SF6 gas monitoring data online processing system

Abstract

The invention discloses an SF6 gas monitoring data online processing system. The system comprises a multi-mode sensing unit, an edge calculation preprocessing unit, an Internet of Things communication transmission unit, a cloud deep reinforcement learning and Transform fusion processing unit, a decision analysis unit and an interactive display unit. The multi-mode sensing unit collects SF6 gas pressure, temperature and other parameters in real time, and the parameters are uploaded to the cloud through the Internet of Things communication transmission unit after edge calculation preprocessing. The cloud end utilizes a deep reinforcement learning and Transform fusion model to mine a complex relationship and a change trend among parameters, the decision analysis unit evaluates a gas state according to the complex relationship and the change trend, and the interactive display unit visually displays a result. The system overcomes the problems of low manual efficiency, shallow data processing, poor self-adaption and the like of traditional monitoring, realizes accurate, real-time and intelligent monitoring of the SF6 gas density, and provides reliable guarantee for safe operation and intelligent operation and maintenance of substation equipment.

Description

Technical Field

[0001] The present invention relates to the field of SF6 gas monitoring in transformer substations, and in particular to an online processing system for SF6 gas monitoring data. Background Art

[0002] The safe and stable operation of substation equipment is crucial to the development of smart grids. SF6 gas is widely used in high-voltage electrical equipment due to its excellent insulation and arc-extinguishing properties. However, abnormal SF6 gas density can threaten the safe operation of equipment and even cause failures. Traditional monitoring methods and some existing online monitoring technologies are no longer able to meet the needs of precise and intelligent operation and maintenance, making the development of new monitoring systems urgent.

[0003] There are two major outstanding problems with existing technologies. First, data processing and analysis capabilities are insufficient. Traditional monitoring technologies mostly rely on simple threshold judgments, which are unable to explore the complex correlations between multiple parameters such as SF6 gas pressure, temperature, and humidity, and it is difficult to capture subtle changes in gas density trends. Under the interference of environmental factors, misjudgments or missed judgments are prone to occur, and potential hidden dangers of equipment cannot be discovered in advance. Second, the system lacks adaptive adjustment capabilities. The existing online monitoring system has fixed parameters such as acquisition frequency and transmission rate, and cannot be dynamically adjusted according to data change characteristics and system load. When the data volume is large, the transmission efficiency is low and the cloud is prone to overload; when the data is stable, there is a waste of resources, and it is difficult to adapt to the complex and changing operating environment of the substation. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of the prior art, the present invention provides an SF6 gas monitoring data online processing system.

[0005] The technical solution adopted by the present invention is an SF6 gas monitoring data online processing system, comprising: A multimodal sensing unit integrates multiple types of high-precision sensors to sense and collect the pressure, temperature, density, humidity, moisture content, and gas leakage rate of SF6 gas in the substation in real time, and converts the collected analog signals into digital signals. An edge computing preprocessing unit, which is connected to the multimodal sensing unit, receives the digital signal transmitted by the multimodal sensing unit, performs preliminary noise reduction, filtering and data format standardization on the digital signal, and performs frame processing on the data; The IoT communication transmission unit is connected to the edge computing pre-processing unit and the cloud processing unit respectively. It uses multiple communication protocols to encapsulate the framed data processed by the edge computing pre-processing unit and transmits the data stably and reliably to the cloud processing unit via a wireless network or a wired network. The cloud-based deep reinforcement learning and Transformer fusion processing unit receives data transmitted by the IoT communication transmission unit and builds a model architecture that combines deep reinforcement learning and Transformer. This architecture is used to perform feature extraction, time series analysis, and complex relationship mining on the data. The decision analysis unit is connected to the cloud-based deep reinforcement learning and Transformer fusion processing unit, receives the processed data analysis results, and conducts a comprehensive assessment and decision on the status of the substation's SF6 gas based on pre-set rules and thresholds; The interactive display unit is connected to the decision analysis unit to display the evaluation and decision results of the decision analysis unit in the form of visual charts, graphs and data lists, and supports users to query, analyze and export monitoring data through interactive operations.

[0006] Furthermore, the fusion model constructed in the cloud-based deep reinforcement learning and Transformer fusion processing unit uses the following formula for data processing: in, Represents the output data after processing by the fusion model; represents the action-value function output by the deep reinforcement learning module, The current state, including the pressure of SF6 gas ,temperature ,density ,humidity , trace water content , gas leakage rate The state space composed of parameters, For the actions taken, are the parameters of the deep reinforcement learning model; Transformer It is the Transformer processing module; It is a function for characteristic transformation of various parameters of SF6 gas.

[0007] Furthermore, the cloud-based deep reinforcement learning and Transformer fusion processing unit also includes the following formula for feature optimization: in, For the Optimized weights of individual data features; is the association weight between data; is the number of data samples; is the activation function; For SF6 gas A function that uses a set of parameters to perform feature extraction and transformation.

[0008] Furthermore, the decision analysis unit adopts the following decision formula based on deep reinforcement learning and Transformer model: in, Indicates the decision result; is the indicator function; is the number of decision data samples; For the The weight of individual data in decision making; is the decision threshold; For SF6 gas A function that takes group parameters for feature processing.

[0009] Furthermore, there is a data collaborative optimization mechanism between the edge computing preprocessing unit and the cloud-based deep reinforcement learning and Transformer fusion processing unit. The formula is as follows: in, Represents the data after collaborative optimization; Data processed by the edge computing pre-processing unit; is the fusion coefficient; They are the state, action, and parameters of deep reinforcement learning in edge computing scenarios; This is a function for performing feature transformation on SF6 gas parameters collected by edge computing.

[0010] Furthermore, the sensor data collection frequency of the multimodal perception unit is correlated with the model update of the cloud-based deep reinforcement learning and Transformer fusion processing unit, and the formula is: in, Update frequency for cloud models; is the adjustment coefficient; is the number of sensors; is the acquisition frequency of the sensth sensor; is the minimum acquisition frequency threshold; are the states, actions, and parameters of deep reinforcement learning related to the acquisition frequency, respectively.

[0011] Furthermore, there is an adaptive adjustment relationship between the data transmission rate of the IoT communication transmission unit and the computing load of the cloud-based deep reinforcement learning and Transformer fusion processing unit, as shown in the following formula: in, is the data transmission rate; is the adjustment factor; Calculate the load for the current cloud; Calculate the load for average; are the states, actions, and parameters of deep reinforcement learning related to transmission, respectively; It is a function for performing feature processing on the SF6 gas parameters involved in the transmission data.

[0012] Furthermore, there is a dynamic mapping relationship between the visual interface layout of the interactive display unit and the decision result of the decision analysis unit, and the formula is: in, It is the layout parameter of the visual interface; is the layout adjustment coefficient; is the number of decision results; For the The weight of each decision outcome; are the states, actions, and parameters of deep reinforcement learning related to decision outcomes; It is a function for performing characteristic processing on SF6 gas parameters related to the decision result.

[0013] Furthermore, the system also includes an abnormal data tracing module, which uses the following formula to locate abnormal data: in, is the location index of abnormal data; is the mean vector of normal data; For the Function for performing characteristic processing on a group of SF6 gas parameters.

[0014] An SF6 gas monitoring data online processing system, the operation of which includes the following steps: In the first step, the multimodal sensing unit periodically collects the pressure, temperature, density, humidity, moisture content, and gas leakage rate parameters of the SF6 gas in the substation in real time according to the set sampling strategy, and converts the collected analog signals into digital signals. The second step is to transmit the digital signal collected by the multimodal sensing unit to the edge computing preprocessing unit, which performs noise reduction, filtering, and data format standardization on the digital signal, and frames the data according to preset rules. In the third step, the framed data processed by the edge computing pre-processing unit is encapsulated by the IoT communication transmission unit using a specific communication protocol and stably transmitted to the cloud processing unit via a wireless or wired network. In the fourth step, the cloud-based deep reinforcement learning and Transformer fusion processing unit in the cloud processing unit receives the transmitted data and uses the constructed deep reinforcement learning and Transformer combined model architecture to perform feature extraction, time series analysis, and complex relationship mining on the data; In the fifth step, the decision analysis unit receives the data analysis results from the cloud-based deep reinforcement learning and Transformer fusion processing unit, and conducts a comprehensive assessment and decision on the status of the substation's SF6 gas based on pre-set rules and thresholds. In the sixth step, the interactive display unit presents the evaluation and decision results of the decision analysis unit in the form of visual charts, graphs and data lists, and provides an interactive operation interface so that users can query, analyze and export the monitoring data.

[0015] Beneficial effects: The present invention proposes an online processing system for SF6 gas monitoring data. The system adopts a multimodal sensing unit and an edge computing preprocessing unit to realize real-time and accurate collection and preliminary processing of multiple parameters such as SF6 gas pressure, temperature, and density. Compared with traditional manual inspections, it not only greatly improves the timeliness and continuity of data collection, but also reduces data errors through standardized processing, and solves the problem of manual measurement being interfered with by environmental and human factors. At the data processing and analysis level, the cloud-based deep reinforcement learning and Transformer fusion processing unit can deeply explore the complex correlation and time series characteristics between multiple parameters of SF6 gas, changing the limitations of traditional monitoring technology that relies only on simple threshold judgment, accurately identifying subtle changes in gas density trends, and significantly improving the accuracy and predictability of anomaly detection. In response to the problem of poor adaptability of existing systems, this system realizes dynamic coordination of data collection frequency, transmission rate and cloud computing load. For example, the sensor collection frequency and data transmission rate are automatically adjusted according to the data change characteristics and system load, while ensuring effective data transmission and processing, avoiding resource waste, improving system operation efficiency, and better adapting to the complex and changeable operating environment of substations. In addition, the design of the decision analysis unit and the interactive display unit ensures that the monitoring results can be presented in an intuitive and interactive form, providing comprehensive and timely decision-making basis for operation and maintenance personnel, and effectively improving the intelligence level and operation and maintenance efficiency of substation SF6 gas monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a diagram of the system unit composition of the present invention; Figure 2 It is a flow chart of the system operation steps of the present invention. DETAILED DESCRIPTION

[0017] It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of this application can be combined with each other. The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1 As shown, an SF6 gas monitoring data online processing system includes: A multimodal sensing unit integrates multiple types of high-precision sensors to sense and collect various parameters of SF6 gas in the substation, including pressure, temperature, density, humidity, trace water content, and gas leakage rate, in real time, and convert the collected analog signals into digital signals. Specifically, the multimodal sensing unit, as the "data source" of the entire monitoring system, integrates a variety of high-precision sensors. Among them, the pressure sensor uses a high-precision piezoresistive sensor with a range covering 0-1.0MPa and an accuracy of up to ±0.2%FS, which can accurately capture changes in SF6 gas pressure; the temperature sensor uses a platinum resistance temperature sensor with a temperature measurement range of -40℃-125℃ and an accuracy of ±0.1℃, which can effectively eliminate the interference of ambient temperature on gas density measurement; the density sensor uses an oscillating densitometer with a measurement accuracy of ±0.1kg / m³, ensuring the reliability of gas density data; the humidity and trace water content sensor uses a polymer thin film capacitive sensor, which can achieve accurate measurement of trace water content in the range of 0-1000μL / L with an accuracy of ±5%; the gas leakage rate sensor is based on the ultrasonic principle and can detect the minimum leakage of These sensors collect data in real time at a sampling frequency of 100Hz and convert analog signals into digital signals via built-in A / D converters, providing the raw data foundation for subsequent data processing. The units are installed directly at key nodes of SF6 gas equipment in substations, such as circuit breaker and disconnector chambers, to ensure that the collected data truly reflects the gas status within the equipment.

[0019] An edge computing preprocessing unit, which is connected to the multimodal sensing unit, receives the digital signal transmitted by the multimodal sensing unit, performs preliminary noise reduction, filtering and data format standardization on the digital signal, and performs frame processing on the data; Specifically, the edge computing preprocessing unit is closely connected to the multimodal sensing unit and undertakes the crucial task of initial data processing. For data noise reduction, a composite filtering algorithm combining median and mean filtering is employed to effectively remove impulse and random noise from sensor data. Data format standardization, based on the International Electrotechnical Commission (IEC) power system data communication standards, converts raw data collected by different sensor types into a standardized format. Furthermore, the unit frames the data, averaging every 100 sampling points, to optimize data transmission efficiency and cloud processing performance. In terms of hardware implementation, the edge computing preprocessing unit utilizes an embedded microprocessor platform equipped with a high-performance ARM processor and a dedicated digital signal processing chip. This platform offers powerful real-time data processing capabilities, enabling data preprocessing at the data collection site. This reduces data transmission volume and cloud computing pressure, lowering overall system latency and improving the real-time performance and responsiveness of the monitoring system.

[0020] The IoT communication transmission unit is connected to the edge computing pre-processing unit and the cloud processing unit respectively. It uses multiple communication protocols to encapsulate the framed data processed by the edge computing pre-processing unit and transmits the data stably and reliably to the cloud processing unit via a wireless network or a wired network. Specifically, the IoT communication transmission unit builds a data bridge between the edge computing pre-processing unit and the cloud processing unit. This unit supports 4G / 5G wireless network communication protocols, with transmission rates up to 1 Gbps. It is also compatible with wired network communication methods such as Ethernet and fiber optics, allowing for flexible transmission methods tailored to the substation's actual network environment. During data encapsulation, it follows the TCP / IP protocol stack and encapsulates the framed data in the JSON data format. A header containing information such as the data source, acquisition time, and frame sequence number is added to ensure data integrity and accuracy during transmission. To ensure data transmission stability, the unit integrates an automatic repeat request (ARQ) mechanism, automatically triggering retransmissions when data transmission errors or loss are detected. Link adaptation technology also dynamically adjusts the transmission rate and encoding method based on network signal strength and bandwidth, such as reducing the transmission rate and increasing encoding redundancy when the network signal is weak, ensuring stable and reliable data transmission. In actual deployment, the IoT communication transmission unit is deployed in the substation's communications room and connected to the edge computing nodes and cloud servers via dedicated communication lines.

[0021] The cloud-based deep reinforcement learning and Transformer fusion processing unit receives data transmitted by the IoT communication transmission unit and builds a model architecture that combines deep reinforcement learning and Transformer. This architecture is used to perform feature extraction, time series analysis, and complex relationship mining on the data. Specifically, the cloud-based deep reinforcement learning and Transformer fusion processing unit is the core data processing hub of the entire system. The deep reinforcement learning module constructs a neural network architecture containing multiple fully connected layers, takes the pressure, temperature, density and other parameters of SF6 gas as state input, and optimizes the action strategy through continuous interactive learning with the environment to predict the trend of gas state changes. The Transformer module uses a multi-head attention mechanism to model the complex relationship between the time series and parameters of the input data and effectively extract data features. In actual operation, this unit adopts a distributed computing architecture and is deployed on a high-performance server cluster in the cloud data center, using GPU acceleration technology to improve computing efficiency. Through training and learning of massive historical data, the system can accurately identify the change pattern of SF6 gas parameters. For example, in the early stage of a small leak in the equipment, it can promptly capture the subtle change trend of parameters such as pressure and density, providing accurate data support for subsequent decision-making. Compared with traditional data processing methods, it significantly improves the depth of data processing and analytical capabilities.

[0022] The decision analysis unit is connected to the cloud-based deep reinforcement learning and Transformer fusion processing unit, receives the processed data analysis results, and conducts a comprehensive assessment and decision on the status of the substation's SF6 gas based on pre-set rules and thresholds; Specifically, the decision analysis unit receives analysis results from the cloud-based deep reinforcement learning and Transformer fusion processing unit and performs a comprehensive assessment and decision-making based on pre-set multi-level decision rules and thresholds. These rules and thresholds are derived from power industry standards and statistical analysis of extensive historical data. For example, a Level 1 alarm is triggered when the SF6 gas density decreases at a rate exceeding 0.05 kg / m³ / day and the pressure drops by more than 0.02 MPa within 24 hours; a Level 2 alarm is triggered when the gas moisture content exceeds 500 μL / L. The decision analysis unit utilizes a hierarchical decision-making mechanism, classifying gas status into multiple levels—normal, warning, and fault—and formulates corresponding action recommendations for each level. In terms of hardware implementation, the decision analysis unit is deployed on a cloud server. Using efficient algorithms and data processing logic, it rapidly evaluates and makes decisions based on input data, providing operations and maintenance personnel with timely and accurate equipment status information so they can implement targeted operations and maintenance measures to ensure the safe and stable operation of substation equipment.

[0023] The interactive display unit is connected to the decision analysis unit to display the evaluation and decision results of the decision analysis unit in the form of visual charts, graphs and data lists, and supports users to query, analyze and export monitoring data through interactive operations.

[0024] Specifically, the interactive display unit serves as the user interface between the monitoring system and the user, presenting monitoring data and decision-making results in an intuitive and visual manner. This unit utilizes a browser-based architecture, allowing users to access the system interface through a browser without installing additional client software. For data display, visualization technologies such as Echarts and D3.js are used to display SF6 gas parameter changes in various chart formats, including line charts, bar charts, and dashboards. Historical data comparison and analysis are also available, allowing users to view data trends within a customizable time range. To facilitate user operation, the interactive display unit offers a variety of interactive features, such as data query and export. Users can query specific data by entering criteria such as device name and time range, and export data to formats such as Excel and CSV. Furthermore, it features real-time alert notification. When the decision analysis unit triggers an alarm, the system promptly notifies relevant operations and maintenance personnel via pop-up windows, email, and SMS messages. In practical applications, this unit can be deployed in locations such as substation monitoring centers and operations and maintenance department offices, meeting the needs of diverse users and enhancing the usability and practicality of the monitoring system.

[0025] Preferably, the fusion model constructed in the cloud-based deep reinforcement learning and Transformer fusion processing unit uses the following formula for data processing: in, Represents the output data after processing by the fusion model; represents the action-value function output by the deep reinforcement learning module, The current state, including the pressure of SF6 gas ,temperature ,density ,humidity , trace water content , gas leakage rate The state space composed of parameters such as For the actions taken, are the parameters of the deep reinforcement learning model; Transformer It is the Transformer processing module; It is a function for characteristic transformation of various parameters of SF6 gas.

[0026] Specifically, the deep reinforcement learning and Transformer fusion model has been optimized for the characteristics of SF6 gas monitoring. The deep reinforcement learning module adopts a hierarchical state space design to integrate pressure (0.3-0.7MPa), temperature (-20℃-60℃), density (5.7-6.2kg / m³), humidity (0-1000μL / L), trace water content (0-500μL / L), leakage rate The model takes 6 parameters as input state and outputs the optimal action strategy through a 12-layer neural network (8-layer actor + 4-layer critic). The Transformer module uses a 6-encoder-decoder stack, each layer contains 8 heads of multi-head attention mechanism to extract features from time series data. In the implementation of a 500kV substation, the model can detect small leakage. The system's detection accuracy reached 98.3%, a 15.7 percentage point improvement over traditional algorithms, and the false alarm rate dropped from 12% to 3%. The system processes approximately 1.2 million pieces of monitoring data daily, with an average response time of less than 1.5 seconds, meeting real-time monitoring needs.

[0027] Preferably, the cloud-based deep reinforcement learning and Transformer fusion processing unit further includes the following formula for feature optimization: in, For the Optimized weights of individual data features; is the association weight between data; is the number of data samples; is the activation function; For SF6 gas A function that uses a set of parameters to perform feature extraction and transformation.

[0028] Specifically, the feature optimization mechanism improves data characterization capabilities through three-level weight adjustment. In the time dimension, an exponential decay window is used to assign a weight of 0.8 to the data of the last hour, a weight of 0.15 to the data of the past 24 hours, and a weight of 0.05 to the historical data. In the parameter association dimension, the Granger causality test is used to determine that the correlation weight between pressure and density is 0.85, and the correlation weight between temperature and micro-water content is 0.72. In the anomaly detection dimension, when the parameter fluctuation exceeds 3σ, the feature weight of the parameter is automatically increased by 50%. In the application of a 220kV substation, this mechanism increased the early leakage detection rate from 78% to 94%, and improved the robustness to humidity interference by 23%. The system has been in continuous operation for 18 months, successfully warning of equipment anomalies 31 times, avoiding 6 possible equipment failures, and significantly improving the reliability of the monitoring system.

[0029] Preferably, the decision analysis unit adopts the following decision formula based on deep reinforcement learning and Transformer model: in, Indicates the decision result; is the indicator function; is the number of decision data samples; For the The weight of individual data in decision making; is the decision threshold; For SF6 gas A function that takes group parameters for feature processing.

[0030] Specifically, the decision analysis unit constructed a four-layer decision tree architecture. The first layer is for basic threshold judgment, setting six thresholds, including a pressure lower limit of 0.35 MPa and a density lower limit of 5.7 kg / m³. The second layer is for rate of change judgment, calculating the rate of change over 1 hour (threshold 0.01 MPa / h), 6 hours (threshold 0.03 MPa / 6h), and 24 hours (threshold 0.05 MPa / 24h). The third layer is for parameter correlation judgment, analyzing parameter correlation using the Pearson coefficient matrix (threshold 0.7). The fourth layer is for trend prediction judgment, predicting parameter changes over the next 24 hours based on the ARIMA model. In a test at a 110kV substation, this mechanism extended the advance warning time for equipment failures from an average of 8 hours to 42 hours, with a warning accuracy rate of 96.8%. The system processed 1,273 alarm messages throughout the year, of which 1,215 were valid, with a false alarm rate of only 4.5%, significantly reducing ineffective work for operations and maintenance personnel.

[0031] Preferably, there is a data collaborative optimization mechanism between the edge computing preprocessing unit and the cloud-based deep reinforcement learning and Transformer fusion processing unit, and the formula is as follows: in, Represents the data after collaborative optimization; Data processed by the edge computing pre-processing unit; is the fusion coefficient; They are the state, action, and parameters of deep reinforcement learning in edge computing scenarios; This is a function for performing feature transformation on SF6 gas parameters collected by edge computing.

[0032] Specifically, the data collaborative optimization mechanism realizes three-level linkage processing between the edge and the cloud. The edge uses the LZ77 algorithm to compress the pre-processed data with a compression ratio of 4:1; after extracting the 2048-dimensional feature vector, it is reduced to 512 dimensions through PCA; the cloud further refines it to a 128-dimensional feature vector. When the edge detects that the KL divergence of the data distribution is greater than 0.2, it triggers an incremental update of the cloud model. In a 500kV substation cluster application, this mechanism reduces data transmission by 75% and increases the utilization of cloud computing resources by 60%. The model's adaptation period to seasonal environmental changes is shortened from 7 days in traditional methods to 24 hours, significantly improving the adaptability and efficiency of the system.

[0033] Preferably, the sensor data acquisition frequency of the multimodal perception unit is correlated with the model update of the cloud-based deep reinforcement learning and Transformer fusion processing unit, and the formula is: in, Update frequency for cloud models; is the adjustment coefficient; is the number of sensors; is the acquisition frequency of the sensth sensor; is the minimum acquisition frequency threshold; are the states, actions, and parameters of deep reinforcement learning related to the acquisition frequency, respectively.

[0034] Specifically, the sensor data acquisition frequency adaptive mechanism is based on a three-layer feedback control architecture. The basic frequency layer sets the acquisition frequency of the main transformer air chamber at 10Hz and the busbar air chamber at 5Hz according to the importance of the equipment; the change rate response layer increases the frequency by 2 times when the parameter change rate exceeds the threshold (pressure 0.005MPa / h); and the abnormal response layer increases the frequency to 50Hz when an abnormal mode is detected. A PID controller is used for dynamic adjustment, with a proportional coefficient of 0.5, an integral coefficient of 0.2, and a differential coefficient of 0.1. In the operation of a certain 220kV substation, this mechanism reduced the data transmission volume by 63% and extended the sensor life by 40%. The response time to sudden leaks was shortened from 30 minutes with traditional methods to 2 minutes, greatly improving the response speed of the system.

[0035] Preferably, there is an adaptive adjustment relationship between the data transmission rate of the IoT communication transmission unit and the computing load of the cloud-based deep reinforcement learning and Transformer fusion processing unit, and the formula is as follows: in, is the data transmission rate; is the adjustment factor; Calculate the load for the current cloud; Calculate the load for average; are the states, actions, and parameters of deep reinforcement learning related to transmission, respectively; It is a function for performing feature processing on the SF6 gas parameters involved in the transmission data.

[0036] Preferably, there is a dynamic mapping relationship between the visual interface layout of the interactive display unit and the decision result of the decision analysis unit, and the formula is: in, It is the layout parameter of the visual interface; is the layout adjustment coefficient; is the number of decision results; For the The weight of each decision outcome; are the states, actions, and parameters of deep reinforcement learning related to decision outcomes; It is a function for performing characteristic processing on SF6 gas parameters related to the decision result.

[0037] Specifically, the dynamic mapping mechanism of the visual interface implements three-layer adaptive adjustment. The data dimension mapping layer adjusts the chart type (line chart, bar chart, heat map) according to the complexity of the decision result; the user role mapping layer provides a real-time data monitoring interface for operation and maintenance personnel, a trend analysis interface for management personnel, and a parameter correlation analysis interface for technical experts; the interactive depth mapping layer dynamically adjusts the function display according to user operation habits. The interface response time is less than 100ms and supports smooth rendering of 100,000 data points. In an application in a provincial power grid monitoring center, the time for operation and maintenance personnel to obtain key information was shortened from an average of 2 minutes to 15 seconds, and the operation error rate was reduced by 40%, which significantly improved the monitoring efficiency.

[0038] Preferably, the system further includes an abnormal data tracing module, which uses the following formula to locate abnormal data: in, is the location index of abnormal data; is the mean vector of normal data; For the Function for performing characteristic processing on a group of SF6 gas parameters.

[0039] Specifically, the abnormal data tracing mechanism constructs a four-dimensional positioning system. The time dimension uses a sliding time window (1 minute-24 hours) to locate the abnormal period; the spatial dimension determines the physical location through the sensor topology structure; the parameter dimension analyzes the correlation between each parameter to find the key parameters; the historical dimension compares the historical abnormal pattern to determine the type. The improved DBSCAN algorithm is used, with a neighborhood radius of ε=0.1 and a minimum number of points MinPts=5. In a 500kV substation application, this mechanism can accurately locate the anomaly within 3 minutes, with a positioning accuracy of 99.5%. The tracing results of 17 abnormal events throughout 2023 show that the accuracy of locating the root cause of the fault reached 100%, providing strong support for fault diagnosis and equipment maintenance.

[0040] like Figure 2 As shown, an SF6 gas monitoring data online processing system includes the following steps: In the first step, the multimodal sensing unit periodically collects parameters such as pressure, temperature, density, humidity, moisture content, and gas leakage rate of SF6 gas in the substation in real time according to the set sampling strategy, and converts the collected analog signals into digital signals. The second step is to transmit the digital signal collected by the multimodal sensing unit to the edge computing preprocessing unit, which performs noise reduction, filtering, and data format standardization on the digital signal, and frames the data according to preset rules. In the third step, the framed data processed by the edge computing pre-processing unit is encapsulated by the IoT communication transmission unit using a specific communication protocol and stably transmitted to the cloud processing unit via a wireless or wired network. In the fourth step, the cloud-based deep reinforcement learning and Transformer fusion processing unit in the cloud processing unit receives the transmitted data and uses the constructed deep reinforcement learning and Transformer combined model architecture to perform feature extraction, time series analysis, and complex relationship mining on the data; In the fifth step, the decision analysis unit receives the data analysis results from the cloud-based deep reinforcement learning and Transformer fusion processing unit, and conducts a comprehensive assessment and decision on the status of the substation's SF6 gas based on pre-set rules and thresholds. In the sixth step, the interactive display unit presents the evaluation and decision results of the decision analysis unit in the form of visual charts, graphs and data lists, and provides an interactive operation interface so that users can query, analyze and export the monitoring data.

[0041] An SF6 gas monitoring data online processing system, with its innovative architecture design and integration of cutting-edge technologies, demonstrates significant advantages in many aspects.

[0042] To address the low efficiency and poor data accuracy of traditional manual inspections, the system uses a collaborative working mode between a multimodal sensing unit and an edge computing preprocessing unit. The multimodal sensing unit integrates high-precision sensors that can perform real-time, high-frequency acquisition of SF6 gas parameters such as pressure, temperature, and density. The acquisition frequency far exceeds the manual inspection cycle, enabling uninterrupted monitoring of the gas status. The edge computing preprocessing unit performs noise reduction, filtering, and format standardization on the raw data to eliminate environmental interference and human errors, ensuring data accuracy and consistency. Compared to manual inspections, this system not only greatly improves the timeliness of data acquisition, but also significantly reduces the data error rate, effectively avoiding monitoring deviations caused by non-standard manual operations.

[0043] To address the shortcomings of existing online monitoring systems, such as insufficient data processing capabilities and weak adaptive adjustment capabilities, the system achieves breakthroughs through cloud-based deep reinforcement learning and Transformer fusion processing units, as well as a dynamic coordination mechanism between multiple units. Utilizing deep reinforcement learning and the Transformer model, the fusion processing unit can deeply explore the complex relationships and changing trends between various SF6 gas parameters, accurately identifying minor leaks, abnormal parameter fluctuations, and other situations. Compared with traditional threshold judgment methods, the accuracy of anomaly detection is greatly improved. At the same time, the system has adaptive adjustment capabilities. The IoT communication transmission unit can dynamically adjust the transmission rate based on network status and data priority. The sensor data acquisition frequency can also be automatically adjusted based on parameter changes, and data collaborative optimization is achieved between edge computing and cloud processing units. This dynamic coordination mechanism enables the system to ensure efficient data transmission and processing in a complex and changing operating environment, while also rationally allocating computing resources to avoid resource waste, significantly improving the stability and adaptability of the system.

[0044] Furthermore, the system's decision analysis unit and interactive display unit further enhance its application value. The decision analysis unit, based on multi-layered decision rules and thresholds, scientifically assesses gas conditions and makes precise decisions, effectively improving the lead time for fault warnings. The interactive display unit, with its intuitive visual interface and rich interactive features, allows operators to quickly access critical information, improving O&M efficiency. The entire system provides a comprehensive, efficient, and intelligent solution for substation SF6 gas monitoring, effectively ensuring the safe and stable operation of substation equipment.

[0045] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to mechanical connections or electrical connections; they may refer to direct connections or indirect connections through an intermediate medium; and they may refer to internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0046] Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An SF6 gas monitoring data online processing system, characterized in that: include: A multimodal sensing unit integrates multiple types of high-precision sensors to sense and collect the pressure, temperature, density, humidity, moisture content, and gas leakage rate of SF6 gas in the substation in real time, and converts the collected analog signals into digital signals. An edge computing preprocessing unit, which is connected to the multimodal sensing unit, receives the digital signal transmitted by the multimodal sensing unit, performs preliminary noise reduction, filtering and data format standardization on the digital signal, and performs frame processing on the data; The IoT communication transmission unit is connected to the edge computing pre-processing unit and the cloud processing unit respectively. It uses multiple communication protocols to encapsulate the framed data processed by the edge computing pre-processing unit and transmits the data stably and reliably to the cloud processing unit via a wireless network or a wired network. The cloud-based deep reinforcement learning and Transformer fusion processing unit receives data transmitted by the IoT communication transmission unit and builds a model architecture that combines deep reinforcement learning and Transformer. This architecture is used to perform feature extraction, time series analysis, and complex relationship mining on the data. The decision analysis unit is connected to the cloud-based deep reinforcement learning and Transformer fusion processing unit, receives the processed data analysis results, and conducts a comprehensive assessment and decision on the status of the substation's SF6 gas based on pre-set rules and thresholds; The interactive display unit is connected to the decision analysis unit to display the evaluation and decision results of the decision analysis unit in the form of visual charts, graphs and data lists, and supports users to query, analyze and export monitoring data through interactive operations.

2. The SF6 gas monitoring data online processing system according to claim 1, characterized in that: The fusion model constructed in the cloud-based deep reinforcement learning and Transformer fusion processing unit uses the following formula for data processing: in, Represents the output data after processing by the fusion model; represents the action-value function output by the deep reinforcement learning module, The current state, including the pressure of SF6 gas ,temperature ,density ,humidity , trace water content , gas leakage rate The state space composed of parameters, For the actions taken, are the parameters of the deep reinforcement learning model; Transformer It is the Transformer processing module; It is a function for characteristic transformation of various parameters of SF6 gas.

3. The SF6 gas monitoring data online processing system according to claim 1, characterized in that: The cloud-based deep reinforcement learning and Transformer fusion processing unit also includes the following formula for feature optimization: in, For the Optimized weights of individual data features; is the association weight between data; is the number of data samples; is the activation function; For SF6 gas A function that uses a set of parameters to perform feature extraction and transformation.

4. The SF6 gas monitoring data online processing system according to claim 1, characterized in that: The decision analysis unit adopts the following decision formula based on deep reinforcement learning and Transformer model: in, Indicates the decision result; is the indicator function; is the number of decision data samples; For the The weight of individual data in decision making; is the decision threshold; For SF6 gas A function that takes group parameters for feature processing.

5. The SF6 gas monitoring data online processing system according to claim 1, characterized in that: There is also a data collaborative optimization mechanism between the edge computing preprocessing unit and the cloud-based deep reinforcement learning and Transformer fusion processing unit. The formula is as follows: in, Represents the data after collaborative optimization; Data processed by the edge computing pre-processing unit; is the fusion coefficient; They are the state, action, and parameters of deep reinforcement learning in edge computing scenarios; This is a function for performing feature transformation on SF6 gas parameters collected by edge computing.

6. The SF6 gas monitoring data online processing system according to claim 1, characterized in that: The sensor data collection frequency of the multimodal perception unit is correlated with the model update of the cloud-based deep reinforcement learning and Transformer fusion processing unit, and the formula is: in, Update frequency for cloud models; is the adjustment coefficient; is the number of sensors; is the acquisition frequency of the sensth sensor; is the minimum acquisition frequency threshold; are the states, actions, and parameters of deep reinforcement learning related to the acquisition frequency, respectively.

7. The SF6 gas monitoring data online processing system according to claim 1, characterized in that: The data transmission rate of the IoT communication transmission unit is adaptively adjusted to the computing load of the cloud-based deep reinforcement learning and Transformer fusion processing unit, as shown in the following formula: in, is the data transmission rate; is the adjustment factor; Calculate the load for the current cloud; Calculate the load for average; are the states, actions, and parameters of deep reinforcement learning related to transmission, respectively; It is a function for performing feature processing on the SF6 gas parameters involved in the transmission data.

8. The SF6 gas monitoring data online processing system according to claim 1, characterized in that: There is a dynamic mapping relationship between the visual interface layout of the interactive display unit and the decision result of the decision analysis unit, and the formula is: in, It is the layout parameter of the visual interface; is the layout adjustment coefficient; is the number of decision results; For the The weight of each decision outcome; are the states, actions, and parameters of deep reinforcement learning related to decision outcomes; It is a function for performing characteristic processing on SF6 gas parameters related to the decision result.

9. The SF6 gas monitoring data online processing system according to claim 1, characterized in that: The system also includes an abnormal data tracing module, which uses the following formula to locate abnormal data: in, is the location index of abnormal data; is the mean vector of normal data; For the Function for performing characteristic processing on a group of SF6 gas parameters.

10. An SF6 gas monitoring data online processing system according to any one of claims 1 to 9, characterized in that: The system operation includes the following steps: In the first step, the multimodal sensing unit periodically collects the pressure, temperature, density, humidity, moisture content, and gas leakage rate parameters of the SF6 gas in the substation in real time according to the set sampling strategy, and converts the collected analog signals into digital signals. The second step is to transmit the digital signal collected by the multimodal sensing unit to the edge computing preprocessing unit, which performs noise reduction, filtering, and data format standardization on the digital signal, and frames the data according to preset rules. In the third step, the framed data processed by the edge computing pre-processing unit is encapsulated by the IoT communication transmission unit using a specific communication protocol and stably transmitted to the cloud processing unit via a wireless or wired network. In the fourth step, the cloud-based deep reinforcement learning and Transformer fusion processing unit in the cloud processing unit receives the transmitted data and uses the constructed deep reinforcement learning and Transformer combined model architecture to perform feature extraction, time series analysis, and complex relationship mining on the data; In the fifth step, the decision analysis unit receives the data analysis results from the cloud-based deep reinforcement learning and Transformer fusion processing unit, and conducts a comprehensive assessment and decision on the status of the substation's SF6 gas based on pre-set rules and thresholds. In the sixth step, the interactive display unit presents the evaluation and decision results of the decision analysis unit in the form of visual charts, graphs and data lists, and provides an interactive operation interface so that users can query, analyze and export the monitoring data.

Images