A condensation prevention and control method and system for a power cabinet type device

Abstract

This invention discloses a method and system for preventing condensation in power cabinet equipment. The method includes: real-time acquisition of the time-series signal matrix of a condensation acquisition board, followed by wavelet denoising and normalization preprocessing; multi-dimensional feature extraction of the preprocessed signal to construct a feature vector sequence; inputting the feature vector sequence into a pre-trained CNN-LSTM hybrid deep learning model to predict the number of condensation drops in real time; comparing the predicted value with a warning threshold to determine whether to generate a warning signal and trigger the on-site dehumidification device; and activating an intelligent dehumidification system to condense the humid air inside the cabinet into water and discharge it outside the cabinet. This invention also discloses a condensation prevention and control system, including an intelligent sensing module, an intelligent linkage control module, a condensation dehumidification module, and a communication module. This invention directly predicts the number of condensation drops using a deep learning model, combined with active condensation dehumidification technology, achieving accurate quantification and efficient prevention and control of condensation inside power cabinets.

Description

Technical Field

[0001] This invention relates to a method and system for condensation control, and more particularly to a method and system for condensation control in electrical cabinet equipment. Background Technology

[0002] In power systems, distribution boxes and switchgear are core equipment for ensuring power supply reliability, and the stability of their internal microenvironment is crucial for safe operation. Especially in outdoor ring network units and low-voltage branch boxes, condensation is prone to occur due to factors such as diurnal temperature variations, differences in enclosure airtightness, and internal and external heat and moisture exchange. Condensation not only weakens electrical insulation performance but can also induce short-circuit faults, malfunctions of protection devices, accelerate the aging of insulation materials, and even cause insulation breakdown, seriously threatening the safe operation of the power grid.

[0003] Current condensation control technologies have the following limitations:

[0004] The methods for identifying condensation are limited: Traditional methods mainly rely on monitoring the humidity of the air inside the enclosure and controlling the start and stop of the dehumidifier based on a preset threshold. This method can only reflect the humidity content of the air and cannot quantitatively assess the degree of condensation formation (such as the amount of condensation per unit time), making it difficult to achieve accurate early warning. Furthermore, it is susceptible to environmental interference, limiting the reliability of monitoring.

[0005] Dehumidification technology has shortcomings: existing decondensation solutions mostly use electric heating, forced ventilation, or adsorbent treatment. Among them, electric heating can only increase the air saturation temperature and does not fundamentally remove moisture; forced ventilation has limited effectiveness in high humidity environments and is prone to introducing external pollutants; although adsorption dehumidification can absorb some moisture, its water absorption capacity is limited, the regeneration process is difficult, it is difficult to achieve long-term automatic operation, and the burden of later maintenance is heavy.

[0006] In summary, existing technologies still have shortcomings in quantitative condensation sensing and efficient dehumidification, making it difficult to meet the needs for accurate monitoring and effective control of condensation under varying operating conditions. Summary of the Invention

[0007] Purpose of the invention: The purpose of this invention is to propose a method and system for preventing condensation in electrical cabinet equipment, providing an effective monitoring means and condensation removal measures to overcome the limitations of current condensation prevention systems.

[0008] Technical solution: This invention includes the following steps:

[0009] S1: Real-time acquisition of electrical characteristic parameters of the microelectrode array on the condensation acquisition board, generation of time-series signal matrix, and wavelet denoising and normalization processing of the time-series signal matrix;

[0010] S2: Perform multi-dimensional feature extraction on the denoised and normalized time-series signal matrix to construct a feature vector sequence;

[0011] S3: Input the feature vector sequence into the pre-trained CNN-LSTM hybrid deep learning model. The CNN-LSTM hybrid deep learning model extracts the spatial distribution features of the condensation collection plate signal through a convolutional neural network, models the temporal dynamic process of condensation formation and evolution through a long short-term memory network, and outputs the estimated number of condensation droplets currently attached to the collection plate in real time.

[0012] S4: Compare the estimated number of condensation drops with the preset warning threshold. If the threshold is exceeded, generate a condensation exceeding warning signal and upload it to trigger the on-site dehumidification device.

[0013] S5: Activate the intelligent dehumidification system to condense the humid air inside the cabinet into water and discharge it outside the cabinet, achieving air dehumidification and condensation control.

[0014] In step S1, the microelectrode array is distributed in a grid pattern, and each microelectrode is connected to a different channel of the high-precision impedance / capacitance sensing circuit to form a multi-channel synchronous acquisition architecture.

[0015] The timing signal matrix is ​​a multi-channel signal matrix in which the impedance or capacitance values ​​between each microelectrode change over time.

[0016] The multi-dimensional features in step S2 include time-domain features, frequency-domain features, and spatial features. The time-domain features include mean, variance, peak factor, root mean square, and kurtosis factor. The frequency-domain features are obtained by performing a fast Fourier transform on the time-series signal to extract the main frequency and frequency band energy ratio of the power spectral density. The spatial features include the correlation coefficient matrix of adjacent electrode signals and the spatial gradient distribution features.

[0017] The CNN-LSTM hybrid deep learning model in step S3 includes:

[0018] The CNN part is used to reconstruct the feature vector of each time step into a two-dimensional feature map, and extract the spatial aggregation and dispersion distribution patterns of condensation droplets on the condensation collection plate through convolutional layers and pooling layers.

[0019] The LSTM part is used to input the spatial feature sequences of multiple consecutive time steps into the LSTM layer, and to model the temporal dynamic process of condensation formation, growth, merging and dripping through forget gate, input gate, output gate and cell state update mechanism;

[0020] The fully connected layer is used to flatten the temporal features output by the LSTM and then perform a nonlinear transformation to regress the output to predict the number of dewdrops.

[0021] The training process of the CNN-LSTM hybrid deep learning model includes: supervised training using a historical dataset labeled with real dew drop counts. The real dew drop counts are obtained by synchronously acquiring images of the dew collection board with a high-definition macro camera and manually labeling them. The loss function is mean squared error (MSE), the optimizer is Adam, and the model parameters are iteratively optimized through backpropagation.

[0022] A condensation prevention and control system for electrical cabinet equipment includes:

[0023] Intelligent sensing module: including condensation acquisition board and impedance / capacitance sensing circuit connected thereto, used to acquire and output a time-series signal matrix reflecting the condensation state in real time;

[0024] Intelligent linkage control module: including microcontroller, used to reduce noise, normalize, and extract multi-dimensional features from the signal output by the sensing module, run CNN-LSTM deep learning model to predict the number of dew drops in real time, and generate control commands based on the comparison between the predicted value and the warning threshold;

[0025] Condensation dehumidification module: including air supply system, semiconductor refrigerator, heating circuit and drainage pipe, used to condense the humid air in the cabinet into water and discharge it outside the cabinet;

[0026] Communication module: Used to upload condensation early warning information to a remote monitoring center or cloud platform.

[0027] The CNN-LSTM deep learning model deployed in the intelligent linkage control module is embedded in the microcontroller and supports online inference. It is used to predict the number of condensation droplets and make threshold judgments in real time.

[0028] The heating circuit of the condensation dehumidification module is linked to the semiconductor cooler for control. When the ambient temperature is lower than the preset threshold, the heating circuit is activated to keep the surface temperature of the condenser plate above 0°C and prevent the condensate from freezing.

[0029] The condensation collection plate is installed on the inner surface of the top of the cabinet, and the condensation collection plate is made of copper-clad laminate.

[0030] Beneficial effects: This invention has the following advantages:

[0031] (1) The present invention can realize accurate quantification and intelligent prediction of condensation: the electrical characteristic changes of the condensation collection board are directly sensed by a high-precision impedance / capacitance sensing circuit, and the spatial distribution features and temporal evolution features of condensation are extracted by combining the CNN-LSTM deep learning model. The number of condensation drops can be directly predicted, which greatly improves the accuracy of condensation state perception. Moreover, the model can be embedded in the microcontroller to realize real-time online reasoning.

[0032] (2) This invention can fundamentally eliminate the hidden danger of condensation: by adopting condensation dehumidification technology, the humid air is condensed into water by a semiconductor cooler and actively discharged outside the box / cabinet through the drainage pipe, rather than relying solely on heating or ventilation to temporarily change the air state, thereby achieving the complete removal of humid air inside the box / cabinet, effectively preventing the recurrence of condensation, and ensuring the insulation performance and safe operation of the equipment.

[0033] (3) This invention integrates sensing, analysis, decision-making, execution and communication functions, and can realize full-process automation and intelligence from condensation monitoring to active prevention and control, greatly reducing manual intervention and significantly improving the efficiency and reliability of power equipment operation and maintenance. Attached Figure Description

[0034] Figure 1 This is a flowchart of the present invention;

[0035] Figure 2 This is a schematic diagram of the system structure of the present invention;

[0036] Figure 3 This is a system block diagram of the present invention. Detailed Implementation

[0037] The invention will now be further described with reference to the accompanying drawings.

[0038] Example 1

[0039] like Figure 1 As shown, the condensation prevention method and system for electrical cabinet equipment in this embodiment includes the following steps:

[0040] S1: Real-time acquisition and output of a timing signal matrix reflecting the changes in the electrical characteristics of the entire condensation acquisition plate. The original time-series signal matrix is ​​subjected to wavelet denoising and normalization, and the signal values ​​are mapped to... Interval.

[0041] The sensing module acquires the impedance / capacitance signal of the condensation collection plate at a sampling frequency of 1Hz, forming a time-series matrix X(t). The original time-series signal matrix is ​​denoised using the db4 wavelet basis function. The wavelet decomposition has three levels, and the soft threshold coefficient is set to 0.1 to eliminate signal noise caused by environmental electromagnetic interference and temperature fluctuations. The denoised signal is then normalized using a maximum-minimum method to map the signal values. For intervals, the normalization formula is:

[0042]

[0043] in This is the signal value after noise reduction. , These are the maximum and minimum values ​​of the corresponding signal sequence, respectively, to eliminate the interference of environmental temperature and humidity fluctuations on signal characteristics.

[0044] S2: Perform multi-dimensional feature extraction on the denoised and normalized time-series signal matrix from step 1 to construct a feature vector sequence. ,in, The total number of feature dimensions is 45. The feature vector sequence forms a model input sample window every 10 seconds, with a window sliding step of 1 second. The extracted features include:

[0045] S21. Temporal characteristics

[0046] Five time-domain features of the signal are extracted: mean, variance, peak factor, root mean square (RMS), and kurtosis factor. The peak factor is calculated as the ratio of the signal peak value to the RMS value. The kurtosis factor reflects the steepness of the signal probability density distribution. The time-domain features reflect the time-domain statistical regularity of the sensing signal during the condensation formation process.

[0047] S22. Frequency Domain Characteristics

[0048] Fast Fourier Transform (FFT) was performed on the time-series signal to extract the top three dominant frequencies and three frequency band energy ratios of the power spectral density, resulting in a total of six frequency domain features. The frequency bands were divided into three intervals: 0.1-0.5Hz, 0.5-2Hz, and 2-5Hz. The frequency band energy ratio was the ratio of the power spectral density integral within each frequency band to the total power spectral density integral. The frequency domain features reflected the frequency distribution characteristics of the sensing signal during the condensation evolution process.

[0049] S23. Spatial Features

[0050] The correlation coefficient matrix between adjacent electrodes in 8×8 dimensions and the spatial gradient distribution features calculated by the Sobel operator were extracted, resulting in a total of 34 spatial features. The correlation coefficient matrix reflects the spatial correlation of the sensing signals of different electrodes on the condensation acquisition plate and characterizes the spatial aggregation state of condensation droplets. The Sobel operator uses a 3×3 template to calculate the spatial gradient of the signal along the x and y directions, respectively, characterizing the spatial distribution gradient of condensation droplets on the acquisition plate. The spatial features reflect the spatial distribution pattern of condensation.

[0051] S3: Input the feature vector sequence constructed in step S2 into the pre-trained CNN-LSTM hybrid deep learning model. The model outputs the estimated number of dew drops attached to the dew collection plate in real time. This model integrates the spatial feature extraction capabilities of Convolutional Neural Networks (CNNs) and the temporal dynamic modeling capabilities of Long Short-Term Memory Networks (LSTMs). The training and prediction process of the model is as follows:

[0052] S31. Model Structure Parameters

[0053] The CNN part consists of 2 convolutional layers and 2 max pooling layers. The kernel size of the convolutional layers is 3×3. The first convolutional layer has 32 kernels and the second convolutional layer has 64 kernels. The pooling kernel size is 2×2 with a stride of 2. The activation function used is ReLU.

[0054] LSTM section: A single LSTM layer is used with 64 hidden units. The forget gate, input gate, and output gate all use the sigmoid activation function. The candidate cell state uses the tanh activation function. The sequence input length is 10 (corresponding to a 10-second sample window).

[0055] Fully connected layer: It contains two fully connected layers. The first fully connected layer has 32 neurons and uses the ReLU function as the activation function. The second fully connected layer is a linear output layer with 1 neuron and outputs the predicted value of the number of dew drops.

[0056] Model optimizer: Adam optimizer is used, with a learning rate of 0.001, a weight decay coefficient of 0.0001, and a training batch size of... The value is 32, and the number of iteration training rounds is 200.

[0057] S32. Spatial Feature Extraction (CNN Part)

[0058] The feature vector sequence at each time step is reconstructed into a 9×5 2D feature map (simulating the planar distribution of the condensation collection plate), and input into the CNN part for local spatial feature extraction; The formula for layer convolution is:

[0059]

[0060] in, For the first Layer convolution kernel weights, For the first Layer bias, This represents a two-dimensional convolution operation. It is the ReLU activation function. This is the output of the previous layer. For the first layer, This is the reconstructed two-dimensional feature map;

[0061] A convolutional layer is followed by a max pooling layer for feature dimensionality reduction, preserving key spatial features. After two convolutional-pooling layers, a high-order spatial feature representation is obtained. It captures the spatial aggregation and dispersion patterns of condensed water droplets on the collection plate.

[0062] S33. Temporal Dynamic Modeling (LSTM Part)

[0063] The high-order spatial feature sequence of 10 consecutive time steps Inputting into an LSTM layer models the temporal dynamics of condensation formation, growth, merging, or dripping; LSTM cells at time steps The core calculation formula is as follows:

[0064] Forgotten Gate:

[0065] Input Gate:

[0066] Candidate cell status:

[0067] Cell status update:

[0068] Output gate:

[0069] Hidden output:

[0070] in, The spatial features input at the current time step, This is the hidden state from the previous moment. This represents the cell state at the previous moment. For the trainable weights corresponding to the gating, This is the trainable bias corresponding to the gating. This represents element-wise multiplication. The LSTM layer uses the sigmoid activation function; its final output is a temporal feature representation. It contains time-dependent information on the evolution of condensation.

[0071] S34. Dewdrop Count Regression Prediction (Fully Connected Layer)

[0072] Temporal features output by LSTM layer Flattened into a one-dimensional vector, the input is fed into two fully connected layers for nonlinear transformation. Finally, a linear output layer outputs the predicted dewdrop count, with the prediction formula as follows:

[0073]

[0074] in, The output features of the last fully connected layer, This is the output layer weight matrix. This is the output layer bias.

[0075] S35. Model Training

[0076] The CNN-LSTM model was trained under supervision using a historical dataset with real labels, and the dataset contained the actual number of dew drops. Images of the condensation collection plate were simultaneously acquired using a 2-megapixel high-definition macro camera and manually labeled. The image acquisition frame rate was 1 frame / second, synchronized with the sensor signal acquisition frequency. The loss function used for model training was mean squared error (MSE), calculated using the following formula:

[0077]

[0078] By minimizing the loss function using the backpropagation algorithm combined with the Adam optimizer, all trainable parameters of the model are iteratively optimized until the model converges. After training, the root mean square error (RMSE) of the model test set is ≤2.5 and the mean absolute error (MAE) is ≤1.8, ensuring the accuracy of condensation droplet prediction.

[0079] S4: The predicted number of condensation drops With the preset warning threshold Compare. If If this occurs, a warning signal for excessive condensation is generated and uploaded to the monitoring center via the communication module, triggering the on-site dehumidification device. Specifically:

[0080] Condensation warning threshold Set to 10 drops (the effective area of ​​the condensation collection plate is 10cm×10cm);

[0081] like If the system determines that the condensation exceeds the standard, it will immediately generate a condensation exceeding the standard warning signal. The warning signal will be uploaded to the remote monitoring center through the RS485 communication module at a baud rate of 9600bps. At the same time, it will trigger the semiconductor cooling dehumidification device inside the power box / cabinet to start. The start-up delay of the dehumidification device is ≤0.5 seconds.

[0082] like The system maintains normal monitoring status, continues to collect signals in real time, and predicts the number of condensation droplets.

[0083] S5: Executes the intelligent dehumidification system. First, a heating circuit is created to ensure that the ambient temperature of the condenser plate is always kept above 0℃, preventing the condensate from freezing due to low temperatures. At this time, the intelligent dehumidification system actively draws humid air from the space into the dehumidification duct under the action of the fan. The water vapor in the air is condensed into water after passing through the semiconductor refrigeration mechanism. After the condenser plate collects a certain amount of water, the water flows out through the water pipe and is discharged outside the cabinet, achieving the purpose of air dehumidification.

[0084] Example 2

[0085] like Figure 2As shown, the condensation control system for the power cabinet-type equipment in this embodiment includes a cabinet 1, a condensation collection board 2, a temperature sensor 3, a control module, a communication module 5, and a condensation dehumidification module. The condensation collection board 2 is installed on the inner top surface of the cabinet 1. The condensation collection board 2 is made of copper-clad laminate, and multiple microelectrode arrays are evenly distributed on its surface. The microelectrode arrays are distributed in a grid pattern, and each microelectrode is connected to a different channel of a high-precision impedance / capacitance sensing circuit, forming a multi-channel synchronous acquisition architecture. The high-precision impedance / capacitance sensing module is connected to each microelectrode through a flexible cable, and collects the changes in electrical parameters between each electrode in real time, outputting a multi-channel timing signal matrix. The temperature sensor 3 is a high-precision temperature sensor used to collect the temperature inside the cabinet 1. The temperature sensor 3 is connected to the condensation collection board 2 through a 485 bus, and transmits the collected information to the control module for analysis. The temperature sensor 3 and the condensation collection board 2 together constitute an intelligent monitoring and sensing module.

[0086] The control module is a microcontroller 4, installed on the inner side wall of the cabinet. It receives information from the temperature sensor 3 and the condensation collection board 2. This embedded microcontroller integrates signal conditioning circuitry and a communication interface. The control module has a pre-trained CNN-LSTM model built-in, supporting online inference and parameter updates. The communication module 5, integrated on the microcontroller 4, can upload condensation warning information, predicted drop counts, temperature and humidity data to a cloud platform, supporting remote monitoring and historical data review.

[0087] The condensation dehumidification module includes a heating device 6, a dehumidifying fan 7, a condensing device 8, and a drain pipe 9. The heating device 6 is used to create a heating circuit to prevent condensate from freezing due to low temperature. The dehumidifying fan 7 is mainly used to absorb water vapor in the air. The condensing device 8 adopts the thermoelectric refrigeration principle based on the Seebeck effect and Peltier effect. It applies direct current to the semiconductor device for refrigeration, thereby cooling water molecules in the humid air to below the dew point and then condensing them on the condensing plate. The drain pipe 9 is used to drain the water collected by the condensing plate to the outside of the cabinet 1.

[0088] Example 3

[0089] like Figure 3 As shown, the condensation prevention system for the power distribution cabinet equipment in this embodiment includes:

[0090] Intelligent Sensing Module: Used for real-time monitoring of condensation levels inside the enclosure. This module uses a condensation acquisition board as its core and integrates a high-precision impedance / capacitance detection circuit. When condensation occurs on the surface of the acquisition board, its electrical parameters change, which are then converted into a timing signal matrix output by the detection circuit, providing fundamental data support for subsequent analysis.

[0091] Intelligent linkage control module: Responsible for data processing, status recognition, and control decision-making. Using a microcontroller as its core, it primarily implements the following functions:

[0092] Signal processing and feature extraction: Noise reduction, normalization transformation and multi-dimensional feature extraction are performed on the raw signal output by the sensing module.

[0093] Condensation Quantification and Trend Prediction: Based on a pre-trained CNN-LSTM deep learning model, the extracted features are used to quantitatively predict the amount of condensation per unit time.

[0094] Intelligent judgment and execution decision: The predicted condensation amount and real-time temperature and humidity data are compared with the set threshold. When the threshold is exceeded, the control command is triggered to start the condensation dehumidification module or generate alarm information and upload it.

[0095] Condensation dehumidification module: Used to actively reduce the humidity of the air inside the chamber, it consists of an air supply system, a semiconductor refrigeration unit, a heating component, and a drainage channel. The air supply system introduces high-humidity air from the sealed space into the dehumidification duct; the semiconductor refrigeration unit condenses the air entering the duct; the heating component maintains the surface temperature of the condenser plate above 0°C to prevent condensate from freezing; and the drainage channel directs the condensed moisture to the outside of the chamber, achieving continuous dehumidification.

[0096] Communication module: Using 4G wireless communication, it is responsible for uploading condensation anomaly information and equipment status data to the remote monitoring center or cloud platform, so that maintenance personnel can respond and handle them in a timely manner.

[0097] Example 4

[0098] To verify the anti-condensation effect of the developed device, a comparative experiment was designed in this embodiment. Two cable distribution boxes of identical specifications and environments were used as simulation boxes. Under the same high-humidity environment (initial temperature 28℃, humidity 85%RH), one box was equipped with and activated with the dehumidification and anti-condensation device (experimental group), while the other box was left untreated (control group). Subsequently, conditions for condensation generation were created by creating a temperature difference, and the internal conditions of the two boxes were observed and recorded.

[0099] Group 1 (control group, no dehumidification device used): The humidity inside the chamber was consistently maintained at a high level of 85% RH. When the ambient temperature inside the chamber dropped slightly due to changes in experimental conditions, a large amount of condensation quickly appeared on the inner wall of the chamber and persisted. This indicates that in a humid environment, conventional chambers cannot prevent condensation from occurring.

[0100] Group 2 (experimental group, using a dehumidifier): After the device was powered on, the humidity inside the chamber began to decrease steadily within a short period of time. Throughout the experiment, although the external conditions were exactly the same as the control group, the continuous operation of the device kept the humidity inside the chamber at a low level of around 60%RH, and the inner wall of the chamber remained dry without any condensation.

[0101] Table 1 Comparison of Experimental Results

[0102]

[0103] As shown in Table 1, this invention can effectively control condensation in power distribution boxes / cabinets, keeping the condensation level within a small range without affecting equipment operation. Therefore, for situations where condensation easily occurs on the inner walls of boxes / cabinets, installing this device is beneficial for long-term reliable operation and improves the safety of equipment operation.

Claims

1. A method for preventing condensation in electrical control cabinet equipment, characterized in that, Includes the following steps: S1: Real-time acquisition of electrical characteristic parameters of the microelectrode array on the condensation acquisition board, generation of time-series signal matrix, and wavelet denoising and normalization processing of the time-series signal matrix; S2: Perform multi-dimensional feature extraction on the denoised and normalized time-series signal matrix to construct a feature vector sequence; S3: Input the feature vector sequence into the pre-trained CNN-LSTM hybrid deep learning model. The CNN-LSTM hybrid deep learning model extracts the spatial distribution features of the condensation collection plate signal through a convolutional neural network, models the temporal dynamic process of condensation formation and evolution through a long short-term memory network, and outputs the estimated number of condensation droplets currently attached to the collection plate in real time. S4: Compare the estimated number of condensation drops with the preset warning threshold. If the threshold is exceeded, generate a condensation exceeding warning signal and upload it to trigger the on-site dehumidification device. S5: Activate the intelligent dehumidification system to condense the humid air inside the cabinet into water and discharge it outside the cabinet, achieving air dehumidification and condensation control.

2. The method for preventing condensation in electrical cabinet-type equipment according to claim 1, characterized in that, In step S1, the microelectrode array is distributed in a grid pattern, and each microelectrode is connected to a different channel of the high-precision impedance / capacitance sensing circuit to form a multi-channel synchronous acquisition architecture.

3. The method for preventing condensation in electrical cabinet-type equipment according to claim 1, characterized in that, The timing signal matrix is ​​a multi-channel signal matrix in which the impedance or capacitance values ​​between each microelectrode change over time.

4. The method for preventing condensation in electrical cabinet-type equipment according to claim 1, characterized in that, The multi-dimensional features in step S2 include time-domain features, frequency-domain features, and spatial features. The time-domain features include mean, variance, peak factor, root mean square, and kurtosis factor. The frequency-domain features are obtained by performing a fast Fourier transform on the time-series signal to extract the main frequency and frequency band energy ratio of the power spectral density. The spatial features include the correlation coefficient matrix of adjacent electrode signals and the spatial gradient distribution features.

5. The method for preventing condensation in electrical cabinet-type equipment according to claim 1, characterized in that, The CNN-LSTM hybrid deep learning model in step S3 includes: The CNN part is used to reconstruct the feature vector of each time step into a two-dimensional feature map, and extract the spatial aggregation and dispersion distribution patterns of condensation droplets on the condensation collection plate through convolutional layers and pooling layers. The LSTM part is used to input the spatial feature sequences of multiple consecutive time steps into the LSTM layer, and to model the temporal dynamic process of condensation formation, growth, merging and dripping through forget gate, input gate, output gate and cell state update mechanism; The fully connected layer is used to flatten the temporal features output by the LSTM and then perform a nonlinear transformation to regress the output to predict the number of dewdrops.

6. The method for preventing condensation in electrical cabinet-type equipment according to claim 5, characterized in that, The training process of the CNN-LSTM hybrid deep learning model includes: supervised training using a historical dataset with real dew droplet labels, and iterative optimization of model parameters through backpropagation.

7. A condensation control system for electrical cabinet-type equipment, the system being applicable to the condensation control method for electrical cabinet-type equipment as described in any one of claims 1 to 6, characterized in that, include: Intelligent sensing module: including condensation acquisition board and impedance / capacitance sensing circuit connected thereto, used to acquire and output a time-series signal matrix reflecting the condensation state in real time; Intelligent linkage control module: including microcontroller, used to reduce noise, normalize, and extract multi-dimensional features from the signal output by the sensing module, run CNN-LSTM deep learning model to predict the number of dew drops in real time, and generate control commands based on the comparison between the predicted value and the warning threshold; Condensation dehumidification module: including air supply system, semiconductor refrigerator, heating circuit and drainage pipe, used to condense the humid air in the cabinet into water and discharge it outside the cabinet; Communication module: Used to upload condensation early warning information to a remote monitoring center or cloud platform.

8. A condensation control system for electrical cabinet equipment according to claim 7, characterized in that, The CNN-LSTM deep learning model deployed in the intelligent linkage control module is embedded in the microcontroller and supports online inference.

9. A condensation control system for electrical cabinet equipment according to claim 7, characterized in that, The heating circuit of the condensation dehumidification module is linked to the semiconductor cooler for control. When the ambient temperature is lower than a preset threshold, the heating circuit is activated to keep the surface temperature of the condenser plate above 0°C.

10. A condensation control system for electrical cabinet equipment according to claim 7, characterized in that, The condensation collection plate is installed on the inner surface of the top of the cabinet.

Images