Integrated low-power-consumption monitoring method for hydrogen, temperature and oil level in distribution transformer

Abstract

The invention relates to an integrated low-power-consumption monitoring method for hydrogen and temperature oil level in a distribution transformer, and belongs to the technical field of state monitoring and fault diagnosis of power equipment. According to the technical scheme, the monitoring terminal integrates a hydrogen sensing module, a temperature sensing module and an oil level sensing module, a deep dormancy and data aggregation reporting mechanism is adopted, the average power consumption is smaller than or equal to 30 mW, and battery power supply for more than 5 years can be achieved. The data analysis platform performs multi-source fusion and intelligent analysis on the received hydrogen concentration, oil temperature and oil level data; electric / thermal faults are distinguished by calculating the hydrogen change rate and associating the oil temperature trend; oil level data is introduced for concentration compensation; and judging by adopting a dynamic threshold value based on a historical baseline, and finally realizing graded early warning of early-stage insulation defects. According to the method, breakthrough balance is achieved in the four dimensions of monitoring cost, early warning accuracy, deployment convenience and diagnosis intelligence, and a feasible technical path is provided for real-time, reliable and general monitoring of the insulation state of massive distribution transformers.

Description

Technical Field

[0001] This invention relates to a low-power integrated monitoring method for hydrogen, temperature, and oil level inside a distribution transformer, belonging to the field of power equipment condition monitoring and fault diagnosis technology. Background Technology

[0002] Distribution transformers (hereinafter referred to as "transformers") are key equipment in power distribution networks for voltage transformation and power distribution. Their operational reliability directly affects the regional power supply quality and grid security. The insulation system of distribution transformers typically uses an oil-paper composite structure, which is subjected to multiple stresses—electrical, thermal, and mechanical—during long-term operation. This leads to gradual aging of the insulation material and may cause latent defects such as partial discharge and overheating. In their early stages, these defects produce characteristic gases (mainly hydrogen, as well as methane, acetylene, and ethylene) that dissolve in the insulating oil. Therefore, effective monitoring of characteristic gases in the insulating oil, especially hydrogen, which is the earliest and most sensitive to be produced, is a core means of diagnosing early insulation degradation and preventing faults.

[0003] Currently, monitoring the insulation condition of distribution transformers mainly relies on two technical approaches: offline testing and online monitoring. However, both approaches have significant bottlenecks when dealing with the large-scale, universal monitoring needs of numerous distribution transformers.

[0004] (1) Inherent limitations of offline experiments

[0005] Offline testing methods are represented by periodic power outage maintenance and laboratory oil chromatography (DGA). Although laboratory DGA is considered the standard for insulation fault diagnosis, it is essentially a periodic, offline sampling method. Its limitations include: (a) large monitoring blind spots: typically measured in years or several years, it cannot capture sudden or rapidly developing insulation defects between tests; (b) high economic and maintenance costs: power outages affect power supply, and the sampling and testing process is complex, requiring significant manpower and resources; (c) poor data timeliness: the time from sampling to obtaining results is long, making it impossible to support real-time status assessment and early warning. This method can no longer meet the requirements of modern smart distribution networks for real-time equipment status perception and proactive early warning.

[0006] (2) The main shortcomings of existing online monitoring technologies

[0007] To overcome the shortcomings of offline testing, online monitoring technology has emerged. However, its current development level still makes it difficult to apply on a large scale in the distribution transformer field, mainly due to the following problems:

[0008] The detection is performed using a full-component online chromatographic monitoring instrument; the laboratory DGA process is brought online, enabling continuous monitoring of multiple characteristic gases. However, to achieve chromatographic separation, it is necessary to integrate complex modules such as gas path system, carrier gas source, and detector, resulting in a large equipment size, high cost, high power consumption, and the need for regular carrier gas replenishment and professional maintenance.

[0009] Low-cost single hydrogen monitoring devices are used for detection. Given that hydrogen is the earliest and most sensitive characteristic gas in the formation of most insulation defects, various single hydrogen online monitoring products based on electrochemical and semiconductor principles have emerged on the market, significantly reducing costs. However, these devices generally suffer from the following problems: (a) High false alarm rate: The hydrogen production rate is significantly positively correlated with the insulating oil temperature. Existing devices mostly use simple fixed concentration thresholds for alarms, making it difficult to effectively distinguish between "abnormal hydrogen production caused by insulation faults" and "changes in hydrogen production rate caused by normal fluctuations in oil temperature." This leads to frequent false alarms under specific operating conditions, resulting in poor reliability. (b) Risk of missed alarms: When the transformer oil volume decreases, the equilibrium state of dissolved gases in the oil changes, which may cause the measured hydrogen concentration value to not accurately reflect the gas production per unit oil volume, potentially masking the true gas production trend and causing missed or delayed alarms. (c) Lack of diagnostic function: It provides real-time data or over-limit signals of hydrogen concentration in one dimension, but lacks the ability to mine data trends and the ability to correlate with other state parameters such as oil temperature. It cannot make a preliminary distinction between the nature of defects (such as electrical faults and thermal faults), and cannot provide forward-looking early warnings with preventive significance. The level of intelligence is low. (3) Common problem: obstacles to terminal deployment.

[0010] In addition to the aforementioned functional deficiencies, existing online monitoring terminals face common challenges in engineering deployment: power consumption and power supply issues. Most terminals consume significant power, relying on power from the transformer itself or nearby. This poses a significant challenge for outdoor pole-mounted transformers without pre-installed power supplies, as well as for older, cramped distribution rooms with difficult wiring, making installation difficult, costly, or even impossible, thus severely limiting their application scope. Summary of the Invention

[0011] The purpose of this invention is to provide a low-power integrated monitoring method for hydrogen, temperature, and oil level inside distribution transformers. It achieves a breakthrough balance in four dimensions: monitoring cost, early warning accuracy, deployment convenience, and intelligent diagnosis. It provides a feasible technical path for real-time, reliable, and universal monitoring of the insulation status of massive distribution transformers, realizes early warning of distribution transformer insulation status in a cost-effective, highly reliable, and easy-to-deploy manner, and solves the aforementioned technical problems of existing technologies.

[0012] The technical solution of this invention is:

[0013] A low-power integrated monitoring method for hydrogen, temperature, and oil level inside a distribution transformer is proposed. This method is based on a monitoring terminal and a remote data analysis platform. The monitoring terminal, deployed in the distribution transformer's oil circuit, sends the monitored parameters to the remote data analysis platform for processing. The method includes the following steps:

[0014] S1. Multi-source data synchronous acquisition: The monitoring terminal synchronously acquires the dissolved hydrogen concentration H2, top oil temperature T and oil level L in the transformer insulating oil in the first sampling period T1 to obtain the original data sequence;

[0015] S2. Local preprocessing and caching: The original data sequence is digitally filtered to eliminate transient interference, and the processed valid data is stored in local non-volatile memory;

[0016] S3. Low-power aggregation upload: The monitoring terminal wakes up at a timed second upload cycle T2, where T2 > T1, and packages multiple sets of valid data stored in the memory within the most recent T2 cycle, and uploads them to the remote data analysis platform through the low-power wide area network wireless communication module.

[0017] S4. Cloud-based multi-parameter fusion analysis: The remote data analysis platform performs the following analysis steps:

[0018] S41. Trend feature extraction: Calculate the first rate of change dH2 / dt of the hydrogen concentration value H2 within a preset time window, and the second rate of change dT / dt of the oil temperature value T within the same time window;

[0019] S42. Hydrogen-Temperature Correlation Judgment: Establish a correlation model. When the conditions are met: dH2 / dt > δ1 and |dT / dt| < δ2 for a continuous period of time, it is determined that the early insulation defect tends to be an electrical fault, and a first-level warning sign is generated; where δ1 is the positive threshold of hydrogen change rate and δ2 is the stability threshold of oil temperature change rate.

[0020] S43. Dynamic oil level compensation: Based on the received oil level height value L, the hydrogen concentration value H2 is corrected using the pre-stored oil level-concentration compensation function F(L) to obtain the corrected hydrogen concentration value H2' = H2 × F(L), where F(L) > 1 when L is lower than the rated oil level.

[0021] S5. Dynamic threshold-based early warning: Based on the historical hydrogen concentration baseline data of the transformer, the current oil temperature value T, and the first-level early warning identifier, a dynamic early warning threshold Y is generated; the corrected hydrogen concentration value H2' or its changing trend is compared with the dynamic early warning threshold Y, and if it exceeds the threshold, a second-level early warning information containing the defect type tendency is generated and output.

[0022] Furthermore, the monitoring terminal operates in an intermittent deep sleep mode, with a static power consumption of ≤1mW in deep sleep state and an average power consumption of ≤30mW during the active working cycle including data acquisition, processing, and communication, so as to achieve a battery life of more than 5 years for the built-in battery.

[0023] Furthermore, the hydrogen temperature correlation determination in step S4 further includes: when dH2 / dt > δ1 and dT / dt > δ3 (δ3 is the oil temperature rise threshold), the cause of hydrogen production is determined to be a thermal fault correlation; when only dH2 / dt > δ1 and |dT / dt| < δ2, the conclusion of electrical fault determination is strengthened; the thresholds δ1, δ2, and δ3 are adaptively adjusted according to the transformer model, operating history, and seasonal characteristics of ambient temperature.

[0024] Furthermore, the method for generating the dynamic early warning threshold Y in step S5 is specifically as follows:

[0025] Y = μ + kσ + f(T) + g(Flag)

[0026] Where μ and σ are the mean and standard deviation of the historical hydrogen concentration data of the transformer under similar oil temperature conditions, respectively; k is the sensitivity coefficient; f(T) is the oil temperature correction term, reflecting the influence of temperature on the background of gas production; g(Flag) is the warning level adjustment term, and when the first-level warning flag exists, g(Flag) > 0.

[0027] Furthermore, step S5 is followed by step S6: early warning visualization and report generation: the remote data analysis platform integrates the second-level early warning information, related trend curves, correlation analysis conclusions and maintenance suggestions to generate a structured early warning report, and pushes it to the user terminal through a graphical interface and preset communication channels.

[0028] A monitoring terminal for implementing the above-mentioned low-power integrated monitoring method for hydrogen gas, temperature, and oil level inside a distribution transformer includes: a fully metal shielded sealed housing, an integrated sensing module, a main control and data processing module, a low-power wireless communication module, and a power management module; the fully metal shielded sealed housing constitutes the main body for electromagnetic shielding and environmental protection, and one end of it is provided with a mechanical interface adapted to the transformer oil valve to achieve plug-and-play functionality and oil circuit sealing; the integrated sensing module is built into the fully metal shielded sealed housing and includes:

[0029] (1) Hydrogen sensing unit, used to detect the concentration of dissolved hydrogen in oil;

[0030] (2) Temperature sensing unit, used to detect oil temperature;

[0031] (3) Oil level detection unit, used to detect oil level height;

[0032] The main control and data processing module is electrically connected to the integrated sensing module and is used to control the sampling timing, perform data preprocessing and caching; the low-power wireless communication module is used to interact with an external gateway; the power management module is connected to a high-energy-density battery and is used to provide stable power supply to each functional module and implement precise power consumption state management; wherein, the main control and data processing module is configured to execute steps S1 to S3 of the above method.

[0033] The hydrogen sensing unit of the integrated sensing module is an electrochemical or thermal conductivity type hydrogen sensor based on microelectromechanical systems (MEMS) technology; the protection level of the all-metal shielded sealed housing is not lower than IP67; the mechanical interface is a standard threaded or flanged structure.

[0034] The high-energy-density battery is a lithium-thionyl chloride battery.

[0035] The power management module includes a high-efficiency DC-DC conversion circuit and a load switch array. The main control and data processing module controls the on / off state of the load switch array to completely power off the low-power wireless communication module and the integrated sensing module during non-working periods, maintaining only the real-time clock and wake-up circuit.

[0036] A remote data analysis platform for implementing the above-mentioned low-power integrated monitoring method for hydrogen, temperature, and oil level inside a distribution transformer includes a network transmission layer and a data analysis platform. The network transmission layer is used to aggregate data from the monitoring terminal. The data analysis platform is used to receive, store, and process data from the network transmission layer, and to execute steps S4 to S5 in the above method to implement intelligent early warning of early insulation defects in distribution transformers using single hydrogen detection.

[0037] The data analysis platform also includes a machine learning model, which uses historical multidimensional data sequences and corresponding fault records as a training set to perform in-depth analysis of the input real-time data features and output insulation health status scores and defect probability predictions to optimize the decision-making of graded early warning and improve the accuracy of early warning.

[0038] This invention constructs a complete technical solution through a three-layer collaborative architecture of low-power multi-parameter monitoring terminal + low-power wide area network transmission + cloud-based intelligent analysis platform, which enables accurate data acquisition, low-power reliable transmission, multi-parameter intelligent diagnosis, and hierarchical early warning output. It solves the four core defects of existing technologies and provides a low-cost, low-power, highly reliable, easy-to-deploy, and intelligent complete solution for early warning of distribution transformer insulation status.

[0039] Key points of the invention concept:

[0040] 1. Based on single hydrogen detection, it integrates multi-parameter analysis of oil temperature and oil level;

[0041] 2. Ultra-low power terminal design with deep sleep mode and data aggregation reporting;

[0042] 3. A hydrogen-temperature correlation discrimination model to distinguish between electrical and thermal faults;

[0043] 4. Dynamic oil level compensation algorithm to correct concentration measurement distortion;

[0044] 5. Personalized dynamic threshold early warning based on historical baselines;

[0045] 6. Plug and play all-metal sealed terminal structure.

[0046] The beneficial effects of this invention are as follows: (1) A revolutionary reduction in monitoring costs and scalability: By focusing on the most effective single hydrogen detection, using commercial micro sensors and minimal circuit design, the terminal hardware cost is reduced by one to two orders of magnitude compared to online chromatographs. Ultra-low power consumption and battery power supply completely avoid power costs, making it possible to monitor a large number of distribution transformers across the network, especially pole-mounted transformers. (2) Significantly improves the accuracy and reliability of early warning: The false alarm rate is greatly reduced: The "hydrogen temperature correlation discrimination" model effectively eliminates the hydrogen production interference caused by temperature changes, allowing the alarm to focus on "abnormal independent hydrogen production", fundamentally solving the industry problem of false alarms in summer. Effectively avoids missed alarms: The "oil level dynamic compensation" model corrects the measurement deviation caused by changes in oil volume, ensuring the authenticity of the data and avoiding early warning delays or failures caused by equipment conditions (such as oil leakage). Overcomes the terminal deployment problem: The "plug-and-play" mechanical interface and IP67 fully sealed metal shell design eliminate the need for professional wiring. With an average power consumption of ≤30mW and a high-capacity battery, it ensures maintenance-free operation for more than 5 years, greatly reducing the total life cycle maintenance cost. (3) It realizes the intelligent diagnosis process and early warning: Intelligent diagnosis: The system output is no longer raw data, but a structured early warning report that integrates trend analysis, preliminary judgment of fault type (electrical / thermal) and oil level compensation conclusion, providing direct decision support. Early warning: Based on the rate of change analysis and personalized dynamic threshold, the system can identify the dangerous upward trend before the absolute value of hydrogen reaches the traditional fixed threshold, truly realizing early warning.

[0047] This invention achieves an excellent balance in four dimensions: cost, accuracy, ease of use, and intelligence, providing a complete and efficient technical solution for building a new generation of online monitoring system for the insulation status of distribution transformers. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the overall architecture of an embodiment of the present invention;

[0049] Figure 2This is a hardware system block diagram of the monitoring terminal according to an embodiment of the present invention;

[0050] Figure 3 This is a flowchart illustrating the core data processing and early warning logic of an embodiment of the present invention;

[0051] Figure 4 This is a schematic diagram illustrating the correlation analysis between hydrogen concentration and faults in an embodiment of the present invention;

[0052] Figure 5 This is a schematic diagram illustrating the correlation analysis between temperature and fault in an embodiment of the present invention. Detailed Implementation

[0053] The invention will be further illustrated below with reference to the accompanying drawings and examples.

[0054] See attached document Figure 1-5 A low-power integrated monitoring method for hydrogen, temperature, and oil level inside a distribution transformer is proposed. This method is based on a monitoring terminal and a remote data analysis platform. The monitoring terminal, deployed in the distribution transformer's oil circuit, sends the monitored parameters to the remote data analysis platform for processing. The method includes the following steps:

[0055] (1) Multi-source data synchronous sensing for early defects

[0056] The monitoring terminal is configured to synchronously acquire three key state variables within the transformer tank using its internally integrated sensors during a highly consistent first sampling period (T1):

[0057] Dissolved hydrogen concentration in oil (H2): Measured directly using a sensor with high selectivity for hydrogen.

[0058] Top layer oil temperature T: Measured at a position close to thermal equilibrium, reflecting the thermal state of the insulation system;

[0059] Oil level height value L: represents the volume of insulating oil in the oil tank;

[0060] Synchronization refers to completing the sampling of three parameters within the same sampling trigger time, ensuring strict alignment of data points on the time axis, providing a time-consistent data foundation for subsequent correlation analysis, and avoiding analysis errors caused by asynchronous sampling.

[0061] (2) Localized preprocessing and low-power data management

[0062] The controller of the monitoring terminal performs real-time preprocessing on the collected raw data sequences, mainly including:

[0063] Digital filtering: Using moving average filtering, median filtering, or a combination thereof, instantaneous pulse interference and random noise caused by transient electromagnetic processes in the field are filtered out, thereby improving the reliability of single sample values;

[0064] Validity verification: Set a reasonable range based on physical probability, and mark or remove data that exceeds the range;

[0065] Data caching: Processed, valid data is stored locally in non-volatile memory. This design moves the raw data cleaning task forward, reducing the burden on the cloud and ensuring data quality.

[0066] (3) Extremely energy-efficient communication based on aggregation reporting

[0067] To achieve ultra-low power consumption and long lifespan, the monitoring terminal adopts a communication strategy that prioritizes deep sleep, utilizes timed wake-up, and aggregates and reports data. Specifically, this includes:

[0068] Working status management: The terminal is in deep sleep mode most of the time, at which time only the real-time clock (RTC) and power management unit are maintained to maintain basic operation, and the overall static power consumption is ≤1mW;

[0069] Periodic wake-up and data acquisition: The RTC wakes up the main control and sensing modules at a timer of period T1, executes steps S1 and S2, and then quickly returns to deep sleep after completion;

[0070] Data aggregation and uploading: The RTC triggers a complete work cycle with a second uploading period T2 that is significantly longer than T1. In this cycle, the terminal first reads all valid data packets stored in the previous T2 period (number N = T2 / T1) from the memory, and then packages these N sets of data, terminal device ID, battery voltage status, and other information into an aggregated data frame for aggregation.

[0071] Low-power wireless transmission: Activating low-power wide-area network (LPWAN) communication modules such as LoRa, the system transmits aggregated data frames to a remote gateway at low power. After transmission, all modules immediately re-enter deep sleep mode. This mode minimizes high-power wireless radio frequency activity and is a core design feature enabling the terminal to achieve an average power consumption of ≤30mW and a battery life of over 5 years thanks to its built-in battery.

[0072] (4) Cloud-based multi-parameter fusion and intelligent feature extraction

[0073] After receiving the data, the remote data analysis platform starts the core analysis engine and performs the following deep fusion analysis:

[0074] (a) Dynamic Trend Feature Extraction: The remote data analysis platform not only monitors the instantaneous values ​​of parameters, but more importantly, extracts their dynamic trend features. For hydrogen concentration H2 and oil temperature T, their rates of change within a preset sliding time window are calculated, i.e., dH2 / dt and dT / dt. Simultaneously, the mean, standard deviation, and cumulative gas production within a certain period can be calculated. The rate of change is a more sensitive and leading indicator for identifying early abnormal trends and distinguishing them from normal fluctuations.

[0075] (b) Hydrogen Temperature-Related Fault Type Identification: This is the core of distinguishing the nature of the fault. The remote data analysis platform establishes a real-time correlation model and sets the following criteria: Electrical Fault Tendency Criterion: If, within multiple consecutive analysis cycles, dH2 / dt > δ1 and |dT / dt| < δ2, then the probability that hydrogen production originates from electrical faults such as partial discharge is significantly increased. Here, δ1 is the positive threshold for hydrogen change rate, and δ2 is the stability threshold for oil temperature change rate. This criterion accurately captures the early key characteristic of electrical faults: "hydrogen concentration independent of oil temperature anomalies and a continuous increase." When this condition is met, the remote data analysis platform generates a first-level warning flag.

[0076] If dH2 / dt > δ1 and dT / dt > δ3, then hydrogen production is likely to be associated with overheating faults or normal overloads. The thresholds δ1, δ2, and δ3 are not completely fixed and can be fine-tuned based on the transformer model, historical operating data baseline, and seasonal characteristics of ambient temperature to give the model adaptive capabilities.

[0077] (c) Dynamic Oil Level Compensation Model: To mitigate the risk of underreporting, the remote data analysis platform introduces an oil level compensation function F(L). The correction formula is: H2' = H2 × F(L). Here, F(L) is a piecewise or continuous function with the following basic characteristics: when the real-time oil level L is equal to or higher than the rated oil level L0, F(L) ≈ 1; when L < L0, F(L) > 1, and the value of F(L) increases as the oil level decreases. Through this correction, the corrected hydrogen concentration value H2' is obtained, which more accurately reflects the gas production concentration converted to the standard oil level, eliminating the interference of oil quantity changes.

[0078] (5) Hierarchical early warning based on dynamic baseline and risk bonus

[0079] The remote data analysis platform abandons a single fixed threshold and establishes a personalized dynamic early warning threshold Y for each transformer, generating the following model:

[0080] Y = μ + kσ + f(T) + g(Flag)

[0081] μ and σ are the mean and standard deviation of the historical hydrogen concentration data of the transformer under operating conditions similar to the current oil temperature, respectively; k is the sensitivity coefficient; f(T) is the oil temperature correction term, reflecting the influence of temperature on the background of gas production; g(Flag) is the early warning level adjustment term. When the first-level early warning flag exists, g(Flag) > 0, actively lowering the early warning threshold.

[0082] Finally, the remote data analysis platform compares the corrected hydrogen concentration value H2' or its trend with the dynamic threshold Y. If it exceeds the threshold, it triggers a second-level warning containing the tendency of defect types and pushes it through a structured report.

[0083] A low-power multi-parameter monitoring terminal for implementing the above method, which is key to solving deployment and power supply problems, includes:

[0084] Environmentally adaptable structure: Utilizing a fully metal shielded sealed housing (protection rating ≥ IP67), it combines electromagnetic shielding, mechanical protection, and heat dissipation. One end of the housing is designed with a mechanical interface compatible with standard transformer oil valves, enabling true plug-and-play functionality and reliable oil circuit sealing.

[0085] Highly integrated sensing unit: compactly integrated within a housing:

[0086] Hydrogen sensing unit: Employs an electrochemical or thermal conductivity miniature sensor based on MEMS technology, specifically designed for the detection of dissolved hydrogen in oil;

[0087] Temperature sensing unit: High-precision Pt100 platinum resistance thermometer, in direct contact with oil flow;

[0088] Oil level detection unit: Radio frequency capacitive or ultrasonic sensor;

[0089] Ultra-low power hardware system:

[0090] Main control module: Employs an ultra-low power microcontroller responsible for scheduling all tasks;

[0091] Precision Power Management Module: Connects to high-energy-density batteries such as lithium-thionyl chloride batteries. This module includes a high-efficiency DC-DC conversion circuit and a load switch array. The main controller can completely cut off the power supply to the sensor and communication module during non-data acquisition periods through the switch array, maintaining only the core timing circuit, thereby achieving an average operating power consumption of ≤30mW and a battery life of more than 5 years.

[0092] A remote data analysis platform for implementing the above-mentioned low-power integrated monitoring method for hydrogen, temperature, and oil level inside a distribution transformer includes:

[0093] (1) Sensing layer: It consists of low-power multi-parameter monitoring terminals, as described above, which are widely deployed in various distribution transformers;

[0094] (2) Network transmission layer: It consists of a low-power wide area network gateway (such as a LoRa gateway) and a backhaul channel, which is responsible for reliably and economically aggregating massive amounts of terminal data and transmitting it to the cloud;

[0095] (3) Platform Application Layer: This is a software platform deployed on a cloud server or private data center, including data receiving service, time series database, feature calculation engine, early warning rule engine, machine learning model library, data visualization module, and alarm push service. The machine learning model can be trained on historical data to further optimize dynamic threshold parameters and serve as an auxiliary decision-making basis for early warning.

[0096] Example 1: Complete system implementation for a typical 10kV outdoor pole-mounted transformer.

[0097] This embodiment uses an oil-immersed outdoor pole-mounted distribution transformer of model S13-M-400 / 10 as the monitoring object to fully demonstrate the implementation of the method, terminal and system of the present invention.

[0098] (1) Hardware implementation and deployment of monitoring terminals

[0099] Terminal structure: The housing is made of 316 stainless steel, with dimensions of Φ45mm×160mm and a protection rating of IP68. The front end is an M20×1.5 external thread interface, equipped with two fluororubber O-rings.

[0100] Sensing unit:

[0101] Hydrogen sensing unit: Employs a MEMS thermal conductivity hydrogen sensor with a range of 0-2000 ppm, an accuracy of ±50 ppm, and a response time of <30 seconds;

[0102] Temperature sensing unit: PT100 platinum resistance thermometer, encapsulated in a stainless steel sheath;

[0103] Oil level detection unit: adopts radio frequency capacitive sensor, outputs 4-20mA signal corresponding to 0-100% oil level.

[0104] Core circuit:

[0105] Main control MCU: STMicroelectronics' STM32L072 series ultra-low power microcontroller;

[0106] Communication module: Uses Semtech SX1276 LoRa chip, operating frequency band 470MHz, transmit power +14dBm;

[0107] Power management: Employs a TI BQ25504 high-efficiency energy harvester, paired with one ER34615 lithium-thionyl chloride battery (capacity 3.6V, 19Ah).

[0108] Deployment: With the transformer powered on, screw it directly into the low-voltage side sampling valve or dedicated monitoring valve to replace the original plug and ensure a seal. The terminal will automatically activate and enter working mode after installation.

[0109] (2)Terminal working process and parameter configuration

[0110] The terminal firmware is configured according to the above steps S1 to S3:

[0111] Data acquisition and caching: Set the first sampling period T1 = 5 minutes. After each wake-up, synchronously acquire H2, T, and L data, and store them in FRAM after median filtering. The average power consumption for single acquisition and processing is about 20 mW, and the time-consuming is about 10 seconds. Subsequently, the main control MCU controls all peripheral modules to power off and enters the deep sleep mode by itself. At this time, the static power consumption of the whole machine is less than 1 μA

[0112] Aggregate reporting: Set the second upload period T2 = 1 hour. After waking up every hour, pack the past 12 groups of data and send them to the gateway 500 meters away through LoRa. The power consumption during the sending process is about 120 mW and lasts for about 3 seconds;

[0113] Power consumption accounting: After accounting, in this working mode, the average power consumption of the terminal is much lower than the design target of 30 mW, and the theoretical battery life can exceed 5 years.

[0114] (3)Implementation of cloud platform analysis and early warning

[0115] The data analysis platform configures the algorithm according to the above steps S4 to S5 and S6:

[0116] Trend extraction: The platform calculates dH2 / dt and dT / dt within the past 6-hour window with a 1-hour step (using linear fitting to obtain the slope);

[0117] Association discrimination model: Set the thresholds δ1 = 2 ppm / h, δ2 = 0.5 ℃ / h, δ3 = 1 ℃ / h. If dH2 / dt > 2 and |dT / dt| < 0.5 are satisfied for 3 consecutive cycles (3 hours), it is determined as an electrical fault tendency, and the first-level early warning flag Flag = 1 is generated; the thresholds δ1, δ2, and δ3 can be obtained based on the statistical data of the historical operation of a large number of normal and faulty transformers, or can be initially set individually according to the transformer model, insulating oil characteristics, and seasonal laws of ambient temperature, and dynamically optimized through the self-learning algorithm of the data analysis platform during operation.

[0118] Oil level compensation: Set the rated oil level L0 = 80%. The compensation function is simplified as: F(L) = L0 / L (when L < L0); F(L) = 1 (when L ≥ L0). If L = 70% is measured, then F(L) = 1.14;

[0119] Dynamic threshold and early warning: The dynamic threshold model is Y = μ + 2σ + 0.05 (T - 40) + 0.1 μ Flag;

[0120] μ, σ: Calculated using historical hydrogen concentration data of the transformer oil temperature within the current T±2℃ range over the past 30 days;

[0121] f(T) = 0.05 (T-40): Oil temperature compensation item, 40℃ is the reference temperature;

[0122] g(Flag) = 0.1 μ Flag: Risk bonus, takes effect when Flag=1.

[0123] 4. Simulation Operation and Early Warning Process

[0124] Assume the transformer has a history of stable operation, with a baseline μ = 35 ppm and σ = 5 ppm (at an oil temperature of around 45°C);

[0125] Days 1-7 (normal condition): Data is stable, H2 fluctuates between 30-40 ppm, T varies with daily load between 40-50℃, and L remains stable at 82%. The platform calculates the dynamic threshold Y to be between 45-50 ppm, with no warnings.

[0126] Starting from day 8 (simulating early electrical failure): assume that hydrogen production begins continuously due to slight deterioration of internal insulation.

[0127] H2 increased from 40 ppm to 58 ppm (dH2 / dt ≈ 3 ppm / h), while T remained at 48℃ (dT / dt ≈ 0). Platform calculations yielded: Flag = 1 (electric tendency), corrected concentration H2' ≈ 58 ppm (L normal), and dynamic threshold Y = 48.9 ppm. Since 58 > 48.9, a level-two warning was triggered.

[0128] Warning Report Output: The platform generates a report including: Level 2 Warning: Transformer A (ID: ZS-001) at 14:00 on day X, the corrected hydrogen concentration is 58 ppm, exceeding the dynamic threshold (48.9 ppm). Characteristics: Hydrogen production rate is independent of oil temperature rise, showing an early electrical fault tendency. Recommendation: Conduct a comprehensive review using infrared thermography, ultrasonic partial discharge detection, or laboratory chromatographic analysis of oil samples.

[0129] Example 2: Targeting transformers in distribution rooms exhibiting a tendency for thermal failure.

[0130] This embodiment simulates a transformer overheating due to a loose connection, demonstrating how the present invention distinguishes thermal faults.

[0131] Scenario: An indoor oil-immersed transformer (model SCB10-630 / 10).

[0132] Simulated data: One afternoon, due to a surge in load and overheating of the joint, the oil temperature T rose from 65℃ to 85℃ in 4 hours (dT / dt=5 ℃ / h), and the hydrogen concentration H2 rose from 50ppm to 90ppm simultaneously (dH2 / dt=10 ppm / h).

[0133] Platform Analysis:

[0134] Association discrimination: Although dH2 / dt(10) > δ1 (2), dT / dt(5) > δ3 (1), the platform determines it as a thermal fault association and does not set an electrical fault flag (Flag=0).

[0135] Oil level compensation: Oil level is normal, H2' = H2;

[0136] Dynamic threshold calculation: Assuming historical baseline μ (65℃) = 45ppm, σ = 8ppm, Y = 63.25ppm;

[0137] Result: Since H2' (90ppm) > Y (63.25ppm), a level 2 warning will still be triggered. However, the warning report clearly concludes that the warning trigger is mainly synergistic with a significant increase in oil temperature, exhibiting characteristics of a thermal fault. It is recommended to focus on checking the load and cooling system. This is completely different from the conclusion regarding electrical faults.

[0138] Example 3: Verification of the compensation effect of a transformer with a drop in oil level

[0139] This embodiment verifies the necessity of the oil level compensation model.

[0140] Scenario: The transformer in Example 1, after running for a period of time, showed slight oil leakage, and the oil level L slowly dropped from 82% to 70%.

[0141] Comparison of two scenarios:

[0142] Scenario A: The platform does not use the compensation model (F(L)=1). When the measured H2 value increases from 35ppm to 50ppm (actual gas production increases), this value may be underestimated due to the "dilution" effect of the oil level drop. Assuming there is no compensation, the platform calculates a dynamic threshold of approximately 48ppm, which may not trigger an alert, resulting in a missed report.

[0143] Scenario B: The platform activates the compensation model. H2 is measured at 50 ppm, L at 70%, and H2' is calculated to be 57.1 ppm. Using H2' in the dynamic threshold comparison makes it more likely to exceed the threshold, triggering a timely warning. The warning report will indicate: Current oil level is low (70%), and hydrogen concentration has been compensated.

[0144] Table 1. Differences in effectiveness between the present invention and existing technologies.

[0145]

[0146] Comparative Example 1: Traditional fixed threshold single hydrogen monitoring device.

[0147] This comparative example is used to highlight the advantages of the present invention.

[0148] Device: A commercially available fixed-threshold hydrogen detector, with the alarm point set at the warning value of 150 ppm according to the DL / T 722 standard.

[0149] Scenario: Exactly the same as the early electrical failure scenario in Example 1.

[0150] Process: On the 8th day, when the invention had triggered the warning (H2=58ppm), the conventional device displayed a reading of 58ppm, which was far from the 150ppm alarm value, and there was no alarm.

[0151] Result: The transformer may continue to operate with the defect for weeks or even months until the hydrogen concentration slowly accumulates to more than 150 ppm before triggering an alarm, thus missing the best opportunity for early intervention.

[0152] Explanation of related terms

[0153] 1. Distribution transformer (distribution transformer): The core equipment in the power distribution network that realizes voltage transformation and power distribution, and is the monitoring object of this invention. Its insulation system usually adopts an oil-paper composite structure, which is susceptible to defects caused by electrical, thermal and mechanical stress during long-term operation.

[0154] 2. Insulation defects: Latent damage to the transformer insulation system (oil-paper composite structure) caused by factors such as aging, partial discharge, and overheating will produce characteristic gases that dissolve in the insulating oil in the early stage, which is the early warning target of this invention.

[0155] 3. Dissolved hydrogen (H2): The core characteristic gas generated in the early stage of transformer insulation defects. It is characterized by early generation time and high sensitivity. It is the core monitoring indicator of the single hydrogen detection scheme of this invention. It is dissolved in insulating oil.

[0156] 4. Single hydrogen detection: This method uses only hydrogen as the core characteristic gas, combined with auxiliary parameters such as temperature and oil level, to monitor the insulation status of distribution transformers. Unlike the detection of all components of gas, this is the core design of this invention to achieve low-cost monitoring.

[0157] 5. Low-power multi-parameter monitoring terminal: A field sensing device deployed in the distribution transformer oil circuit, integrating hydrogen, temperature, and oil level sensing modules as well as main control, communication, and power management modules. It adopts a deep sleep and data aggregation reporting mechanism and is the core hardware for data acquisition and low-power implementation of this invention.

[0158] 6. Remote Data Analysis Platform: A software system deployed in the cloud, responsible for receiving data uploaded by monitoring terminals, and realizing intelligent early warning through multi-parameter fusion analysis, fault identification, threshold calculation, etc. It is the core carrier of the intelligent diagnosis of this invention.

[0159] 7. Multi-source fusion analysis: The core technical idea of ​​this invention is to comprehensively process data from three different sources, namely hydrogen concentration, oil temperature and oil level, collected by the monitoring terminal, including trend extraction, correlation discrimination and compensation correction, in order to improve the accuracy of early warning.

[0160] 8. Hydrogen-Temperature Correlation Discrimination: By calculating the rate of change of hydrogen concentration (dH2 / dt) and the rate of change of oil temperature (dT / dt), an algorithm is established to distinguish fault types based on the correlation model. This algorithm can identify electrical faults and thermal fault tendencies, which is a key means of reducing false alarm rate in this invention.

[0161] 9. Dynamic oil level compensation: Based on the distribution transformer oil level height data, the process of correcting the hydrogen concentration measurement value through a preset compensation function is used to eliminate the concentration measurement distortion caused by the drop in oil level and avoid missed reports. This is the core design of this invention to improve monitoring reliability.

[0162] 10. Dynamic early warning threshold (Y): The early warning judgment standard is dynamically generated based on the historical hydrogen concentration baseline data of the distribution transformer, the current oil temperature and fault tendency indicators. Unlike the traditional fixed threshold, it can adapt to different operating conditions. The formula is Y=μ+kσ+f(T)+g(Flag).

[0163] 11. Deep Sleep: Monitors the terminal's low-power operating mode. In this mode, only the real-time clock and wake-up circuit are maintained. The static power consumption is ≤1mW, which is the core guarantee for the terminal to achieve more than 5 years of battery life.

[0164] 12. Data aggregation and reporting: The monitoring terminal integrates the effective data from multiple sampling periods into a single data frame and uploads it to the cloud by using short-cycle sampling data and long-cycle packaged upload. This reduces the number of high-power communication operations and lowers the overall power consumption.

[0165] 13. Low Power Wide Area Network (LPWAN): A wireless communication network suitable for IoT devices, characterized by low power consumption, long distance, and wide coverage. In this invention, it is used to monitor data transmission between the terminal and the gateway (such as LoRa network).

[0166] 14. LoRa Gateway: A core device for low-power wide area networks, used to aggregate data uploaded by multiple monitoring terminals and transmit it to a remote data analysis platform via a backhaul network, thereby enabling communication between the perception layer and the platform layer.

[0167] 15. MEMS process: Micro-Electro-Mechanical Systems process, used to manufacture the hydrogen sensor in this invention, enables sensor miniaturization, low power consumption, and adaptability to the integrated design of monitoring terminals.

[0168] 16. Electrochemical sensor: A device for detecting gas concentration based on electrochemical principles. In this invention, it can be used as a hydrogen sensing unit, and has the characteristics of high selectivity and fast response.

[0169] 17. Thermal conductivity sensor: A device for detecting gas concentration based on the difference in thermal conductivity of gases. In this invention, it can be used as a hydrogen sensing unit to meet the detection requirements of dissolved hydrogen in oil.

[0170] 18. PT100 platinum resistance thermometer: a high-precision temperature sensor used in this invention to detect the top layer oil temperature of the distribution transformer. It directly contacts the oil flow and can accurately reflect the thermal state of the insulation system.

[0171] 19. Radio frequency capacitive sensor: A device for detecting oil level based on the principle of capacitance change. In this invention, it is used in the oil level detection unit and outputs an electrical signal (e.g., 4-20mA) corresponding to the oil level.

[0172] 20. Non-volatile memory: The built-in storage device of the monitoring terminal can retain data after power failure. It is used to cache pre-processed valid data such as hydrogen, temperature and oil level to avoid data loss.

[0173] 21. DC-DC Conversion Circuit: A core component of the monitoring terminal power management module, used to convert the battery voltage into a stable voltage required by various functional modules. It features high efficiency and can reduce power loss.

[0174] 22. Load switch array: A component of the power management module, controlled by the main control module. It can cut off the power supply to the sensors and communication modules during non-working periods, maintaining only the operation of the core circuit, thereby further reducing power consumption.

[0175] 23. Real-time Clock (RTC): The timing component that monitors the terminal, used to control the sampling period (T1) and upload period (T2), and to wake up the terminal at regular intervals to perform tasks such as data acquisition, aggregation and upload. It is the core of low-power scheduling.

[0176] 24. Oil chromatography (DGA): A standard method for traditional insulation fault diagnosis, which determines faults by analyzing various characteristic gases (such as hydrogen, methane, acetylene, etc.) dissolved in insulating oil in the laboratory. This invention improves upon its limitations in offline detection.

[0177] 25. Electrical faults: Transformer insulation defects caused by electrical stress such as partial discharge. The core characteristic is that the hydrogen concentration rises independently of the oil temperature (dH2 / dt>δ1 and |dT / dt|<δ2), which is one of the fault types that can be distinguished by this invention.

[0178] 26. Thermal faults: Transformer insulation defects caused by thermal stress such as overheating and overload. The core characteristic is that the hydrogen concentration and oil temperature rise synchronously (dH2 / dt>δ1 and dT / dt>δ3), which is one of the fault types that can be distinguished by this invention.

[0179] 27. Partial discharge: A discharge phenomenon caused by the local electric field strength in the distribution transformer insulation system exceeding the dielectric breakdown field strength. It is a type of electrical fault and will continuously produce hydrogen gas in the early stages.

[0180] 28. Insulating oil: The insulating medium filled in the transformer oil tank, which has both insulation and heat dissipation functions. Characteristic gases (such as hydrogen) generated by insulation defects will dissolve in it, and it is the direct object of sensor detection.

[0181] 29. Rated oil level: The standard oil level height (denoted as L0) when the distribution transformer is operating normally is the reference benchmark for the oil level dynamic compensation function. When the actual oil level is lower than this value, concentration correction is required.

[0182] 30. Historical baseline data: Historical statistical data on hydrogen concentration (including mean μ and standard deviation σ) under similar oil temperature conditions when the distribution transformer is in normal operation is the core basis for calculating the dynamic early warning threshold.

[0183] 31. Rate of change (dH2 / dt, dT / dt): The rate of change of hydrogen concentration and oil temperature within a preset time window (units are ppm / h and ℃ / h, respectively). These are sensitive indicators for identifying early abnormal trends and are used to determine the type of fault.

[0184] 32. Thresholds (δ1, δ2, δ3): Key parameters for hydrogen-temperature correlation discrimination. δ1 is the positive threshold for hydrogen change rate, δ2 is the stability threshold for oil temperature change rate, and δ3 is the oil temperature rise threshold. These parameters can be adaptively adjusted according to the transformer model and operating history.

[0185] 33. Machine Learning Model: An algorithmic model deployed on a data analysis platform, using historical multidimensional data sequences and fault records as training sets, outputs insulation health status scores and defect probability predictions to optimize early warning decisions.

[0186] 34. Structured Early Warning Report: A standardized report generated by the data analysis platform, containing information such as early warning level, fault type tendency, data trend, and maintenance suggestions, which can be pushed to user terminals through a graphical interface or preset channels.

[0187] 35. Plug and play: The deployment characteristics of the monitoring terminal are that it can be directly installed without power interruption through a standard mechanical interface that is compatible with transformer oil valves, without the need for professional wiring, thus enabling rapid deployment.

[0188] 36. Protection rating (IP67 / IP68): The protection standard of the monitoring terminal housing. IP67 means that it is completely protected against dust intrusion and can be immersed in 1 meter of water for 30 minutes without damage; IP68 has a higher protection rating and can meet the needs of use in complex environments such as outdoors and humid environments.

[0189] 37. Lithium-thionyl chloride battery: A high-energy-density battery adapted to the low-power design of monitoring terminals, providing more than 5 years of battery life, and is the core power supply component in scenarios where the terminal has no external power source.

[0190] 38. Digital filtering: Preprocessing methods for raw sampled data by monitoring terminals, including moving average filtering, median filtering, etc., to eliminate instantaneous interference and random noise and improve data reliability.

Claims

1. A low-power integrated monitoring method for hydrogen gas, temperature, and oil level inside a distribution transformer, characterized in that: Based on monitoring terminals and a remote data analysis platform, the monitoring terminals deployed in the distribution transformer oil circuit send the monitored parameters to the remote data analysis platform for processing, including the following steps: S1. Multi-source data synchronous acquisition: The monitoring terminal synchronously acquires the dissolved hydrogen concentration H2, top oil temperature T and oil level L in the distribution transformer insulating oil in the first sampling period T1 to obtain the original data sequence; S2. Local preprocessing and caching: The original data sequence is digitally filtered to eliminate transient interference, and the processed effective data is stored in local non-volatile memory; S3, Low-power aggregation upload: The monitoring terminal wakes up at a timed interval of the second upload cycle T2, where T2 > T1, and packages multiple sets of valid data stored in the memory within the most recent T2 cycle, and uploads them to the remote data analysis platform through the low-power wide area network wireless communication module. S4. Cloud-based multi-parameter fusion analysis: The remote data analysis platform performs the following analysis steps: S41. Trend feature extraction: Calculate the first rate of change dH2 / dt of the hydrogen concentration value H2 within a preset time window, and the second rate of change dT / dt of the oil temperature value T within the same time window; S42. Hydrogen-Temperature Correlation Judgment: Establish a correlation model. When the conditions are met: dH2 / dt > δ1 and |dT / dt| < δ2 for a continuous period of time, it is determined that the early insulation defect tends to be an electrical fault, and a first-level warning sign is generated; where δ1 is the positive threshold of hydrogen change rate and δ2 is the stability threshold of oil temperature change rate. S43. Dynamic oil level compensation: Based on the received oil level height value L, the hydrogen concentration value H2 is corrected using the pre-stored oil level-concentration compensation function F(L) to obtain the corrected hydrogen concentration value H2' = H2 × F(L), where F(L) > 1 when L is lower than the rated oil level. S5. Dynamic threshold-based early warning: Based on the historical hydrogen concentration baseline data of the transformer, the current oil temperature value T, and the first-level early warning identifier, a dynamic early warning threshold Y is generated; the corrected hydrogen concentration value H2' or its changing trend is compared with the dynamic early warning threshold Y, and if it exceeds the threshold, a second-level early warning information containing the defect type tendency is generated and output.

2. The low-power integrated monitoring method for hydrogen, temperature, and oil level inside a distribution transformer according to claim 1, characterized in that: The monitoring terminal operates in intermittent deep sleep mode. Its static power consumption in deep sleep mode is ≤1mW, and its average power consumption during the active working cycle, which includes data acquisition, processing, and communication, is ≤30mW, so as to achieve a battery life of more than 5 years with the built-in battery.

3. The low-power integrated monitoring method for hydrogen, temperature, and oil level inside a distribution transformer according to claim 1, characterized in that: The hydrogen temperature correlation judgment in step S4 further includes: when dH2 / dt > δ1 and dT / dt > δ3, δ3 is the oil temperature rise threshold, and the cause of hydrogen production is determined to be a thermal fault correlation; when only dH2 / dt > δ1 and |dT / dt| < δ2, the conclusion of electrical fault judgment is strengthened; the thresholds δ1, δ2, and δ3 are adaptively adjusted according to the transformer model, operating history, and seasonal characteristics of ambient temperature.

4. The low-power integrated monitoring method for hydrogen, temperature, and oil level inside a distribution transformer according to claim 1, characterized in that: The method for generating the dynamic early warning threshold Y in step S5 is as follows: Y = μ + kσ + f(T) + g(Flag) Where μ and σ are the mean and standard deviation of the historical hydrogen concentration data of the transformer under similar oil temperature conditions, respectively; k is the sensitivity coefficient; f(T) is the oil temperature correction term, reflecting the influence of temperature on the background of gas production; g(Flag) is the warning level adjustment term, and when the first-level warning flag exists, g(Flag) > 0.

5. The low-power integrated monitoring method for hydrogen, temperature, and oil level inside a distribution transformer according to claim 1, characterized in that: Step S5 is followed by step S6: Early warning visualization and report generation: The remote data analysis platform integrates the second-level early warning information, related trend curves, correlation analysis conclusions and maintenance suggestions to generate a structured early warning report, and pushes it to the user terminal through a graphical interface and preset communication channels.

6. A monitoring terminal for implementing the integrated low-power monitoring method for hydrogen, temperature, and oil level inside a distribution transformer as described in any one of claims 1-5. Its features include: a fully metal shielded sealed housing, an integrated sensing module, a main control and data processing module, a low-power wireless communication module, and a power management module; the fully metal shielded sealed housing constitutes the main body for electromagnetic shielding and environmental protection, and one end of it is provided with a mechanical interface adapted to a transformer oil valve, realizing plug-and-play and oil circuit sealing; the integrated sensing module is built into the fully metal shielded sealed housing and includes: Hydrogen sensing unit is used to detect the concentration of dissolved hydrogen in oil; Temperature sensing unit for detecting oil temperature; Oil level detection unit, used to detect oil level height; The main control and data processing module is electrically connected to the integrated sensing module and is used to control the sampling timing, perform data preprocessing and caching; the low-power wireless communication module is used to interact with an external gateway; the power management module is connected to a high-energy-density battery and is used to provide stable power supply to each functional module and implement precise power consumption status management; wherein, the main control and data processing module is configured to execute steps S1 to S3 of the integrated low-power monitoring method for hydrogen, temperature and oil level inside the distribution transformer as described in claim 1.

7. A low-power integrated monitoring terminal for hydrogen, temperature, and oil level inside a distribution transformer according to claim 6, characterized in that: The hydrogen sensing unit of the integrated sensing module is an electrochemical or thermal conductivity type hydrogen sensor based on microelectromechanical systems (MEMS) technology; the protection level of the all-metal shielded sealed housing is not lower than IP67; the mechanical interface is a standard threaded or flanged structure. The high-energy-density battery is a lithium-thionyl chloride battery.

8. A low-power integrated monitoring terminal for hydrogen, temperature, and oil level inside a distribution transformer according to claim 6, characterized in that: The power management module includes a high-efficiency DC-DC conversion circuit and a load switch array. The main control and data processing module controls the on / off state of the load switch array to completely power off the low-power wireless communication module and the integrated sensing module during non-working periods, maintaining only the real-time clock and wake-up circuit.

9. A remote data analysis platform for implementing the integrated low-power monitoring method for hydrogen, temperature, and oil level inside a distribution transformer as described in any one of claims 1-5, characterized in that: It includes a network transmission layer and a data analysis platform. The network transmission layer is used to aggregate data from the monitoring terminal. The data analysis platform is used to receive, store and process data from the network transmission layer, and execute steps S4 to S5 in the above method to implement intelligent early warning of early insulation defects in distribution transformers using single hydrogen detection.

10. The integrated low-power remote data analysis platform for hydrogen, temperature, and oil level inside a distribution transformer according to claim 9, characterized in that: The data analysis platform also includes a machine learning model, which uses historical multidimensional data sequences and corresponding fault records as a training set to perform in-depth analysis of the input real-time data features and output insulation health status scores and defect probability predictions to optimize the decision-making of graded early warning and improve the accuracy of early warning.

Images