Drainage type switch cabinet humidity management system and method

Abstract

The invention relates to the technical field of humidity control of drainage type switch cabinets, in particular to a humidity management system and method of a drainage type switch cabinet, by arranging a humidity sensor network in the switch cabinet, humidity data of different point positions in the switch cabinet are obtained, a humidity data matrix is generated, and external temperature and humidity information obtained by an environment collection module is combined to obtain humidity information of the switch cabinet. An environment parameter matrix is generated, the monitoring precision is improved, and the limitation of local monitoring is overcome. A real-time judgment unit based on a residual neural network and a prediction judgment unit based on deep learning are used for analyzing a humidity data matrix and an environment parameter matrix and accurately distinguishing non-humidity, natural humidity and accidental humidity, so that a proper dehumidification strategy is selected, the equipment fault risk caused by misjudgment and delay reaction is effectively reduced, and the equipment quality is improved. And the energy consumption cost of the dehumidification device is saved.

Description

Technical Field

[0001] The invention relates to the technical field of humidity control of a drainage type switch cabinet, and in particular to a humidity management system and method for a drainage type switch cabinet. Background Art

[0002] In modern power systems, switchgear, as an important power distribution and control equipment, is widely used in various industrial and civil places. The stable operation of switchgear plays a vital role in the safety and reliability of power systems. However, in humid or rainy environments, the inside of the switchgear is prone to excessive humidity, resulting in reduced insulation performance of the equipment, increased electrical failures, and even serious accidents such as short circuits and fires.

[0003] Traditional humidity control methods mainly rely on simple dehumidification devices, such as installing fixed dehumidifiers or fans, and dehumidifying through manual or timed control. However, this method has the problems of slow response speed and limited dehumidification effect when facing complex environmental changes. In addition, traditional methods cannot determine the specific humidity type in the cabinet, and cannot make intelligent adjustments based on real-time changes in environmental parameters. The energy consumption cost of the dehumidification device is high. At the same time, the monitoring of humidity has the limitation of local monitoring, which makes it difficult to effectively avoid equipment damage and failure caused by unexpected humidity. Summary of the invention

[0004] To solve the above problems, the present invention provides a humidity management system and method for a drainage type switch cabinet. By setting a humidity sensor network in the cabinet and combining residual network training and learning with environmental acquisition parameters, the humidity type is identified to match the drainage and dehumidification strategy, thereby realizing the determination of humidity type and prediction of humidity conditions and taking corresponding measures, saving the energy consumption cost of the dehumidification device and effectively avoiding equipment damage and failures caused by humidity.

[0005] To achieve the above object, the technical solution adopted by the present invention is: A moisture management system for a drainage switch cabinet, comprising a humidity sensor network, a humidity analysis module and a drainage control module connected in sequence, and an environment acquisition module, wherein the environment acquisition module is connected to the humidity analysis module; The humidity sensor network is used to obtain humidity data at different points in the switch cabinet, construct a humidity data matrix according to the spatial position of the points, and collect humidity data matrices of continuous time lines to construct a humidity matrix set; The environment acquisition module is used to collect temperature data and humidity data outside the switch cabinet, and to construct an environment parameter matrix based on the temperature data and humidity data outside the switch cabinet; The humidity analysis module includes a real-time determination unit and a prediction determination unit; The real-time determination unit is used to analyze the humidity matrix set and the environmental parameter matrix based on the residual neural network to determine the humidity type, and the humidity type includes non-humidity, natural humidity and accidental humidity; The prediction and determination unit is used to perform the following steps: Based on deep learning, the humidity matrix set is correlated and analyzed to predict the humidity change trend of each point in the future time period and extract potential humidity anomaly areas; Determine whether the humidity abnormal area coincides with the switch in the cabinet and send the coincidence determination result to the drainage control module; The drainage control module is used to match the preset drainage strategy according to the moisture type and the coincidence judgment result, and control the drainage facilities in the cabinet to drain and dehumidify according to the drainage strategy.

[0006] Further, the humidity sensor network includes a processor and a plurality of humidity sensors disposed in different areas of the switch cabinet, the processor being connected to the humidity sensors; The humidity sensor is used to obtain humidity data; The processor is used to record the spatial position information of all humidity sensors, integrate all humidity data with the same time stamp and arrange them according to the spatial position information of the humidity sensors to construct a humidity data matrix.

[0007] Furthermore, the construction of the environmental parameter matrix based on the temperature data and humidity data outside the switch cabinet includes the following steps: Obtain temperature and humidity data at several different points on the outer surface of the switch cabinet; The temperature data and humidity data are arranged in the preset point order to construct the environmental parameter matrix.

[0008] Furthermore, the analysis of the humidity matrix set and the environmental parameter matrix based on the residual neural network includes the following steps: Performing data preprocessing on the humidity matrix set and the environmental parameter matrix to generate a preprocessed data matrix, wherein the data preprocessing includes data cleaning and normalization; A residual neural network model is constructed based on the preprocessed data matrix, and high-dimensional features are extracted layer by layer through several residual blocks to generate feature maps; The feature map is weightedly analyzed through the attention mechanism to generate the moisture type judgment result.

[0009] Furthermore, the residual neural network model is trained by the following steps: Construct a training data set, take the historical humidity matrix set and the environmental parameter matrix as input data, take the humidity type and humidity change trend as target output, and generate a labeled data sample set; Initialize the parameters of the residual neural network model, use the stochastic gradient descent method for initial training, and calculate the error of each iteration through the cross entropy loss function; Perform error back propagation training on the residual neural network model by adjusting the weight parameters of the residual block layer by layer; The trained residual neural network model is verified by the K-fold cross-validation method.

[0010] Furthermore, the formula of the residual neural network model is as follows: ; in, For input data, including preprocessed data matrix; Output results to the network, including the determination results of humidity types or the prediction of humidity change trends; and is the weight matrix of the residual block; and is the bias term; is the rectified linear unit activation function; is the Sigmoid activation function.

[0011] Furthermore, the attention mechanism is trained by the following steps: Taking the historical humidity matrix set, environmental parameter matrix and corresponding humidity type as input data, the time series feature set is constructed through time series to generate the training sample set; Initialize the query, key, and value matrices of the attention mechanism model, calculate the feature weights for the input data, and assign weights based on the relative importance of different humidity feature points; Based on the training samples and weight calculation results, continue to adjust the query, key, and value matrices.

[0012] Furthermore, the drainage facilities in the cabinet include: a drying device, a fan, a plurality of vents and a drain port at the bottom of the cabinet.

[0013] Furthermore, the preset drainage strategy includes: If the humidity type is non-humidity and the coincidence determination result is no coincidence, turn off the drying and dehumidifying device, fan and drain port at the bottom of the cabinet, and open the vent; If the humidity type is non-humidity and the coincidence determination result is coincidence, turn on the drying and dehumidifying device and close the fan, vents and drain at the bottom of the cabinet; If the humidity type is natural humidity, turn on the drying and dehumidifying device, and close the fan, vents, and drain at the bottom of the cabinet; If the humidity type is unexpected, turn on the fan, vents and drain at the bottom of the cabinet, and turn off the drying and dehumidifying device.

[0014] A method for humidity management of a drainage type switch cabinet, applied to the aforementioned humidity management system of a drainage type switch cabinet, comprises the following steps: Obtain humidity data at different points in the switch cabinet, construct a humidity data matrix according to the spatial position of the points, and collect the humidity data matrix of the continuous time line to construct a humidity matrix set; Collect temperature and humidity data outside the switch cabinet, and build an environmental parameter matrix based on the temperature and humidity data outside the switch cabinet; Based on the residual neural network, the humidity matrix set and the environmental parameter matrix are analyzed to determine the humidity type, which includes non-humidity, natural humidity and accidental humidity; Based on deep learning, the humidity matrix set is correlated and analyzed to predict the humidity change trend of each point in the future time period and extract potential humidity anomaly areas; Determine whether the abnormal humidity area coincides with the switch in the cabinet; The preset drainage strategy is matched according to the moisture type and the coincidence judgment result, and the drainage facilities in the control cabinet are controlled according to the drainage strategy to drain and dehumidify.

[0015] The beneficial effect of the present invention is that the present invention solves the problems of slow response speed and limited dehumidification effect of traditional humidity control methods in complex environments by setting up a humidity sensor network, an environmental acquisition module and an intelligent humidity analysis module. The humidity sensor network can accurately obtain humidity data at different points inside the switch cabinet, construct a humidity data matrix, and generate an environmental parameter matrix in combination with the external temperature and humidity information provided by the environmental acquisition module. This data acquisition method not only improves the accuracy of monitoring, but also overcomes the local monitoring limitations in traditional monitoring. The real-time judgment unit based on the residual neural network and the prediction judgment unit based on deep learning perform in-depth analysis on the collected humidity and environmental data. Through the real-time judgment unit, the present solution can accurately distinguish different types of humidity, including non-humidity, natural humidity and accidental humidity, so as to make corresponding dehumidification strategy selections for different types of humidity. This intelligent analysis method effectively reduces the risk of equipment failure caused by misjudgment or delayed response. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a structural schematic diagram of a drainage type switch cabinet humidity management system in the present invention.

[0017] Figure 2 It is a flow chart of the steps of analyzing the humidity matrix set and the environmental parameter matrix based on the residual neural network in the present invention. DETAILED DESCRIPTION

[0018] See also Figure 1-2As shown, the present invention relates to a moisture management system for a drainage type switch cabinet, comprising a humidity sensor network, a humidity analysis module and a drainage control module connected in sequence, and also comprising an environment acquisition module, wherein the environment acquisition module is connected to the humidity analysis module; The humidity sensor network is used to obtain humidity data at different points in the switch cabinet, construct a humidity data matrix according to the spatial position of the points, and collect humidity data matrices of continuous time lines to construct a humidity matrix set; The environment acquisition module is used to collect temperature data and humidity data outside the switch cabinet, and to construct an environment parameter matrix based on the temperature data and humidity data outside the switch cabinet; The humidity analysis module includes a real-time determination unit and a prediction determination unit; The real-time determination unit is used to analyze the humidity matrix set and the environmental parameter matrix based on the residual neural network to determine the humidity type, and the humidity type includes non-humidity, natural humidity and accidental humidity; The prediction and determination unit is used to perform the following steps: Based on deep learning, the humidity matrix set is correlated and analyzed to predict the humidity change trend of each point in the future time period and extract potential humidity anomaly areas; Determine whether the humidity abnormal area coincides with the switch in the cabinet and send the coincidence determination result to the drainage control module; The drainage control module is used to match the preset drainage strategy according to the moisture type and the coincidence judgment result, and control the drainage facilities in the cabinet to drain and dehumidify according to the drainage strategy.

[0019] It should be noted that the humidity sensor network consists of several humidity sensors, which are evenly arranged at various key positions inside the switch cabinet, including the top and bottom of the cabinet and around key electrical components. Each humidity sensor is equipped with an independent embedded processing unit for real-time acquisition of humidity data and transmission of the data to the central processor. The central processor constructs a humidity data matrix according to the preset three-dimensional coordinate system based on the position data of the sensor, and samples the data at fixed time intervals to generate a humidity matrix set of continuous time lines. The environment acquisition module uses a highly sensitive temperature and humidity sensor to collect temperature and humidity data of the external environment of the switch cabinet in real time. The module is connected to the humidity analysis module to transmit the collected external environment data to the system central database and construct an environmental parameter matrix based on the external temperature and humidity data. The environmental parameter matrix is ​​used to supplement the humidity data matrix and provide environmental background information of humidity changes, thereby improving the accuracy of data analysis. The humidity analysis module is embedded with a computing unit, including a real-time determination unit and a prediction determination unit. The real-time determination unit uses a hardware accelerator based on a residual neural network to determine the humidity type in the cabinet by synchronously analyzing the humidity matrix set and the environmental parameter matrix. The real-time judgment unit can quickly process large amounts of environmental and humidity data, complete calculation tasks within milliseconds, and classify the data according to humidity types to ensure real-time performance.

[0020] Further, the humidity sensor network includes a processor and a plurality of humidity sensors disposed in different areas of the switch cabinet, the processor being connected to the humidity sensors; The humidity sensor is used to obtain humidity data; The processor is used to record the spatial position information of all humidity sensors, integrate all humidity data with the same time stamp and arrange them according to the spatial position information of the humidity sensors to construct a humidity data matrix.

[0021] In some embodiments, each humidity sensor has independent spatial positioning information for identifying its specific position inside the switch cabinet. These sensors are installed in key areas such as the main switch position, branch wiring terminals, and around the control module to fully cover the humidity conditions of the switch cabinet. All humidity sensors synchronously collect humidity data within a preset time interval, and perform preliminary data processing, including data filtering and signal denoising, through the microprocessor embedded in the sensor to reduce errors. The central processing unit is used to integrate the humidity data of each humidity sensor. Specifically, the processor receives and records the data of each sensor and its corresponding spatial position information. In order to ensure the timeliness and uniformity of the data, the central processing unit timestamps the data from different sensors, integrates the humidity data with the same timestamp, and forms a time series data set. In the process of data integration, the central processing unit arranges the data of each sensor according to the three-dimensional spatial coordinates according to the spatial position information of the humidity sensor, and constructs a humidity data matrix. The matrix presents the humidity distribution in the switch cabinet in the form of a three-dimensional grid, and the data of each grid node represents the humidity value of a specific position at a specific time point. By constructing a humidity data matrix of a continuous timeline, the system can monitor the changing trend of humidity in real time, providing a basis for subsequent data analysis and prediction.

[0022] Furthermore, the construction of the environmental parameter matrix based on the temperature data and humidity data outside the switch cabinet includes the following steps: Obtain temperature and humidity data at several different points on the outer surface of the switch cabinet; The temperature data and humidity data are arranged in the preset point order to construct the environmental parameter matrix.

[0023] In some embodiments, environmental data at different points are obtained in real time by means of a number of high-precision temperature and humidity sensors arranged on the outer surface of the switch cabinet. These sensors are evenly distributed in different external areas of the switch cabinet, such as the top, side and bottom, according to the preset layout rules, so as to fully capture the temperature and humidity conditions of the environment in which the switch cabinet is located. Each sensor uses a high-sensitivity element, which can accurately respond to small changes in temperature and humidity to ensure the reliability and accuracy of the data. Secondly, the system arranges the collected temperature data and humidity data according to the preset point order. During the data collection process, the data of each sensor is accompanied by its specific spatial coordinate information. The central processor sorts the temperature and humidity data of all sensors in a pre-set order according to these coordinate information. The sorting method is based on the geometric structure of the switch cabinet to ensure that the spatial distribution of the data in the environmental parameter matrix is ​​consistent with the position of the actual collection point. Next, the environmental parameter matrix is ​​constructed according to the sorted data. The environmental parameter matrix is ​​a two-dimensional array, in which each row represents the temperature and humidity values ​​of a certain collection point, and the columns correspond to different collection time points. The matrix structure enables the system to analyze the environmental changes at different points in a continuous time period in real time. Through the dynamic changes of the data in the matrix, the potential impact of the external environment on the humidity inside the switch cabinet can be effectively monitored and predicted.

[0024] Furthermore, the analysis of the humidity matrix set and the environmental parameter matrix based on the residual neural network includes the following steps: Performing data preprocessing on the humidity matrix set and the environmental parameter matrix to generate a preprocessed data matrix, wherein the data preprocessing includes data cleaning and normalization; A residual neural network model is constructed based on the preprocessed data matrix, and high-dimensional features are extracted layer by layer through several residual blocks to generate feature maps; The feature map is weightedly analyzed through the attention mechanism to generate the moisture type judgment result.

[0025] In some embodiments, the humidity matrix set and the environmental parameter matrix are first preprocessed to ensure the accuracy and consistency of the data. The first step of preprocessing is data cleaning, including detecting and eliminating possible outliers or sensor failure data, such as completing or smoothing mutation values ​​and missing data. Next, the processed data is normalized to reduce the dimensional differences between different data sources. Normalization uses the Min-Max method to map all data to the range of [0,1] to generate a preprocessed data matrix to provide a unified input format for subsequent neural network analysis. Based on the above preprocessed data matrix, a residual neural network model is constructed. The core of the model is to use multiple residual blocks to extract high-dimensional features layer by layer. Each residual block includes two layers of convolution operations with a convolution kernel size of 3x3, which is intended to extract local spatial features. After each layer of convolution operation, batch normalization is performed to accelerate convergence and prevent gradient disappearance. The residual block directly superimposes the input to the output through a shortcut connection to form a jump connection to maintain feature consistency and prevent the degradation problem of the deep network. The entire residual neural network model uses a strategy of increasing depth layer by layer according to the complexity of the input data to generate a feature map to describe the potential feature relationship of different humidity types. The generated feature map is further weighted by the attention mechanism. The attention mechanism aims to enhance key features and suppress irrelevant or noise data. The specific implementation method is to use the feature channels in the feature map as query, key and value to calculate the importance weight of each feature point. The weight is generated based on the similarity measurement between feature points, and the weight is normalized using the Softmax function. Finally, the attention mechanism readjusts the feature map according to the weighted results to highlight the data points that have a key impact on the determination of humidity types. According to the above analysis process, the output generated by the system is the determination result of humidity type, which includes three types: non-humidity, natural humidity and accidental humidity. The determination result serves as the basis for the subsequent drainage strategy selection, ensuring the scientificity and pertinence of the drainage plan, and improving the efficiency and reliability of humidity control of the switch cabinet under variable environmental conditions.

[0026] Furthermore, the residual neural network model is trained by the following steps: Construct a training data set, take the historical humidity matrix set and the environmental parameter matrix as input data, take the humidity type and humidity change trend as target output, and generate a labeled data sample set; Initialize the parameters of the residual neural network model, use the stochastic gradient descent method for initial training, and calculate the error of each iteration through the cross entropy loss function; Perform error back propagation training on the residual neural network model by adjusting the weight parameters of the residual block layer by layer; The trained residual neural network model is verified by the K-fold cross-validation method.

[0027] In some embodiments, first, a training data set is constructed. The training data set consists of a historical humidity matrix set and an environmental parameter matrix as input data sources. Specifically, the humidity matrix set records the humidity distribution inside the switch cabinet at different time points, while the environmental parameter matrix contains the external temperature and humidity environment changes during the corresponding period. The target output includes two parts: one is the humidity type inside the cabinet, covering three categories: non-humidity, natural humidity and accidental humidity; the other is the prediction of humidity change trend. In order to ensure the quality of the training data, all data samples are manually reviewed and labeled to ensure the accuracy of the labels, and finally form a labeled data sample set. After the data set is constructed, the parameters of the residual neural network model are initialized. The parameter initialization uses the Xavier or He initialization method to optimize the initial state of the neural network and ensure that the gradient remains stable during the forward propagation and back propagation process. The model optimization method uses the stochastic gradient descent method (SGD), which can effectively reduce the computational overhead on large-scale data sets. In the initial training stage, the system calculates the output result through forward propagation, and uses the cross-entropy loss function (Cross-Entropy Loss) to evaluate the difference between the output result and the target label. The cross entropy loss function is suitable for multi-classification problems and can quantify the classification accuracy of the model, providing a basis for subsequent error adjustment. Next, the model enters the error back propagation training stage. The back propagation algorithm adjusts the weight parameters of the residual block layer by layer by calculating the gradient information. Specifically, in each iteration, the system calculates the partial derivative of the loss function for each weight parameter and updates the parameters according to the learning rate to gradually reduce the prediction error. The residual block retains the basic information of the input features through shortcut connections, allowing the model to maintain an efficient feature learning process in the deep network and avoid the gradient disappearance problem. Finally, the trained model is verified by the K-fold cross-validation method. The verification process divides the data set into K subsets, selects one subset as the verification set each time, and the remaining subsets as the training set, and performs multiple rounds of iterative verification to comprehensively evaluate the generalization ability of the model. The final performance of the model is determined by calculating indicators such as accuracy, precision, and recall rate of each round of verification. If the verification result does not meet the expected standards, the system will adjust the structure or hyperparameters of the network based on the verification feedback to further optimize the model.

[0028] Furthermore, the formula of the residual neural network model is as follows: ; in, For input data, including preprocessed data matrix; Output results to the network, including the determination results of humidity types or the prediction of humidity change trends; and is the weight matrix of the residual block; and is the bias term; is the rectified linear unit activation function; is the Sigmoid activation function.

[0029] Specifically, the input data Represents the humidity matrix set and environmental parameter matrix after data preprocessing as the input features of the neural network. The input data has been normalized and cleaned to ensure data consistency and stability. The weight matrix in the residual block and , the parameters used for the convolution operation. The learning of the weight matrix is ​​optimized by gradient descent during training, aiming to maximize the representation ability of the features. and The role of is to ensure that the neural network can produce appropriate output responses even when there is no input signal, and to prevent the output value of the activation function from being fixed in a certain state. After each convolutional layer, nonlinear activation is performed to truncate the negative value to zero, which effectively enhances the nonlinear representation ability of the model. The Sigmoid activation function is used to normalize the final output value to the interval [0,1], which is suitable for probabilistic prediction tasks, such as the classification of humidity types. In the above formula, the design of the residual block is the key feature of the model. By introducing shortcut connections between network layers, the input data is directly passed to the output of the deep network, thereby avoiding the gradient vanishing problem. This structure allows the network to retain the original information of the input data while extracting deep features, improving the learning efficiency of the model on complex data sets. By stacking multiple layers of residual blocks, the model can gradually capture the high-dimensional correlation features in humidity and environmental data, providing accurate decision-making basis for subsequent dehumidification and drainage strategies.

[0030] Furthermore, the attention mechanism is trained by the following steps: Taking the historical humidity matrix set, environmental parameter matrix and corresponding humidity type as input data, the time series feature set is constructed through time series to generate the training sample set; Initialize the query, key, and value matrices of the attention mechanism model, calculate the feature weights for the input data, and assign weights based on the relative importance of different humidity feature points; Based on the training samples and weight calculation results, continue to adjust the query, key, and value matrices.

[0031] In some embodiments, the first step of the training process is to construct a time series feature set. By obtaining a historical humidity matrix set, an environmental parameter matrix and its corresponding humidity type, these data are used as input data sources. When constructing a training sample set, the continuous humidity and environmental data are organized into a time series data format by a time series analysis method. Specifically, each time series sample contains a humidity matrix of multiple time steps and external environmental parameters at the corresponding time to ensure that the sample set can reflect the dynamic characteristics of humidity changes. These time series samples not only record the humidity information in the past, but also clearly mark the humidity type at that time for the supervised learning of subsequent models. Secondly, in the training stage, the core parameters of the attention mechanism model are initialized, including the query matrix (Query), the key matrix (Key) and the value matrix (Value). The initialization of these matrices adopts a random distribution method to ensure the diversity of parameters. The role of the query, key and value matrices is to calculate the feature weights of the input data. During the training process, the model encodes the input humidity matrix and environmental parameter matrix, and each feature point generates its corresponding query, key and value vectors. The query and key vectors are used to calculate the similarity between feature points to form attention weights. The model calculates the relative importance of each humidity feature point through the inner product operation, and then normalizes the weights through the Softmax function to obtain the importance ratio of each feature point in the whole. Finally, based on the generated training samples and the feature weights calculated above, the parameters of the query, key, and value matrices are further adjusted. In each iteration, the model calculates the prediction error of the attention weight through the error backpropagation algorithm (Backpropagation), and adjusts the matrix parameters according to the gradient information to reduce the error. This process ensures that the model can accurately identify the importance of different humidity feature points and assign appropriate weights to highlight the features that have a key impact on the determination of humidity types. At the end of training, the attention mechanism model can dynamically focus on the correlation between different humidity and environmental feature points, improving the accuracy and robustness of humidity type prediction.

[0032] Furthermore, the drainage facilities in the cabinet include: a drying device, a fan, a plurality of vents and a drain port at the bottom of the cabinet.

[0033] In some embodiments, the drying device uses an integrated heating unit and an intelligent control module and is installed at a key position inside the switch cabinet. The drying device automatically starts when the humidity is too high by precisely controlling the power output of the heating element to increase the temperature of the air in the cabinet and accelerate the evaporation and discharge of water vapor. The device has built-in temperature and humidity sensors that can monitor the environmental status in the cabinet in real time to ensure that the heating process does not cause local overheating or damage electronic equipment. In an emergency with high humidity, the drying device can operate in a high-power mode to quickly reduce the humidity to a safe range. Secondly, the fan system is used to accelerate the circulation of air in the cabinet and assist the dehumidification effect of the drying device. The fan is installed in the upper and lower areas of the cabinet. Through the multi-wind direction design, it can effectively promote air circulation and guide the humid air to the drainage channel. At the same time, the speed and operation mode of the fan can be intelligently adjusted through the drainage control module to match different humidity conditions. For example, when the humidity type is natural humidity, the fan runs at a medium speed to increase air circulation; while in the case of accidental humidity, the fan runs at a high speed to quickly remove moisture. Several vents are set at the top and side of the cabinet, and are designed with sealed regulating valves to prevent free entry of outside air under normal conditions. The opening and closing status of the vents is controlled in real time by the humidity analysis module, and dynamically adjusted according to humidity changes to ensure that the humidity in the cabinet is always maintained within a stable range. The vents are also equipped with dust screens and filters to prevent external pollutants from entering the cabinet. Finally, the drain at the bottom of the cabinet is the core component of the drainage facility, which is used to collect and discharge excess condensed water in the cabinet. The drain uses an automatically opened and closed electric control valve, which can automatically open and discharge accumulated water when the humidity sensor detects condensation of water vapor, and automatically close when the humidity returns to normal levels. The drain is also equipped with an anti-blocking device to ensure long-term reliability.

[0034] Furthermore, the preset drainage strategy includes: If the humidity type is non-humidity and the coincidence determination result is no coincidence, turn off the drying and dehumidifying device, fan and drain port at the bottom of the cabinet, and open the vent; If the humidity type is non-humidity and the coincidence determination result is coincidence, turn on the drying and dehumidifying device and close the fan, vents and drain at the bottom of the cabinet; If the humidity type is natural humidity, turn on the drying and dehumidifying device, and close the fan, vents, and drain at the bottom of the cabinet; If the humidity type is unexpected, turn on the fan, vents and drain at the bottom of the cabinet, and turn off the drying and dehumidifying device.

[0035] In some embodiments, when the humidity type is determined to be non-humidity and the result of the coincidence determination is no coincidence, it means that the overall humidity in the switch cabinet is at a normal level, and the predicted humidity abnormal area does not involve key equipment. In this case, in order to save energy, the system will turn off all dehumidification equipment, including drying devices, fans, and cabinet bottom drains, and open the vents to maintain internal humidity stability by natural air circulation. This strategy ensures that no unnecessary energy consumption is generated under normal humidity conditions while maintaining good air exchange. If the humidity type is non-humidity, but the coincidence determination result shows that there is a humidity abnormal area that coincides with key equipment, the system adopts a more cautious strategy. At this time, the system will turn on the drying device to accelerate dehumidification to prevent humidity from causing potential hidden dangers to key equipment. At the same time, the fan, vents, and cabinet bottom drains are closed to ensure that unexpected moisture continues to invade. In the case of natural humidity, the increase in humidity is due to natural changes in the environment, such as seasonal humidity increases. At this time, the system will only start the drying device to reduce the humidity in the air by heating and heating. All other equipment, including fans, vents, and cabinet bottom drains, will remain closed to prevent external humid air from entering the cabinet and ensure the dehumidification effect. This strategy focuses on the gentle regulation of the overall humidity to maintain a stable environment inside the cabinet. When the humidity type is determined to be unexpected humidity, it means that there may be a sudden increase in humidity inside the cabinet, which may be caused by equipment failure or sudden changes in the external environment. The system will urgently start all strong dehumidification measures, including turning on fans, vents and drains at the bottom of the cabinet to quickly remove moisture from the cabinet. At this time, the drying device remains closed to avoid excessive energy consumption caused by heating. The fan will run at a high speed, the vents will be fully opened to ensure that moisture can be quickly discharged, and the drain at the bottom of the cabinet will continue to remove accumulated water until the humidity returns to the normal range. These preset drainage strategies are based on intelligent judgment of humidity type and coincidence judgment results, realizing refined management under different humidity scenarios, thereby ensuring the safe operation of the switch cabinet. Each strategy is scientifically designed to minimize the potential threat of humidity to the equipment with minimal energy consumption.

[0036] The present invention also includes a moisture management method for a drainage type switch cabinet, comprising the following steps: Obtain humidity data at different points in the switch cabinet, construct a humidity data matrix according to the spatial position of the points, and collect the humidity data matrix of the continuous time line to construct a humidity matrix set; Collect temperature and humidity data outside the switch cabinet, and build an environmental parameter matrix based on the temperature and humidity data outside the switch cabinet; Based on the residual neural network, the humidity matrix set and the environmental parameter matrix are analyzed to determine the humidity type, which includes non-humidity, natural humidity and accidental humidity; Based on deep learning, the humidity matrix set is correlated and analyzed to predict the humidity change trend of each point in the future time period and extract potential humidity anomaly areas; Determine whether the abnormal humidity area coincides with the switch in the cabinet; The preset drainage strategy is matched according to the moisture type and the coincidence judgment result, and the drainage facilities in the control cabinet are controlled according to the drainage strategy to drain and dehumidify.

[0037] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A drainage switch cabinet humidity management system, characterized in that: It includes a humidity sensor network, a humidity analysis module and a drainage control module connected in sequence, and also includes an environment acquisition module, the environment acquisition module is connected to the humidity analysis module; The humidity sensor network is used to obtain humidity data at different points in the switch cabinet, construct a humidity data matrix according to the spatial position of the points, and collect humidity data matrices of continuous time lines to construct a humidity matrix set; The environment acquisition module is used to collect temperature data and humidity data outside the switch cabinet, and to construct an environment parameter matrix based on the temperature data and humidity data outside the switch cabinet; The humidity analysis module includes a real-time determination unit and a prediction determination unit; The real-time determination unit is used to analyze the humidity matrix set and the environmental parameter matrix based on the residual neural network to determine the humidity type, and the humidity type includes non-humidity, natural humidity and accidental humidity; The prediction and determination unit is used to perform the following steps: Based on deep learning, the humidity matrix set is correlated and analyzed to predict the humidity change trend of each point in the future time period and extract potential humidity anomaly areas; Determine whether the humidity abnormal area coincides with the switch in the cabinet and send the coincidence determination result to the drainage control module; The drainage control module is used to match the preset drainage strategy according to the moisture type and the coincidence judgment result, and control the drainage facilities in the cabinet to drain and dehumidify according to the drainage strategy; The analysis of the humidity matrix set and the environmental parameter matrix based on the residual neural network comprises the following steps: Performing data preprocessing on the humidity matrix set and the environmental parameter matrix to generate a preprocessed data matrix, wherein the data preprocessing includes data cleaning and normalization; A residual neural network model is constructed based on the preprocessed data matrix, and high-dimensional features are extracted layer by layer through several residual blocks to generate feature maps; The feature map is weightedly analyzed through the attention mechanism to generate the moisture type determination result; The formula of the residual neural network model is as follows: ； in, For input data, including preprocessed data matrix; Output results to the network, including the determination results of humidity types or the prediction of humidity change trends; and is the weight matrix of the residual block; and is the bias term; is the rectified linear unit activation function; is the Sigmoid activation function.

2. A drainage switch cabinet humidity management system according to claim 1, characterized in that: The humidity sensor network includes a processor and a plurality of humidity sensors disposed in different areas of the switch cabinet, wherein the processor is connected to the humidity sensors; The humidity sensor is used to obtain humidity data; The processor is used to record the spatial position information of all humidity sensors, integrate all humidity data with the same time stamp and arrange them according to the spatial position information of the humidity sensors to construct a humidity data matrix.

3. A drainage switch cabinet humidity management system according to claim 1, characterized in that: The construction of the environmental parameter matrix based on the temperature data and humidity data outside the switch cabinet comprises the following steps: Obtain temperature and humidity data at several different points on the outer surface of the switch cabinet; The temperature data and humidity data are arranged in the preset point order to construct the environmental parameter matrix.

4. A drainage switch cabinet humidity management system according to claim 1, characterized in that: The residual neural network model is trained by the following steps: Construct a training data set, take the historical humidity matrix set and the environmental parameter matrix as input data, take the humidity type and humidity change trend as target output, and generate a labeled data sample set; Initialize the parameters of the residual neural network model, use the stochastic gradient descent method for initial training, and calculate the error of each iteration through the cross entropy loss function; Perform error back propagation training on the residual neural network model by adjusting the weight parameters of the residual block layer by layer; The trained residual neural network model is verified by the K-fold cross-validation method.

5. The moisture management system for a drainage switch cabinet according to claim 1, characterized in that: The attention mechanism is trained by the following steps: Taking the historical humidity matrix set, environmental parameter matrix and corresponding humidity type as input data, the time series feature set is constructed through time series to generate the training sample set; Initialize the query, key, and value matrices of the attention mechanism model, calculate the feature weights for the input data, and assign weights based on the relative importance of different humidity feature points; Based on the training samples and weight calculation results, continue to adjust the query, key, and value matrices.

6. A drainage switch cabinet humidity management system according to claim 1, characterized in that: The drainage facilities in the cabinet include: a drying device, a fan, a plurality of vents and a drain port at the bottom of the cabinet.

7. A drainage switch cabinet humidity management system according to claim 7, characterized in that: The preset drainage strategy includes: If the humidity type is non-humidity and the coincidence determination result is no coincidence, turn off the drying and dehumidifying device, fan and drain port at the bottom of the cabinet, and open the vent; If the humidity type is non-humidity and the coincidence determination result is coincidence, turn on the drying and dehumidifying device and close the fan, vents and drain at the bottom of the cabinet; If the humidity type is natural humidity, turn on the drying and dehumidifying device, and close the fan, vents, and drain at the bottom of the cabinet; If the humidity type is unexpected, turn on the fan, vents and drain at the bottom of the cabinet, and turn off the drying and dehumidifying device.

8. A method for managing humidity in a drainage type switch cabinet, applied to a humidity management system for a drainage type switch cabinet according to any one of claims 1 to 7, characterized in that: The following steps are involved: Obtain humidity data at different points in the switch cabinet, construct a humidity data matrix according to the spatial position of the points, and collect the humidity data matrix of the continuous time line to construct a humidity matrix set; Collect temperature and humidity data outside the switch cabinet, and build an environmental parameter matrix based on the temperature and humidity data outside the switch cabinet; Based on the residual neural network, the humidity matrix set and the environmental parameter matrix are analyzed to determine the humidity type, which includes non-humidity, natural humidity and accidental humidity; Based on deep learning, the humidity matrix set is correlated and analyzed to predict the humidity change trend of each point in the future time period and extract potential humidity anomaly areas; Determine whether the abnormal humidity area coincides with the switch in the cabinet; The preset drainage strategy is matched according to the moisture type and the coincidence judgment result, and the drainage facilities in the control cabinet are controlled according to the drainage strategy to drain and dehumidify.

Images