Moisture absorber intelligent early warning decision system

Abstract

This invention provides an intelligent early warning and decision-making system for dehumidifiers, relating to the field of power equipment condition monitoring and early warning technology. It includes a multi-source data acquisition module, a data transmission module, a data processing module, an intelligent early warning and decision-making engine, an early warning classification module, and an application module. The multi-source data acquisition module collects multi-dimensional data about the dehumidifier and its surrounding environment. The data transmission module transmits the collected data to a cloud platform, supporting multiple communication protocols. This invention employs multi-source data fusion, integrating multi-dimensional data such as the dehumidifier's internal humidity, ambient temperature and humidity, transformer oil level, internal equipment pressure, and silica gel color image. Combined with fusion algorithms such as Kalman filtering and weighted averaging, it effectively solves the problems of single monitoring dimensions and low data accuracy in existing technologies. This significantly improves the accuracy and reliability of dehumidifier condition monitoring, avoids misjudgments caused by single data deviations, and provides high-quality data support for subsequent early warning and decision-making.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and early warning technology, and in particular to an intelligent early warning decision system for dehumidifiers. Background Technology

[0002] Dehumidifiers are crucial auxiliary components of oil-filled electrical equipment such as transformers and reactors. Their core function is to absorb moisture from the air, preventing moisture from entering the equipment and causing deterioration of the insulating oil and reduced insulation performance, thereby avoiding equipment failure and ensuring the safe and stable operation of the electrical equipment. Traditional dehumidifiers generally use silica gel as the moisture-absorbing material. When the silica gel reaches saturation, its color changes noticeably, from blue to pink. Maintenance personnel observe this color change during regular inspections to determine if the dehumidifier needs replacement. With the application of IoT, big data, and AI technologies in the field of power equipment maintenance, some intelligent monitoring solutions for dehumidifiers have emerged. These solutions often use a single sensor to monitor the dehumidifier's status, assisting maintenance personnel in assessing the equipment's condition.

[0003] However, traditional manual inspection methods have long inspection cycles, cannot promptly detect the saturation state of dehumidifiers, and rely on human experience to judge the color of silica gel, which is greatly affected by ambient light, highly subjective, and prone to misjudgment. Existing intelligent monitoring solutions have limited monitoring dimensions, collecting data only through a single sensor, resulting in low data accuracy and reliability, and fixed warning thresholds that cannot adapt to different environmental conditions. At the same time, they lack effective trend prediction capabilities and intelligent decision-making mechanisms, making it impossible to predict changes in the state of dehumidifiers, hindering preventative maintenance, and only enabling reactive fault responses. Furthermore, the lack of a unified monitoring and management platform prevents remote monitoring and centralized management of dehumidifier status, leads to untimely alarm information transmission, and results in low operation and maintenance efficiency. Therefore, this invention proposes an intelligent early warning and decision-making system for dehumidifiers to solve the problems existing in the prior art. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes an intelligent early warning and decision-making system for dehumidifiers. This system employs multi-source data fusion, integrating multi-dimensional data such as internal humidity of the dehumidifier, ambient temperature and humidity, transformer oil level, internal pressure of the equipment, and silica gel color images. Combined with fusion algorithms such as Kalman filtering and weighted averaging, it effectively solves the problems of single monitoring dimensions and low data accuracy in existing technologies. This significantly improves the accuracy and reliability of dehumidifier status monitoring, avoids misjudgments caused by single data deviations, and provides high-quality data support for subsequent early warning and decision-making. Furthermore, through environmental data correction, the system can adapt to different environmental conditions, expanding its applicability.

[0005] To achieve the objectives of this invention, the invention is implemented through the following technical solution: a dehumidifier intelligent early warning decision system, comprising a multi-source data acquisition module, a data transmission module, a data processing module, an intelligent early warning decision engine, an early warning classification module, and an application module. The multi-source data acquisition module is used to collect multi-dimensional data of the dehumidifier and its surrounding environment; the data transmission module is used to transmit the collected data to a cloud platform, supporting multiple communication protocols; the data processing module is used to preprocess and fuse the multi-source data to generate a comprehensive status index.

[0006] The intelligent early warning decision engine is used to realize data anomaly judgment, trend prediction, risk assessment and intelligent decision-making; the early warning classification module is used to classify early warning levels and match response strategies; the application module is used to realize multi-channel alarm notification and remote monitoring management. The modules work together to realize real-time monitoring, early warning and operation and maintenance decision-making of the dehumidifier status.

[0007] Further improvements include: the multi-source data acquisition module includes a humidity sensor, a temperature sensor, an oil level sensor, a pressure sensor, a vision sensor, and an environmental sensor; the humidity sensor is used to detect the humidity inside the dehumidifier, with a range of 0-100%RH and an accuracy of ±2%RH; the temperature sensor is used to detect the ambient temperature, with a range of... The accuracy is ±0.5℃; the oil level sensor is used to detect changes in transformer oil level, with a range of 0-100mm and an accuracy of ±1mm; the pressure sensor is used to detect internal pressure of the equipment, with a range of 0-1MPa and an accuracy of ±0.01MPa; the vision sensor is used to acquire silicone color images with a resolution of not less than 1080P; the environmental sensor is used to detect ambient temperature and humidity, with a temperature range of -40~85℃ and a humidity range of 0-100%RH, with accuracies of ±0.5℃ and ±2%RH respectively.

[0008] Further improvements are made in the following aspects: In the multi-source data acquisition module, the sampling frequency of each sensor is as follows: humidity data and temperature data once per minute, oil level data once per 5 minutes, pressure data once per 10 minutes, image data once per hour, and ambient temperature and humidity data once per 3 minutes; The data transmission module uses an edge gateway as a data relay device, supports MQTT, HTTP, and 4G communication protocols, and has a data caching function. When the network is interrupted, it caches data for no less than 72 hours and automatically uploads it after the network is restored.

[0009] Further improvements include: the preprocessing of the data processing module includes data cleaning, temporal alignment, feature engineering, and multi-source data fusion; the data cleaning uses the 3σ criterion to remove outlier data and linear interpolation to supplement missing data; the temporal alignment uses the humidity data sampling time as a benchmark and performs interpolation or downsampling processing on data with different sampling frequencies; the feature engineering is used to extract key feature parameters such as humidity change rate, silica gel color saturation, and temperature compensation coefficient; the multi-source data fusion uses Kalman filtering, weighted averaging, and outlier removal algorithms, and the fusion weights are determined based on the SHAP analysis results of 1000 sets of measured data.

[0010] A further improvement lies in the fact that the state equation of the Kalman filter is:

[0011] ,

[0012] The observation equation is:

[0013] ;

[0014] Where x(k) is the fused data at time k, A is the state transition matrix (1.02), B is the control matrix (0.05), u(k) is the sensor correction coefficient, w(k) is the process noise (N(0,0.01)), z(k) is the original sensor data at time k, H is the observation matrix (1), and v(k) is the observation noise (N(0,0.02)).

[0015] The formula for the weighted average is:

[0016] S=0.35×H+0.28×C+0.18×T+0.12×O+0.07×P,

[0017] Where S is the comprehensive state index, H is the humidity normalized value, C is the silica gel color saturation, T is the temperature compensation value, O is the oil level change rate, and P is the pressure normalized value.

[0018] Further improvements are made in the following aspects: The intelligent early warning decision engine includes a threshold judgment module, a trend prediction module, a risk assessment module, and a decision reasoning module; the threshold judgment module compares the fused data and feature parameters with a preset threshold to determine whether the data is abnormal; the trend prediction module uses an LSTM time-series prediction model, inputting humidity data from the past 72 hours to predict the humidity change trend for the next 24 hours. The model has 5 neurons in the input layer, 2 hidden layers with 64 LSTM units per layer, and 1 neuron in the output layer, with 32 training batches and 100 training iterations; the risk assessment module calculates a comprehensive risk value based on multiple risk factors, and the risk calculation formula is:

[0019] R=0.35×H+0.28×C+0.18×T+0.12×Ht+0.07×E,

[0020] Wherein, R is the comprehensive risk value, unitless, ranging from 0 to 100, used to quantify the operational risk level of the dehumidifier; H is the humidity normalization value, ranging from 0 to 1, obtained by normalizing the humidity data inside the dehumidifier, reflecting the internal humidity level of the dehumidifier; C is the silica gel color saturation, ranging from 0 to 1, extracted by recognizing the silica gel color image collected by the visual sensor, reflecting the moisture absorption degree of the silica gel; T is the temperature compensation value, ranging from 0 to 1, obtained by correcting the humidity data according to the ambient temperature, used to eliminate the influence of temperature on the humidity monitoring results; Ht is the historical trend factor, ranging from 0 to 1, calculated based on the historical operating data and status change trend of the dehumidifier, reflecting the long-term change trend of the dehumidifier's status; E is the environmental correction factor, ranging from 0 to 1, calculated based on surrounding environmental factors such as ambient temperature, humidity, and dust, used to adapt to risk assessment under different environmental conditions.

[0021] Further improvements include: the decision reasoning module adopts a multi-algorithm fusion strategy of rule-based reasoning, fuzzy logic, neural networks, and expert systems, with adaptive weight adjustment; the decision rules are as follows: when R≥80, output immediate shutdown and maintenance suggestion; when 50≤R<80, output switch to backup system and planned maintenance suggestion; when 30≤R<50, output enhanced monitoring and key attention suggestion; when 10≤R<30, output log recording and regular inspection suggestion; and when R<10, output normal monitoring suggestion; multiple risk factors include humidity exceedance level, temperature anomaly amplitude, oil level change rate, duration, and historical fault correlation.

[0022] Further improvements include: the warning classification module divides warning levels into four levels, with clear judgment criteria and response strategies for each level: Level 1 warning R≥80: humidity >95%RH or equipment failure is imminent; the response strategy is immediate shutdown, manual inspection, and notification of responsible personnel through multiple channels; Level 2 warning 50≤R<80: humidity >85%RH or abnormal temperature; the response strategy is automatic switching to the backup system and notification of maintenance personnel to handle the situation within 4 hours; Level 3 warning 30≤R<50: humidity >75%RH or worsening trend; the response strategy is to increase the sampling frequency and notify maintenance personnel to check within 24 hours; Level 4 warning 10≤R<30: humidity >65%RH or slight abnormality; the response strategy is to log and conduct routine inspections; the system supports automatic escalation when a warning is not handled within 30 minutes or is repeatedly triggered.

[0023] Further improvements include: the application module includes a local monitoring terminal, a cloud platform, a mobile APP, and a data dashboard; the local monitoring terminal provides real-time data display, alarm indication, manual control, and parameter setting functions, and stores no less than one month of local data; the cloud platform supports data storage, analysis, forwarding, and device management, and generates visual reports; the mobile APP supports device lists, real-time overviews, alarm push notifications, historical queries, and remote control; and the data dashboard enables centralized display of multiple device statuses and alarm statistics.

[0024] Further improvements are made in the following aspects: The system adopts a five-layer architecture, from bottom to top: data acquisition layer, based on multi-source data acquisition module, data transmission layer, based on data transmission module, data processing layer, based on data processing module, intelligent decision layer, based on intelligent early warning decision engine and early warning classification module, application layer, based on application module; The system workflow includes data acquisition, preliminary verification, data upload, preprocessing and fusion, early warning decision, alarm output, operation and maintenance processing, and status reset.

[0025] The beneficial effects of this invention are as follows:

[0026] 1. This invention employs multi-source data fusion, integrating multi-dimensional data such as the humidity inside the dehumidifier, ambient temperature and humidity, transformer oil level, internal equipment pressure, and silica gel color image. Combined with fusion algorithms such as Kalman filtering and weighted averaging, it effectively solves the problems of single monitoring dimensions and low data accuracy in existing technologies. It significantly improves the accuracy and reliability of dehumidifier status monitoring, avoids misjudgments caused by single data deviations, and provides high-quality data support for subsequent early warning and decision-making. At the same time, through environmental data correction, the system can adapt to different environmental conditions, expanding the system's applicability.

[0027] 2. This invention constructs an integrated intelligent early warning decision engine, integrating threshold judgment, LSTM time series prediction, risk assessment, and multi-algorithm fusion decision-making functions. It realizes the transformation from passively responding to faults to actively preventing faults. Through the LSTM time series prediction model, it can accurately predict the trend of dehumidifier status changes and discover potential fault hazards in advance, providing a scientific basis for preventive maintenance. Through comprehensive evaluation of multiple risk factors and multi-algorithm fusion decision-making, it ensures the rationality and accuracy of decision results, solves the problems of existing technologies being unable to predict trends and lacking intelligent decision-making mechanisms, and reduces the failure rate of power equipment.

[0028] 3. This invention employs a tiered early warning mechanism coupled with multi-channel alarm notifications and remote monitoring functions, achieving precise and intelligent operation and maintenance management of the dehumidifier. Four early warning levels correspond to different response strategies, enabling targeted handling measures based on the severity of the fault, avoiding excessive or untimely maintenance. Multi-channel alarm notifications ensure timely delivery of alarm information to relevant personnel, preventing fault escalation due to information delays. Remote monitoring and mobile management functions allow for monitoring and operation and maintenance scheduling of the dehumidifier anytime, anywhere, significantly reducing manual inspection workload, improving operation and maintenance efficiency, and promoting the development of power equipment operation and maintenance towards less-manned and unmanned operations. Attached Figure Description

[0029] Figure 1 This is a diagram of the overall system architecture of the present invention;

[0030] Figure 2 This is a timing diagram of the system workflow of the present invention;

[0031] Figure 3 This is a comparison chart of the early warning time in the comparative experiment of the present invention;

[0032] Figure 4 This is a comparison chart of the false alarm rate and false negative rate of the present invention;

[0033] Figure 5 This is a risk factor weight distribution diagram of the present invention;

[0034] Figure 6 This is a radar chart comparing the overall performance of the present invention. Detailed Implementation

[0035] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0036] Example 1

[0037] according to Figure 1 , 2 As shown in Figures 3, 4, 5, and 6, this embodiment proposes an intelligent early warning and decision-making system for dehumidifiers. It adopts a five-layer architecture design, consisting of a data acquisition layer, a data transmission layer, a data processing layer, an intelligent decision-making layer, and an application layer from bottom to top. Each layer works in concert to ensure the stable and efficient operation of the system.

[0038] The data acquisition layer, or multi-source data acquisition module, deploys various types of sensors. Specific parameters are as follows: A digital humidity sensor with a range of 0-100%RH and an accuracy of ±2%RH is selected for real-time monitoring of the humidity inside the dehumidifier; a temperature sensor with a range of... The system includes: a thermocouple sensor with an accuracy of ±0.5℃ for detecting ambient temperature; a float-type oil level sensor with a range of 0-100mm and an accuracy of ±1mm for detecting changes in transformer oil level; a pressure transmitter with a range of 0-1MPa and an accuracy of ±0.01MPa for detecting internal equipment pressure; a high-definition industrial camera with a resolution of at least 1080P for capturing silicone color images; and an integrated temperature and humidity sensor with ranges of [missing information - likely related to temperature and humidity measurement]. Temperature and humidity are measured from 0-100%RH, with accuracies of ±0.5℃ and ±2%RH, respectively. Each sensor collects data according to a preset sampling frequency: humidity and temperature data are sampled once per minute, oil level data once every 5 minutes, pressure data once every 10 minutes, image data once per hour, and ambient temperature and humidity data once every 3 minutes, ensuring the real-time nature and completeness of data acquisition.

[0039] The data transmission layer, or data transmission module, uses an edge gateway as a data relay device. The edge gateway integrates multiple communication modules and supports communication protocols such as MQTT, HTTP, and 4G. It can automatically switch communication modes according to the on-site network environment: when a stable local area network exists on-site, the MQTT protocol is used for data transmission with a transmission latency of ≤1s; when there is no local area network on-site, the 4G communication protocol is used with a transmission rate of ≥1Mbps, ensuring the reliability and real-time performance of data transmission. The edge gateway also has a data caching function, which can cache no less than 72 hours of data when the network is interrupted, and automatically upload it to the cloud platform after the network is restored.

[0040] The data processing layer, or data processing module, is deployed on a cloud platform or edge server, employing a distributed processing architecture capable of simultaneously processing multi-source data from multiple devices. The data preprocessing process is as follows: First, data cleaning is performed, using the 3σ criterion to remove outlier data, linear interpolation to supplement missing data, and redundant / duplicate data is deleted. Next, time-series alignment is performed, using the sampling time of humidity data as a benchmark, interpolating or downsampling data with different sampling frequencies to ensure consistent timestamps across all data. Then, feature engineering is performed, extracting key feature parameters such as humidity change rate (the amount of humidity change per unit time), silica gel color saturation (extracted through image recognition algorithms), and temperature compensation coefficient (a coefficient used to correct humidity data based on ambient temperature). Finally, multi-source data fusion is performed, using a Kalman filter algorithm to eliminate sensor noise, combined with a weighted average algorithm to fuse multi-dimensional data and generate a comprehensive status index. The weights of each data point are determined based on the SHAP analysis results of 1000 sets of measured data: humidity data weight 0.35, silica gel color data weight 0.28, temperature data weight 0.18, oil level data weight 0.12, and pressure data weight 0.07.

[0041] The intelligent decision-making layer, also known as the intelligent early warning decision engine, is deployed on a cloud platform using a heterogeneous computing architecture of CPU+GPU to ensure algorithm efficiency. The parameters of the LSTM time-series prediction model are as follows: 5 neurons in the input layer (corresponding to 5 core data types), 2 LSTM layers in the hidden layer with 64 units per layer, 1 neuron in the output layer (corresponding to the predicted humidity value), a batch size of 32, 100 training epochs, the Adam optimizer, a learning rate of 0.001, and training data consisting of 1000 sets of real-world dehumidifier test data, with a prediction error ≤3%.

[0042] The application layer, or application module, includes local monitoring terminals, a cloud platform, a mobile app, and a large data dashboard. The local monitoring terminals are deployed in the power equipment field control room, using industrial touchscreens to provide real-time data display, alarm status indication, manual control, and parameter setting functions. The cloud platform is deployed on cloud servers, supporting data storage, analysis, forwarding, and backup, and can store at least three years of historical data. The mobile app supports iOS and Android systems, providing device lists, real-time overviews, alarm push notifications, historical queries, remote control, and a personal center. The large data dashboard uses a high-definition LED screen, supporting centralized display of multiple device statuses, data visualization analysis, and alarm statistics.

[0043] Example 2

[0044] according to Figure 1 , 2 As shown in Figures 3, 4, 5, and 6, this embodiment proposes an intelligent early warning decision-making system for dehumidifiers. The multi-source data acquisition and fusion process is as follows: First, each sensor collects raw data according to a preset sampling frequency. The humidity sensor collects humidity data inside the dehumidifier, the temperature sensor collects ambient temperature data, the oil level sensor collects transformer oil level data, the pressure sensor collects internal pressure data, the vision sensor collects silicone color image data, and the environmental sensor collects ambient temperature and humidity data. Then, all raw data is transmitted to the edge gateway, which performs preliminary verification of the raw data and removes obviously abnormal data. Next, the preliminarily verified data is transmitted to the data processing module, which cleans, aligns the time sequence, and extracts features from the data. Finally, a multi-source data fusion algorithm is used to fuse the processed data. Specifically, the Kalman filter algorithm is used to denoise the data from each sensor. The state equation and observation equation of the Kalman filter are as follows:

[0045] Equations of state: ,

[0046] Observation equation: z(k) = H·x(k) + v(k),

[0047] Where x(k) is the state vector, representing the fused data at time k; A is the state transition matrix with a value of 1.02; B is the control matrix with a value of 0.05; u(k) is the control vector, representing the sensor's correction coefficient; w(k) is the process noise, following a normal distribution N(0,0.01); z(k) is the observation vector, representing the sensor's raw data at time k; H is the observation matrix with a value of 1; and v(k) is the observation noise, following a normal distribution N(0,0.02). After Kalman filtering, a weighted average algorithm is used to fuse the multi-dimensional data to generate a comprehensive state index. The weighted average formula is: S = 0.35 × H + 0.28 × C + 0.18 × T + 0.12 × O + 0.07 × P, where S is the comprehensive state index, H is the humidity normalized value, C is the silica gel color saturation, T is the temperature compensation value, O is the oil level change rate, and P is the pressure normalized value.

[0048] Example 3

[0049] according to Figure 1 , 2 As shown in Figures 3, 4, 5, and 6, this embodiment proposes an intelligent early warning decision system for dehumidifiers. The intelligent early warning algorithm adopts a three-step process: S1 threshold judgment, S2 trend prediction, and S3 risk level assessment. The specific process is as follows:

[0050] S1: Threshold Judgment Phase. The system compares the comprehensive status indicators and individual dimension data output by the data processing module with preset thresholds. The preset thresholds are adaptively adjusted according to different environmental conditions and equipment models. The humidity thresholds are as follows: Level 1 warning threshold 95%RH, Level 2 warning threshold 85%RH, Level 3 warning threshold 75%RH, and Level 4 warning threshold 65%RH; the temperature anomaly threshold is exceeding the average ambient temperature by ±10℃; the oil level change rate anomaly threshold is exceeding 0.5mm / h; and the pressure anomaly threshold is exceeding 0.8MPa or falling below 0.2MPa. If all data are within the preset threshold range, it is judged as a normal state, and real-time monitoring continues; if any data exceeds the preset threshold, it is judged as an abnormal state, and the system enters the S2 trend prediction phase.

[0051] S2: In the trend prediction stage, an LSTM (Long Short-Term Memory) time-series prediction model is used to predict the changing trend of humidity data inside the dehumidifier. The LSTM model includes an input gate, a forget gate, and an output gate, which can effectively capture long-term dependencies in time-series data. Specifically, the humidity data from the past 72 hours is used as input. The input gate controls the data input, the forget gate discards irrelevant historical data, and the output gate outputs the prediction result, predicting the humidity change trend for the next 24 hours and calculating the humidity change rate. The prediction result, combined with the humidity change rate, outputs a preliminary risk value: Preliminary risk value = (Current humidity - Normal humidity threshold) × Humidity change rate × Duration.

[0052] S3: Risk Level Assessment Phase. Based on the preliminary risk value and multiple risk factors, a comprehensive risk value is calculated. The risk calculation formula is: R = 0.35 × H + 0.28 × C + 0.18 × T + 0.12 × Ht + 0.07 × E, where R is the comprehensive risk value, H is the humidity normalized value (range 0-1), C is the silica gel color saturation (range 0-1, approaching 0 for blue and approaching 1 for pink), T is the temperature compensation value (range 0-1, the larger the value for more abnormal temperatures), Ht is the historical trend factor (range 0-1, the larger the value for more severe trends), and E is the environmental correction factor (range 0-1, the larger the value for more severe environmental conditions). Based on the comprehensive risk value, the warning level is divided into four levels: R ≥ 80 is Level 1 warning, 50 ≤ R < 80 is Level 2 warning, 30 ≤ R < 50 is Level 3 warning, 10 ≤ R < 30 is Level 4 warning, and R < 10 is normal.

[0053] Example 4

[0054] according to Figure 1 , 2As shown in Figures 3, 4, 5, and 6, this embodiment proposes an intelligent early warning and decision-making system for dehumidifiers. The decision engine integrates multiple risk factors for intelligent decision-making. The specific working process is as follows: First, it receives fused data, feature parameters, and trend prediction results output by the data processing module. Then, the risk assessment module calculates a comprehensive risk value based on risk factors such as the degree of humidity exceeding the standard, the magnitude of temperature anomalies, the rate of oil level change, the duration, and historical fault associations. The historical fault association factor is determined based on the historical fault data of this device and similar devices. If the device has experienced dehumidifier-related faults, the historical fault association factor is increased by 0.1-0.2. Next, the decision reasoning module adopts a multi-algorithm fusion strategy, combining the comprehensive risk value and trend prediction results for intelligent decision-making: the rule reasoning module outputs basic decision suggestions based on preset operation and maintenance rules; the fuzzy logic module processes fuzzy data (such as the gradient process of silicone color) and optimizes the decision suggestions; the neural network module performs self-learning based on historical data and adjusts the decision parameters; the expert system module integrates the experience of power operation and maintenance experts to verify and correct the decision suggestions, and finally outputs the optimal operation and maintenance decision suggestions. The decision-making rules are as follows: when R≥80, the decision suggestion is "immediate shutdown and manual maintenance"; when 50≤R<80, the decision suggestion is "switch to the backup system and perform planned maintenance"; when 30≤R<50, the decision suggestion is "strengthen monitoring and focus on key areas"; when 10≤R<30, the decision suggestion is "record logs and conduct regular inspections"; when R<10, the decision suggestion is "normal monitoring and no intervention required".

[0055] Example 5

[0056] according to Figure 1 , 2 As shown in Figures 3, 4, 5, and 6, this embodiment proposes an intelligent early warning decision-making system for dehumidifiers. The early warning hierarchy adopts a pyramid structure, from bottom to top: Level 4 warning (advice), Level 3 warning (attention), Level 2 warning (serious), and Level 1 warning (emergency). Each level of warning corresponds to different response strategies and notification methods, as detailed below:

[0057] Level 1 Warning (Emergency): The comprehensive risk value R ≥ 80. The judgment condition is that the humidity is greater than 95%RH or the equipment failure is about to occur (such as a sudden abnormal pressure). The response strategy is to immediately shut down the machine, cut off the power supply to the relevant equipment, and send alarm information to the operation and maintenance manager and relevant management personnel through all notification methods such as audible and visual alarms, SMS, APP push, email, and telephone calls, requiring them to arrive at the site for manual inspection within 1 hour. After the inspection is completed, the warning status is manually reset.

[0058] Level 2 Warning (Severe): The comprehensive risk value is 50≤R<80. The judgment condition is humidity greater than 85%RH or abnormal temperature. The response strategy is to automatically switch to the backup dehumidifier system to ensure the normal operation of the power equipment. At the same time, alarm information is sent to the operation and maintenance personnel through SMS, APP push, email and other notification methods, requiring them to arrive at the site for inspection and handling within 4 hours. After the handling is completed, the system automatically resets the warning status.

[0059] Level 3 Warning (Note): The overall risk value is 30≤R<50. The judgment condition is that the humidity is greater than 75%RH or the trend is deteriorating. The response strategy is to double the sensor sampling frequency (the sampling frequency of humidity and temperature becomes once every 30 seconds, and the sampling frequency of oil level becomes once every 2.5 minutes) to strengthen monitoring. At the same time, the operation and maintenance personnel will be notified to pay close attention through APP push and platform alarm. The operation and maintenance personnel must complete the equipment status verification within 24 hours.

[0060] Level 4 Warning (Alert): The comprehensive risk value is 10≤R<30. The judgment condition is humidity greater than 65%RH or slight abnormality. The response strategy is to record the equipment operation log, maintain the regular inspection cycle, and record alarm information through the platform alarm method. The maintenance personnel should focus on checking the equipment during the next inspection.

[0061] The system supports an automatic escalation mechanism for warning levels: if a warning level is not handled or is repeatedly triggered within 30 minutes, it will be automatically escalated to a higher warning level, up to the first warning level, to ensure that alarm information is responded to in a timely manner.

[0062] Example 6

[0063] according to Figure 1 , 2 As shown in Figures 3, 4, 5, and 6, this embodiment proposes an intelligent early warning and decision-making system for dehumidifiers. The human-computer interaction of the application module includes interconnection between a local monitoring terminal, a cloud platform, and a mobile APP, enabling comprehensive monitoring and operation and maintenance management. The specific functions are as follows:

[0064] Local monitoring terminal: Deployed in the power equipment field control room, using an industrial touch screen with a screen size of no less than 10 inches, supporting touch operation and keyboard operation, providing real-time data display (including humidity, temperature, oil level, pressure, silica gel color image, etc.), alarm status indication (different alarm levels correspond to different colored indicator lights, level 1 alarm is red, level 2 is orange, level 3 is yellow, and level 4 is blue), manual control (such as manually switching to the backup system and manually resetting the alarm), and parameter setting (such as adjusting the sampling frequency and alarm threshold). It also has local data storage function, which can store no less than one month of real-time data.

[0065] Cloud Platform: Deployed on a cloud server, using a B / S architecture, it supports access via a browser and provides functions such as data storage, data analysis, data forwarding, alarm management, and device management. Data storage uses a distributed database, supporting massive data storage and fast querying, and can store no less than 3 years of historical data and alarm records. Data analysis functions include trend analysis, anomaly analysis, and risk assessment, generating visual reports (such as daily, weekly, and monthly reports). Device management functions support adding, deleting, and modifying device information, and viewing device operating status and historical data.

[0066] Mobile App: Supports iOS and Android systems, can be downloaded and installed from the app store. It provides a device list (displays all dehumidifier devices connected to the system), real-time overview (displays the current status and key data of the devices), alarm push (receive alarm information in real time, supports voice reminders), historical query (query device historical data and alarm records), remote control (such as remote switching to backup system, remote reset of alarms), and personal center (modify personal information, set notification methods), which makes it convenient for maintenance personnel to monitor and maintain devices anytime, anywhere.

[0067] Example 7

[0068] according to Figure 1 , 2 As shown in Figures 3, 4, 5, and 6, this embodiment proposes an intelligent early warning and decision-making system for dehumidifiers. The system's workflow includes the following steps, which are executed sequentially to ensure stable system operation:

[0069] The sensors in the multi-source data acquisition module collect multi-dimensional data such as humidity inside the dehumidifier, ambient temperature, transformer oil level, internal pressure of the equipment, silica gel color image, and ambient temperature and humidity according to the preset sampling frequency;

[0070] The sensor transmits the raw data it collects to the edge gateway. The edge gateway performs preliminary verification on the raw data, removes obviously abnormal data, and caches the data.

[0071] The edge gateway uploads the pre-verified data to the data processing module of the cloud platform through communication protocols such as MQTT, HTTP, and 4G.

[0072] The data processing module preprocesses the uploaded data, including data cleaning, time-series alignment, feature engineering, and multi-source data fusion, to generate comprehensive status indicators and key feature parameters;

[0073] The data processing module transmits comprehensive status indicators and key feature parameters to the intelligent early warning decision engine;

[0074] The threshold judgment module in the intelligent early warning decision engine compares the data with the preset threshold to determine whether it is abnormal; if it is abnormal, the trend prediction module uses the LSTM model to predict the data change trend, and the risk assessment module calculates the comprehensive risk value.

[0075] The graded early warning module classifies early warning levels based on comprehensive risk values, while the decision reasoning module outputs intelligent operation and maintenance decision suggestions based on a multi-algorithm fusion strategy.

[0076] Based on the warning level and decision-making suggestions, the application module outputs alarm information and decision-making suggestions through multiple channels such as local alarms, remote notifications, mobile APP, operation and maintenance platform and data dashboard;

[0077] Maintenance personnel receive alarm information and decision suggestions through local monitoring terminals, mobile apps, or maintenance platforms, and perform corresponding maintenance procedures on the equipment.

[0078] After the maintenance personnel have completed the task, they will report the result on the terminal. After receiving the result, the system will reset the warning status and continue real-time monitoring. If the problem is not resolved within 30 minutes or the warning is triggered repeatedly, the system will automatically upgrade the warning level and repeat steps 8-9 until the problem is resolved.

[0079] Verification data 1:

[0080] Fifty transformer dehumidifiers from 10 substations were selected as test subjects. 25 units were equipped with the system of this invention (test group), 25 units used an existing single-sensor monitoring scheme (control group 1), and 25 units used traditional manual inspection methods (control group 2). The test period was 6 months, and the test environment covered different climatic conditions—high temperature, low temperature, humidity, and dryness. The test results are as follows: The average warning lead time for the test group was 72 hours, enabling the detection of dehumidifier abnormalities 3 days in advance; the average warning lead time for control group 1 was 48 hours, only enabling the detection of abnormalities 2 days in advance; control group 2 had no warning function and could only detect abnormalities during inspections, with an average detection lag of 48-72 hours. Compared to control group 1, the warning lead time for the test group was improved by 50%, and compared to control group 2, the abnormality detection lag time was significantly shortened, effectively achieving preventative maintenance. The test group had a false alarm rate of 2.3%, with only 3 false alarms, all caused by temporary data fluctuations under extreme conditions, and the system automatically corrected these errors. Control group 1 had a false alarm rate of 8.5%, with 11 false alarms, mainly due to a single monitoring dimension and low data accuracy. Control group 2 had no clear concept of false alarms, but there were 7 instances of human misjudgment (mistaking unsaturated silica gel for saturation, or vice versa), resulting in a misjudgment rate of 28%. The test group's false alarm rate was 73% lower than control group 1 and 92% lower than control group 2, significantly improving warning accuracy. The test group had a missed alarm rate of 1.1%, with only 1 missed alarm, caused by a temporary sensor malfunction, which the system promptly detected and corrected in subsequent sampling. Control group 1 had a missed alarm rate of 4.2%, with 5 missed alarms, mainly due to data loss or deviation from a single sensor. Control group 2 had a missed alarm rate of 12.8%, with 16 missed alarms, all caused by the dehumidifier rapidly saturating without being detected during the inspection cycle. The test group's missed alarm rate was 74% lower than control group 1 and 91% lower than control group 2, ensuring no abnormal conditions were missed. The average monthly maintenance time per dehumidifier in the test group was 0.5 hours, mainly for troubleshooting and regular inspections; the average monthly maintenance time per dehumidifier in control group 1 was 1.2 hours, mainly for handling false alarms and regular inspections; and the average monthly maintenance time per dehumidifier in control group 2 was 3.8 hours, mainly for manual inspections and silica gel replacement. The test group showed a 58% improvement in maintenance efficiency compared to control group 1 and an 87% improvement compared to control group 2, significantly reducing maintenance workload. During the 6-month test period, the test group experienced only one dehumidifier-related failure (4% failure rate); control group 1 experienced 6 failures (24% failure rate); and control group 2 experienced 12 failures (48% failure rate). The test group showed an 83% reduction in failure rate compared to control group 1 and a 92% reduction compared to control group 2, effectively ensuring the safe and stable operation of electrical equipment.The test group operated stably under various environmental conditions, including high temperature (above 40℃), low temperature (below -20℃), humid (ambient humidity above 85%), and dry (ambient humidity below 30%), with no significant decrease in early warning accuracy and data reliability. Control group 1, under extreme conditions, showed significant data deviation and an over 30% decrease in early warning accuracy. Control group 2, under extreme conditions, experienced increased difficulty in manual inspection, leading to a substantial increase in false alarm and missed alarm rates. The verification results demonstrate that this invention effectively addresses the shortcomings of existing technologies. Its monitoring accuracy, early warning timeliness, environmental adaptability, and operational efficiency are significantly superior to existing solutions, enabling precise monitoring, proactive prevention, and intelligent operation and maintenance of the dehumidifier's status.

[0081] Verification data 2:

[0082] exist Figure 5 The data provides a risk factor weight distribution chart. Based on 1000 sets of measured data, the SHAP analysis results show that the weights are: humidity change rate 0.35, color saturation 0.28, temperature compensation coefficient 0.18, historical trend 0.12, and environmental factors 0.07.

[0083] exist Figure 3 and attached Figure 4 Detailed comparative experimental results are provided. The experiments compared the performance differences between the system of this invention and the single threshold method and the fixed period method. The results show that the average early warning time of the system of this invention is 72 hours, which is 50% higher than that of the single threshold method (48 hours); the false alarm rate is 2.3%, which is 73% lower than that of the fixed period method (8.5%); and the false alarm rate is 1.1%, which is 74% lower than that of the single threshold method (4.2%). Figure 6 The comprehensive performance radar chart further demonstrates the overall advantages of the system of the present invention in five dimensions: accuracy, timeliness, stability, adaptability, and economy.

[0084] This intelligent early warning and decision-making system for dehumidifiers employs multi-source data fusion, integrating multi-dimensional data such as internal humidity of the dehumidifier, ambient temperature and humidity, transformer oil level, internal pressure of the equipment, and silica gel color images. Combined with fusion algorithms such as Kalman filtering and weighted averaging, it effectively solves the problems of single monitoring dimensions and low data accuracy in existing technologies. This significantly improves the accuracy and reliability of dehumidifier status monitoring, avoiding misjudgments caused by single data biases, and providing high-quality data support for subsequent early warning and decision-making. Simultaneously, through environmental data correction, the system can adapt to different environmental conditions, expanding its applicability. This invention constructs an integrated intelligent early warning and decision-making engine, integrating threshold judgment, LSTM time-series prediction, risk assessment, and multi-algorithm fusion decision-making functions. It realizes a shift from passively responding to faults to proactively preventing faults. The LSTM time-series prediction model can accurately predict the trend of dehumidifier status changes, discover potential fault hazards in advance, and provide a scientific basis for preventative maintenance. Through comprehensive evaluation of multiple risk factors and multi-algorithm fusion decision-making, it ensures the rationality and accuracy of decision results, solving the problems of existing technologies' inability to predict trends and lack of intelligent decision-making mechanisms, thus reducing the failure rate of power equipment. This invention employs a tiered early warning mechanism coupled with multi-channel alarm notifications and remote monitoring functions to achieve precise and intelligent operation and maintenance management of dehumidifiers. Four early warning levels correspond to different response strategies, enabling targeted handling measures based on the severity of the fault, avoiding excessive or untimely maintenance. Multi-channel alarm notifications ensure that alarm information is promptly delivered to relevant responsible persons, preventing fault escalation due to information delays. Remote monitoring and mobile management functions enable monitoring and operation and maintenance scheduling of dehumidifier status anytime, anywhere, significantly reducing manual inspection workload, improving operation and maintenance efficiency, and promoting the development of power equipment operation and maintenance towards less-manned and unmanned operations.

[0085] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A dehumidifier intelligent early warning decision system, comprising a multi-source data acquisition module, a data transmission module, a data processing module, an intelligent early warning decision engine, an early warning classification module, and an application module, characterized in that: The multi-source data acquisition module is used to collect multi-dimensional data of the dehumidifier and its surrounding environment; the data transmission module is used to transmit the collected data to the cloud platform, supporting multiple communication protocols; the data processing module is used to preprocess and fuse the multi-source data to generate comprehensive status indicators. The intelligent early warning decision engine is used to realize data anomaly judgment, trend prediction, risk assessment and intelligent decision-making; the early warning classification module is used to classify early warning levels and match response strategies; the application module is used to realize multi-channel alarm notification and remote monitoring management. The modules work together to realize real-time monitoring, early warning and operation and maintenance decision-making of the dehumidifier status.

2. The intelligent early warning and decision-making system for dehumidifiers according to claim 1, characterized in that: The multi-source data acquisition module includes a humidity sensor, a temperature sensor, an oil level sensor, a pressure sensor, a vision sensor, and an environmental sensor. The humidity sensor detects the humidity inside the dehumidifier, with a range of 0-100%RH and an accuracy of ±2%RH. The temperature sensor detects the ambient temperature, with a range of... The accuracy is ±0.5℃; the oil level sensor is used to detect changes in transformer oil level, with a range of 0-100mm and an accuracy of ±1mm; the pressure sensor is used to detect internal pressure of the equipment, with a range of 0-1MPa and an accuracy of ±0.01MPa; the vision sensor is used to acquire silicone color images, with a resolution of not less than 1080P; the environmental sensor is used to detect ambient temperature and humidity, with a temperature range of... The humidity measurement range is 0-100%RH, with accuracies of ±0.5℃ and ±2%RH, respectively.

3. The intelligent early warning and decision-making system for dehumidifiers according to claim 2, characterized in that: In the multi-source data acquisition module, the sampling frequency of each sensor is as follows: humidity data and temperature data once per minute, oil level data once per 5 minutes, pressure data once per 10 minutes, image data once per hour, and ambient temperature and humidity data once per 3 minutes. The data transmission module uses an edge gateway as a data relay device, supports MQTT, HTTP, and 4G communication protocols, and has a data caching function. When the network is interrupted, it caches data for no less than 72 hours and automatically uploads the data after the network is restored.

4. The intelligent early warning and decision-making system for dehumidifiers according to claim 1, characterized in that: The preprocessing of the data processing module includes data cleaning, time-series alignment, feature engineering, and multi-source data fusion; the data cleaning adopts... The criteria for removing outlier data and using linear interpolation to supplement missing data are as follows: the temporal alignment is based on the sampling time of humidity data, and interpolation or downsampling is performed on data with different sampling frequencies; the feature engineering is used to extract key feature parameters such as humidity change rate, silica gel color saturation, and temperature compensation coefficient; the multi-source data fusion adopts Kalman filtering, weighted averaging, and outlier removal algorithms, and the fusion weight is determined based on the SHAP analysis results of 1000 sets of measured data.

5. The intelligent early warning and decision-making system for dehumidifiers according to claim 4, characterized in that: The state equation for the Kalman filter is: ， The observation equation is: ； Where x(k) is the fused data at time k, A is the state transition matrix (1.02), B is the control matrix (0.05), u(k) is the sensor correction coefficient, w(k) is the process noise (N(0,0.01)), z(k) is the original sensor data at time k, H is the observation matrix (1), and v(k) is the observation noise (N(0,0.02)). The formula for the weighted average is: S=0.35×H+0.28×C+0.18×T+0.12×O+0.07×P, Where S is the comprehensive state index, H is the humidity normalized value, C is the silica gel color saturation, T is the temperature compensation value, O is the oil level change rate, and P is the pressure normalized value.

6. The intelligent early warning and decision-making system for dehumidifiers according to claim 1, characterized in that: The intelligent early warning decision engine includes a threshold judgment module, a trend prediction module, a risk assessment module, and a decision reasoning module. The threshold judgment module compares the fused data and feature parameters with a preset threshold to determine if the data is abnormal. The trend prediction module uses an LSTM time-series prediction model, taking humidity data from the past 72 hours as input, to predict the humidity change trend for the next 24 hours. The model has 5 neurons in the input layer, 2 hidden layers with 64 LSTM units per layer, and 1 neuron in the output layer. It uses 32 training batches and 100 training iterations. The risk assessment module calculates a comprehensive risk value based on multiple risk factors. The risk calculation formula is: R=0.35×H+0.28×C+0.18×T+0.12×Ht+0.07×E, Wherein, R is the comprehensive risk value, unitless, ranging from 0 to 100, used to quantify the operational risk level of the dehumidifier; H is the humidity normalization value, ranging from 0 to 1, obtained by normalizing the humidity data inside the dehumidifier, reflecting the internal humidity level of the dehumidifier; C is the silica gel color saturation, ranging from 0 to 1, extracted by recognizing the silica gel color image collected by the visual sensor, reflecting the moisture absorption degree of the silica gel; T is the temperature compensation value, ranging from 0 to 1, obtained by correcting the humidity data according to the ambient temperature, used to eliminate the influence of temperature on the humidity monitoring results; Ht is the historical trend factor, ranging from 0 to 1, calculated based on the historical operating data and status change trend of the dehumidifier, reflecting the long-term change trend of the dehumidifier's status; E is the environmental correction factor, ranging from 0 to 1, calculated based on surrounding environmental factors such as ambient temperature, humidity, and dust, used to adapt to risk assessment under different environmental conditions.

7. The intelligent early warning and decision-making system for dehumidifiers according to claim 6, characterized in that: The decision-making reasoning module adopts a multi-algorithm fusion strategy of rule-based reasoning, fuzzy logic, neural networks, and expert systems, with adaptive weight adjustment. The decision rules are as follows: when R≥80, it outputs an immediate shutdown and maintenance suggestion; when 50≤R<80, it outputs a switch to a backup system and a planned maintenance suggestion; when 30≤R<50, it outputs an enhanced monitoring and key attention suggestion; when 10≤R<30, it outputs a log recording and regular inspection suggestion; and when R<10, it outputs a normal monitoring suggestion. Multiple risk factors include the degree of humidity exceeding the standard, the magnitude of temperature anomalies, the rate of oil level change, the duration, and the correlation with historical faults.

8. The intelligent early warning and decision-making system for dehumidifiers according to claim 7, characterized in that: The warning classification module divides the warning levels into four levels, with clear judgment criteria and response strategies for each level: Level 1 warning R≥80: Humidity >95%RH or equipment failure is imminent; the response strategy is immediate shutdown, manual inspection, and notification of responsible personnel through multiple channels. Level 2 warning 50≤R<80: Humidity >85%RH or abnormal temperature; the response strategy is automatic switching to the backup system and notification of maintenance personnel to handle the situation within 4 hours. Level 3 warning 30≤R<50: Humidity >75%RH or worsening trend; the response strategy is to increase the sampling frequency and notify maintenance personnel to check within 24 hours. Level 4 warning 10≤R<30: Humidity >65%RH or slight abnormality; the response strategy is to log and conduct routine inspections. The system supports automatic escalation if a warning is not handled within 30 minutes or is repeatedly triggered.

9. The intelligent early warning and decision-making system for dehumidifiers according to claim 1, characterized in that: The application modules include a local monitoring terminal, a cloud platform, a mobile app, and a data dashboard. The local monitoring terminal provides real-time data display, alarm indication, manual control, and parameter setting functions, and stores at least one month of local data. The cloud platform supports data storage, analysis, forwarding, and device management, and generates visual reports. The mobile app supports device lists, real-time overviews, alarm push notifications, historical queries, and remote control. The data dashboard enables centralized display of multiple device statuses and alarm statistics.

10. The intelligent early warning and decision-making system for dehumidifiers according to claim 1, characterized in that: The system adopts a five-layer architecture, from bottom to top: data acquisition layer, based on multi-source data acquisition module; data transmission layer, based on data transmission module; data processing layer, based on data processing module; intelligent decision-making layer, based on intelligent early warning decision engine and early warning classification module; application layer, based on application module. The system workflow includes data acquisition, preliminary verification, data upload, preprocessing and fusion, early warning decision, alarm output, operation and maintenance processing, and status reset.

Images