device for receiving information output from the gas turbine generator set remote data monitoring system.

[0013] Preferably, the working condition sensing module acquires the dynamic resource demand coefficient by including:

[0014] The data vibration signal is decomposed into sub-bands corresponding to multiple frequency bands;

[0015] The energy entropy of the sub-band energy corresponding to the target frequency band is used as the anomalous vibration feature. The calculation formula is specifically expressed as follows:

[0016]

[0017] in, This represents the number of sub-bands corresponding to the target frequency band. This represents the index of the sub-band corresponding to the target frequency band. Represented as the first Energy percentage of each individual;

[0018] The abnormal vibration characteristics are compared with the preset abnormal vibration warning threshold. If the calculated abnormal vibration characteristics exceed the preset abnormal vibration warning threshold, a warning signal based on the vibration characteristics is triggered.

[0019] The temperature gradient change rate of the target area is calculated based on the temperature distribution matrix formed by the temperature values ​​of multiple monitoring points.

[0020] The sliding window technique is used to use the standard deviation of the rate of change of the temperature gradient as a feature of the temperature gradient. Specifically, it is expressed as:

[0021]

[0022] in, This is represented by the number of data points within the sliding window. Represented as the first in the window The rate of change of the temperature gradient at each time step It is expressed as the average value of the rate of change of the temperature gradient within the sliding window;

[0023] The temperature gradient feature is compared with the preset temperature gradient warning threshold. If the calculated temperature gradient feature exceeds the preset temperature gradient warning threshold, a warning signal based on the temperature feature is triggered.

[0024] Preferably, the working condition sensing module, in acquiring the dynamic resource demand coefficient, further includes:

[0025] Dynamic resource demand coefficient The calculation formula is specifically expressed as follows:

[0026]

[0027] in, , , These represent the weights of the normalized abnormal vibration characteristics, the normalized temperature gradient characteristics, and the proportion of early warning signals, respectively. This is represented as abnormal vibration characteristics. This is represented as the abnormal vibration early warning threshold. Represented as a temperature gradient feature, This is represented as the temperature gradient warning threshold. This represents the number of warning signals implemented. This represents the total number of monitored signals.

[0028] Preferably, the monitoring task allocation module divides the computing resources of the edge computing gateway into a fixed resource pool and an elastic resource pool. The fixed resource pool is used for basic data acquisition tasks, and the elastic resource pool dynamically allocates the number of CPU cores based on the dynamic resource demand coefficient.

[0029] Preferably, the edge adaptive collaboration module includes a data processing unit, a communication optimization unit, and a communication optimization unit;

[0030] The data processing unit dynamically deploys lightweight data processing containers in the elastic resource pool through the Kubernetes orchestration engine. The number of replicas of the lightweight data processing containers is positively correlated with the amount of data backlog, and multi-level caching and priority marking are implemented for vibration data based on the improved RED algorithm.

[0031] The communication optimization unit analyzes RSSI volatility and channel duty cycle through the LoRaWAN physical layer, and dynamically adjusts the spreading factor and transmit power using a deep Q-learning algorithm.

[0032] The knowledge transfer unit encodes the abnormal waveforms detected by the lightweight data processing container into differential privacy knowledge vectors, which are then updated to the priority tag of the data processing unit via a federated learning framework.

[0033] Preferably, the edge adaptive collaboration module includes a data processing unit comprising:

[0034] Real-time monitoring of the backlog of data priority queues ,when When the backlog exceeds a preset threshold, the data processing container is horizontally expanded.

[0035] The dynamic adjustment of the number of replicas of the data processing container follows the following:

[0036]

[0037] in, This represents the number of replicas of a horizontally scaled data processing container. This is represented as the backlog in the data priority queue. This represents the estimated single-data-processing-container processing capacity coefficient when a single data-processing container is allocated the corresponding resources. This represents the average size of the data packets.

[0038] Preferably, the edge adaptive collaboration module includes a communication optimization unit comprising a channel state monitoring subunit and a transmission parameter optimization subunit.

[0039] The channel state monitoring subunit constructs an electromagnetic interference spectrum through signal quality analysis and outputs it to the transmission parameter optimization subunit;

[0040] The transmission parameter optimization subunit dynamically adjusts the spreading factor and transmission power based on the electromagnetic interference spectrum, and synchronously triggers the weight update of the knowledge transfer unit.

[0041] Preferably, the edge adaptive collaboration module, the knowledge transfer unit includes:

[0042] For the abnormal waveforms detected by the lightweight data processing container, an autoencoder is used to generate knowledge vectors, and differential privacy processing is performed on the knowledge vectors to obtain the gradient update amount of the lightweight data processing container after adding noise.

[0043] Based on the gradient update amount in the differential privacy knowledge vector, the priority label is updated periodically, specifically as follows:

[0044]

[0045] in, This represents the updated priority flag. This represents the number of lightweight data processing containers. This represents an index for a lightweight data processing container. Represented as the first The gradient update amount of a lightweight data processing container after adding noise. This is represented as the forgetting factor.

[0046] The technical effects and advantages of this invention are as follows:

[0047] 1. This invention generates dynamic resource demand coefficients based on a multi-source heterogeneous data fusion-based operating condition early warning model through an operating condition perception module, predicts resource demand when unit operating conditions change suddenly, realizes forward adjustment of resource allocation, and reduces processing pressure caused by sudden data volume from the source.

[0048] 2. This invention effectively alleviates the shortcomings of the existing fixed resource allocation mechanism in dealing with sudden data processing needs when the unit's operating conditions change, by combining the monitoring task allocation module with dynamic resource demand coefficients and piecewise PID control algorithm.

[0049] 3. This invention achieves adaptive collaborative optimization of data processing and communication transmission by deploying lightweight data processing containers, multi-level caching and priority marking, and optimizing communication parameters through an edge adaptive collaborative module. Attached Figure Description

[0050] Figure 1 This is a block diagram of a remote data monitoring system for a gas turbine generator set provided according to an embodiment of this application.

[0051] Figure 2 This is a block diagram of an edge adaptive collaborative module in a remote data monitoring system for gas turbine generator sets according to an embodiment of this application. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items; and in the description of the embodiments of this application, unless otherwise stated, “a plurality” means two or more.

[0054] As attached Figure 1 The remote data monitoring system for a gas turbine generator set shown includes a database, a central processing module and a user terminal, as well as an operating condition sensing module, a monitoring task allocation module, an edge adaptive collaboration module and a feedback module.

[0055] The database includes all data texts from a gas turbine generator set remote data monitoring system, and collects information texts output by each module in real time.

[0056] The central processing module is used for the text instructions output by each module in the central control system.

[0057] The user terminal is a device for receiving information output from a remote data monitoring system for gas turbine generator sets.

[0058] The operating condition perception module is used to collect equipment operation data of the target gas turbine generator set in real time, build an operating condition early warning model, and generate dynamic resource demand coefficients.

[0059] The monitoring task allocation module is used to dynamically adjust and allocate monitoring tasks for the target gas turbine generator set based on the dynamic resource demand coefficient using a programmable logic architecture.

[0060] The edge adaptive collaboration module is used to receive the trigger signal from the monitoring task allocation module and perform edge adaptive collaboration operations, which include data processing, communication optimization and knowledge transfer.

[0061] The feedback module is used to parse the feedback indicators stored in the database through the PID controller and generate a closed-loop feedback of the resource allocation correction coefficient to the monitoring task allocation module.

[0062] The central processing module is the core computing and control unit of the entire gas turbine generator remote data monitoring system. It includes one or more processing cores and is connected to sub-modules such as the operating condition sensing module and the edge adaptive collaboration module through PCIe bus and SPI interface. By running or executing container orchestration instructions, Q-learning algorithm code set and federated learning instruction set stored in the system operation database, and calling the vibration spectrum feature library stored therein, it can perform core operations such as dynamic resource scheduling, channel parameter optimization and abnormal knowledge transfer to ensure the data integrity rate of edge nodes under peak load.

[0063] The database is used to store a large amount of data related to gas turbine fault diagnosis, including fault prediction model parameters based on hidden Markov models and resource allocation strategy model weights based on reinforcement learning; it also stores historical operating data from the past five years, including vibration spectrum peak sequence, temperature gradient change rate matrix, and abnormal waveform feature vector library; when the system's central processor performs spectrum feature extraction, it will call harmonic energy ratio threshold data from the database in real time to perform operations such as calculating the early warning coefficient of sudden change in operating conditions, dynamically adjusting the number of container instances, and iteratively optimizing PID control parameters, thereby realizing adaptive resource allocation for concurrent high-frequency sampling of multiple units.

[0064] The user terminal connects to a 4K industrial display screen and an infrared thermal imaging camera via an HDMI 2.0 wired interface or an 802.11ax wireless protocol, providing users with a three-dimensional vibration spectrum visualization, real-time alarm information push, and a remote container deployment control interface, supporting cross-platform access.

[0065] The operating condition perception module collects real-time equipment operation data of the target gas turbine generator set, including but not limited to digital vibration signals, temperature data, fuel flow, speed and exhaust back pressure; and integrates multi-source heterogeneous equipment operation data through an operating condition early warning model to assess the operating status of the target gas turbine generator set in real time, issue early warning signals before potential faults occur, and output dynamic resource demand coefficients to guide the monitoring task allocation module in allocating monitoring tasks.

[0066] In this embodiment, the acquisition of equipment operation data includes: deploying an IEPE-type accelerometer in the gas turbine rotor bearing, the accelerometer being used to acquire triaxial vibration signals during gas turbine operation; preprocessing the acquired vibration signals using an anti-aliasing filter, the cutoff frequency of which is set to 20kHz; arranging a K-type thermocouple array at the compressor inlet, combustion chamber shell, and turbine blade root of the gas turbine, the thermocouple array simultaneously acquiring temperature values ​​from multiple monitoring points inside and outside the gas turbine at a sampling rate of 10Hz to form a temperature distribution matrix; measuring fuel flow rate using a Coriolis mass flow meter with a high accuracy of 0.2%; measuring the gas turbine rotational speed using a magnetoelectric sensor with a resolution of 1rpm; and measuring the exhaust back pressure at the gas turbine exhaust port using a piezoresistive sensor with a range of 0-5bar.

[0067] In one possible implementation, the construction of the operating condition early warning model includes: performing time-frequency analysis on the digital vibration signal in the equipment operating data; in this embodiment, wavelet packet decomposition is used to decompose the data vibration signal into sub-bands corresponding to multiple frequency bands; and using the energy entropy of the sub-band energy corresponding to the target frequency band as the abnormal vibration feature. In this embodiment, the target frequency band is 2-4kHz. The calculation formula is specifically expressed as follows:

[0068]

[0069] in, This represents the number of sub-bands corresponding to the target frequency band. This represents the index of the sub-band corresponding to the target frequency band. Represented as the first The energy percentage of each component; the abnormal vibration characteristics are compared with a preset abnormal vibration warning threshold. In this embodiment, the preset abnormal vibration warning threshold is 0.25. If the calculated abnormal vibration characteristics exceed the preset abnormal vibration warning threshold, a warning signal based on the vibration characteristics is triggered; based on the temperature distribution matrix formed by the temperature values ​​of multiple monitoring points, the temperature gradient change rate of the target area is calculated. The temperature gradient change rate of the target area includes, but is not limited to, the temperature gradient change rate along the axial direction of the gas turbine and the critical temperature difference along the radial direction; the standard deviation of the temperature gradient change rate is used as the temperature gradient characteristic using the sliding window technique. In this embodiment, the window width is set to 60 seconds and the step size is 10 seconds, specifically as follows:

[0070]

[0071] in, This is represented by the number of data points within the sliding window. Represented as the first in the window The rate of change of the temperature gradient at each time step This is represented as the average rate of change of the temperature gradient within the sliding window. The temperature gradient feature is compared with a preset temperature gradient warning threshold. In this embodiment, the preset temperature gradient warning threshold is 3℃ / s. If the calculated temperature gradient feature exceeds the preset temperature gradient warning threshold, a warning signal based on the temperature feature is triggered. The abnormal vibration feature, temperature gradient feature, and synchronously collected equipment operation data are fused to construct the feature vector of the input working condition warning model. In this embodiment, the working condition warning model is a lightweight LSTM network, and the Sigmoid activation function is used to output a value between 0 and 1, which is the dynamic resource demand coefficient. Furthermore, the dynamic resource demand coefficient The calculation formula is specifically expressed as follows:

[0072]

[0073] in, , , These represent the weights of the normalized abnormal vibration characteristics, the normalized temperature gradient characteristics, and the proportion of early warning signals, respectively. This is represented as abnormal vibration characteristics. This is represented as the abnormal vibration early warning threshold. Represented as a temperature gradient feature, This is represented as the temperature gradient warning threshold. This represents the number of warning signals implemented. This represents the total number of monitored signals.

[0074] The monitoring task allocation module adopts an FPGA programmable logic architecture, which divides the computing resources of the edge computing gateway into a fixed resource pool and an elastic resource pool. The fixed resource pool accounts for 30% of the total resources and is dedicated to basic data acquisition tasks.

[0075] In this embodiment, the total resources are logically divided into a fixed resource pool and an elastic resource pool. The fixed resource pool occupies 30% of the total computing resources and is allocated fixed hardware resources, including but not limited to three high-performance ARM Cortex-A53 cores and 512MB of DDR4 memory. The fixed resource pool is dedicated to performing basic data acquisition and preprocessing tasks that have extremely high real-time requirements and must run continuously, including but not limited to receiving data from the RS485 bus, performing Modbus-TCP protocol parsing, performing CRC-32 verification, and preliminary data packet encapsulation. The elastic resource pool occupies approximately 70% of the total computing resources and contains dynamically adjustable computing resources, including but not limited to four real-time ARM Cortex-R5 cores and 2GB of DDR4 memory.

[0076] Furthermore, the monitoring task allocation module receives dynamic resource demand coefficients output in real time from the working condition sensing module. The mapping strategy based on the piecewise PID control algorithm will... With the target resource allocation status Compare and calculate the amount of elastic resource allocation that needs to be adjusted. .

[0077] In this embodiment, the calculation process of the piecewise PID control algorithm, which is executed every 200ms, is as follows: Real-time calculation of the proportional term. ,in The proportional gain coefficient; the integral term of the accumulated deviation over a preset time period. in This is the integral gain coefficient. Represented as sampling points within a preset time period; the differential term for predicting future trends. in The differential gain coefficient is used; by adding the proportional, integral, and differential terms, the amount of flexible resource allocation that needs to be adjusted is obtained. ,in It is represented as a clamping function and written to the FPGA configuration register.

[0078] The edge adaptive collaboration module includes a data processing unit, a communication optimization unit, and a communication optimization unit.

[0079] The data processing unit dynamically deploys lightweight data processing containers in the elastic resource pool through the Kubernetes orchestration engine. The number of replicas of the lightweight data processing containers is positively correlated with the amount of data backlog, and multi-level caching and priority marking are implemented for vibration data based on the improved RED algorithm.

[0080] Furthermore, the data processing unit includes a containerized scheduling subunit and a multi-level caching subunit. The containerized scheduling subunit is built on Kubernetes to create a lightweight edge cluster, further refining the computing resources allocated by the monitoring task allocation module from the elastic resource pool and dividing them into allocatable units with a granularity of 500MHz. Simultaneously, a container image for data processing is deployed, with a size less than 15MB and pre-installed with data processing libraries. In this embodiment, the data processing libraries include, but are not limited to, an FFT analysis library for vibration feature extraction and machine learning models. The backlog of the data priority queue is monitored in real time using the Prometheus monitoring tool. ,when When the backlog exceeds a preset threshold, horizontal expansion of the data processing container is triggered, and the dynamic adjustment of the number of replicas of the data processing container follows the following:

[0081]

[0082] in, This represents the number of replicas of a horizontally scaled data processing container. This is represented as the backlog in the data priority queue. This represents the estimated single-data-processing-container processing capacity coefficient when a single data-processing container is reallocated to corresponding resources, with a default value of 2000 data packets / second. This is expressed as the average size of the data packets. This is represented as rounding up; the multi-level cache sub-unit implements the RED algorithm and combines it with a dual-threshold queue mechanism. Data is graded and marked according to the instantaneous length of the data priority queue. For data marked as high priority, memory passthrough technology is used to realize the transfer of data between the cache and the container.

[0083] In this embodiment, the hierarchical marking follows these rules: when the instantaneous queue length is less than 50, the data is marked as L1 priority, indicating that the queue load is low and the data is processed according to the normal process; when the instantaneous queue length is between 50 and 120, it is marked as L2 priority, indicating that the queue load is starting to increase and triggering the LZ4 compression pipeline for compression processing; when the instantaneous queue length is greater than or equal to 120, it is marked as L3 priority, indicating that the queue load is high, the system is facing processing pressure, and data downsampling is triggered.

[0084] The communication optimization unit analyzes RSSI volatility and channel duty cycle through the LoRaWAN physical layer and dynamically adjusts the spreading factor and transmit power using a deep Q-learning algorithm.

[0085] Furthermore, the communication optimization unit includes a channel state monitoring subunit and a transmission parameter optimization subunit. The channel state monitoring subunit constructs an electromagnetic interference spectrum through LoRaWAN physical layer signal quality analysis and outputs it to the transmission parameter optimization subunit. The transmission parameter optimization subunit adopts a deep Q-learning algorithm to dynamically adjust the spreading factor and transmit power according to the interference spectrum. Its parameter update signal synchronously triggers the weight update of the knowledge transfer unit.

[0086] It should be noted that the channel state monitoring subunit integrates a LoRa baseband chip and utilizes data acquisition capabilities to collect the RSSI value of the received LoRaWAN data packets and the busy status of the target LoRaWAN channel in real time, so as to construct an electromagnetic interference map represented by a two-dimensional vector.

[0087] Furthermore, within a preset sliding time window (in this embodiment, the width of the preset sliding time window is 100), the RSSI volatility is calculated. Specifically, it is expressed as:

[0088]

[0089] in, This represents the number of RSSI samples within a preset sliding time window. This represents the index of the RSSI sample within the preset sliding time window. Represented as the first time within the preset sliding time window RSSI samples, This is expressed as the average of all RSSI samples within a preset sliding time window; it should be noted that higher RSSI volatility... This indicates that the channel exhibits strong multipath effects or sudden interference; within a preset statistical period. Within this embodiment, the preset statistical period is 1 second, and the statistical channel is in a busy state. The total duration is used to obtain the channel duty cycle. Specifically, it is expressed as:

[0090]

[0091] It should be noted that the transmission parameter optimization subunit learns to select the optimal combination of spreading factor and transmit power under multiple channel states, and sends a parameter update signal based on the electromagnetic interference spectrum to trigger the weight update of the knowledge transfer unit.

[0092] In this embodiment, the state space of the deep Q-learning algorithm in the transmission parameter optimization subunit includes a target monitoring state vector, which is composed of the spreading factor used by the monitored target, the transmit power of the monitored target, and the real-time sensed electromagnetic interference spectrum. In each decision cycle, the action space selects to execute discrete actions, which include: increasing the spreading factor by one level, decreasing the spreading factor by one level, increasing the transmit power, and decreasing the transmit power. In this embodiment, the spreading factor is maintained in the range of SF7-SF12, the transmit power is maintained in the range of 14-20dBm, and when the execution of the action causes the parameters to exceed the range, the parameters are maintained at the boundary values.

[0093] The knowledge transfer unit encodes the abnormal waveforms detected by the lightweight data processing container into differential privacy knowledge vectors, which are then updated to the priority tag of the data processing unit via a federated learning framework.

[0094] It should be noted that for the abnormal waveforms detected by the lightweight data processing container, an autoencoder is used to generate knowledge vectors, and differential privacy processing is applied to these knowledge vectors to obtain the gradient update amount of the lightweight data processing container after adding noise. Specifically, it is expressed as:

[0095]

[0096] in, This represents the original gradient update amount for the lightweight data processing container. Represented as Laplace noise, This represents the maximum gradient update amount of the lightweight data processing container. This represents the minimum gradient update amount for a lightweight data processing container. This is represented as a privacy budget, used to control the intensity of noise; in this embodiment, ϵ=0.5. Based on the gradient update amount in the differential privacy knowledge vector, the priority label is periodically updated, specifically as follows:

[0097]

[0098] in, This represents the updated priority flag. This represents the number of lightweight data processing containers. This represents an index for a lightweight data processing container. Represented as the first The gradient update amount of a lightweight data processing container after adding noise. This is represented as a forgetting factor, used to prevent old priority tags from adversely affecting new priority tags. In this embodiment... ;

[0099] The feedback module collects feedback indicators from the remote data monitoring system, generates resource allocation correction coefficients through the PID controller, and feeds them back to the elastic resource pool of the monitoring task allocation module in a closed loop.

[0100] In this embodiment, the feedback indicators include, but are not limited to, false alarm rate, data integrity rate, and processing latency. The false alarm rate is the frequency of false alarms generated within a preset statistical period. The data integrity rate is the ratio of successfully uploaded data packets to the total number of data packets that should be uploaded within the statistical period. The processing latency is the end-to-end time delay from the acquisition of raw data from the edge node to its storage in the database. A resource allocation correction coefficient is calculated based on the PID controller. If the absolute value of the resource allocation correction coefficient is greater than 0.3, it indicates that the performance deviates significantly from the target and requires a rapid response; therefore, the proportional and integral gain coefficients are increased. If the absolute value of the resource allocation correction coefficient is greater than 0.1 and less than or equal to 0.3, it indicates that the performance is within a moderate deviation range and no changes are needed. If the absolute value of the resource allocation correction coefficient is less than or equal to 0.1, it indicates that the performance is close to the target value; to avoid excessive oscillation, the proportional and integral gain coefficients are increased and decreased.

[0101] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0102] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A remote data monitoring system for gas turbine generator sets, comprising a database, a central processing module, and a user terminal, characterized in that, Also includes: Operating condition perception module: used to collect equipment operation data of the target gas turbine generator set in real time, build an operating condition early warning model, and generate dynamic resource demand coefficients; Wherein, the dynamic resource demand coefficient The calculation formula is specifically expressed as follows: in, , , These represent the weights of the normalized abnormal vibration characteristics, the normalized temperature gradient characteristics, and the proportion of early warning signals, respectively. This is represented as abnormal vibration characteristics. This is represented as the abnormal vibration early warning threshold. Represented as a temperature gradient feature, This is represented as the temperature gradient warning threshold. This represents the number of warning signals implemented. This represents the total number of monitored signals; Monitoring task allocation module: Used to dynamically adjust and allocate monitoring tasks to the target gas turbine generator set based on the dynamic resource demand coefficient using a programmable logic architecture; The mapping strategy based on the piecewise PID control algorithm will incorporate the dynamic resource demand coefficient. With the target resource allocation status Compare and calculate the amount of elastic resource allocation that needs to be adjusted. The calculation process of the segmented PID control algorithm, which is executed every 200ms, is as follows: Real-time calculation of scale terms ,in This is the proportional gain coefficient; The integral term of the cumulative deviation within the preset time period in This is the integral gain coefficient. These represent sampling points within a preset time period; Differential term for predicting future trends in The differential gain coefficient; Adding the proportional, integral, and differential terms yields the amount of flexible resource allocation that needs adjustment. ,in This is represented as a clamping function and written to the FPGA configuration register; Edge adaptive collaboration module: used to receive the trigger signal from the monitoring task allocation module and perform edge adaptive collaboration operation, which includes data processing, communication optimization and knowledge transfer; Feedback module: used to parse the feedback indicators stored in the database through the PID controller, generate a closed-loop feedback of resource allocation correction coefficients, and feed them back to the monitoring task allocation module; The database includes all data text of a gas turbine generator set remote data monitoring system, and collects information text output by each module in real time. The central processing module is used for information text instructions output by each module in the central control system. The user terminal is a device for receiving information output from the gas turbine generator set remote data monitoring system.

2. The remote data monitoring system for gas turbine generator sets according to claim 1, characterized in that: The operating condition sensing module obtains dynamic resource demand coefficients including: The data vibration signal is decomposed into sub-bands corresponding to multiple frequency bands; The energy entropy of the sub-band energy corresponding to the target frequency band is used as the anomalous vibration feature. The calculation formula is specifically expressed as follows: in, This represents the number of sub-bands corresponding to the target frequency band. This represents the index of the sub-band corresponding to the target frequency band. Represented as the first Energy percentage of each individual; The abnormal vibration characteristics are compared with the preset abnormal vibration warning threshold. If the calculated abnormal vibration characteristics exceed the preset abnormal vibration warning threshold, a warning signal based on the vibration characteristics is triggered. The temperature gradient change rate of the target area is calculated based on the temperature distribution matrix formed by the temperature values ​​of multiple monitoring points. The sliding window technique is used to use the standard deviation of the rate of change of the temperature gradient as a feature of the temperature gradient. Specifically, it is expressed as: in, This is represented by the number of data points within the sliding window. Represented as the first in the window The rate of change of the temperature gradient at each time step It is expressed as the average value of the rate of change of the temperature gradient within the sliding window; The temperature gradient feature is compared with the preset temperature gradient warning threshold. If the calculated temperature gradient feature exceeds the preset temperature gradient warning threshold, a warning signal based on the temperature feature is triggered.

3. The remote data monitoring system for gas turbine generator sets according to claim 1, characterized in that: The monitoring task allocation module divides the computing resources of the edge computing gateway into a fixed resource pool and an elastic resource pool. The fixed resource pool is used for basic data collection tasks, and the elastic resource pool dynamically allocates the number of CPU cores based on the dynamic resource demand coefficient.

4. The remote data monitoring system for gas turbine generator sets according to claim 1, characterized in that: The edge adaptive collaboration module includes a data processing unit, a communication optimization unit, and a knowledge transfer unit. The data processing unit dynamically deploys lightweight data processing containers in the elastic resource pool through the Kubernetes orchestration engine. The number of replicas of the lightweight data processing containers is positively correlated with the amount of data backlog, and multi-level caching and priority marking are implemented for vibration data based on the improved RED algorithm. The communication optimization unit analyzes RSSI volatility and channel duty cycle through the LoRaWAN physical layer, and dynamically adjusts the spreading factor and transmit power using a deep Q-learning algorithm. The knowledge transfer unit encodes the abnormal waveforms detected by the lightweight data processing container into differential privacy knowledge vectors, which are then updated to the priority tag of the data processing unit via a federated learning framework.

5. The remote data monitoring system for gas turbine generator sets according to claim 4, characterized in that: The edge adaptive collaboration module, the data processing unit includes: Real-time monitoring of the backlog of data priority queues ,when When the backlog exceeds a preset threshold, the data processing container is horizontally expanded. The dynamic adjustment of the number of replicas of the data processing container follows the following: in, This represents the number of replicas of a horizontally scaled data processing container. This is represented as the backlog in the data priority queue. This represents the estimated single-data-processing-container processing capacity coefficient when a single data-processing container is allocated the corresponding resources. This represents the average size of the data packets.

6. The remote data monitoring system for gas turbine generator sets according to claim 4, characterized in that: The edge adaptive collaboration module includes a communication optimization unit comprising a channel state monitoring subunit and a transmission parameter optimization subunit. The channel state monitoring subunit constructs an electromagnetic interference spectrum through signal quality analysis and outputs it to the transmission parameter optimization subunit; The transmission parameter optimization subunit dynamically adjusts the spreading factor and transmission power based on the electromagnetic interference spectrum, and synchronously triggers the weight update of the knowledge transfer unit.

7. The remote data monitoring system for gas turbine generator sets according to claim 4, characterized in that: The edge adaptive collaboration module, the knowledge transfer unit includes: For the abnormal waveforms detected by the lightweight data processing container, an autoencoder is used to generate knowledge vectors, and differential privacy processing is performed on the knowledge vectors to obtain the gradient update amount of the lightweight data processing container after adding noise. Based on the gradient update amount in the differential privacy knowledge vector, the priority label is updated periodically, specifically as follows: in, This represents the updated priority flag. This represents the number of lightweight data processing containers. This represents an index for a lightweight data processing container. Represented as the first The gradient update amount of a lightweight data processing container after adding noise. This is represented as the forgetting factor.

Images