Switch cabinet temperature and humidity monitoring method based on multivariate neural network

Abstract

The invention provides a switch cabinet temperature and humidity monitoring method based on a multivariate neural network, and the method specifically comprises the steps: collecting the temperature and humidity data of a switch cabinet, carrying out the preprocessing of the temperature and humidity data, and constructing a corresponding training data set; constructing a CNN-LSTM temperature and humidity prediction model, and training the CNN-LSTM temperature and humidity prediction model in combination with a sparrow optimization algorithm and the training data set; according to the current temperature and humidity data of the switch cabinet, temperature and humidity prediction data of the switch cabinet are obtained through prediction, condensation abnormity judgment of the switch cabinet is carried out, a switch cabinet adjusting strategy is set according to the condensation abnormity judgment result of the switch cabinet, and the temperature and humidity in the switch cabinet are adjusted. According to the invention, the multivariate neural network is used for processing a prediction scene with a large data volume, the over-fitting problem of the model is solved through the sparrow optimization algorithm, the prediction precision is improved, the accuracy of subsequent switch cabinet condensation abnormity judgment is improved, and the operation safety of the switch cabinet is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of switch cabinet monitoring, and in particular to a switch cabinet temperature and humidity monitoring method based on a multivariate neural network. Background Art

[0002] The safety and reliability level of the power system operation is directly related to whether the power equipment components in the switch cabinet are working properly, and the operation of the power equipment components is greatly affected by the environment. For example, under long-term rainy conditions, the moisture in the air can easily form serious condensation on the surface of the electrical equipment components in the cabinet, which in turn promotes the accumulation of dirt, leading to insulation degradation or flashover accidents. Therefore, it is necessary to establish a high-precision temperature and humidity prediction model with temperature and humidity environmental factors as output variables, so as to provide an important decision-making basis for the pre-control of the temperature and humidity control system of the switch cabinet and ensure the normal and safe operation of the switch cabinet.

[0003] Since the switch cabinet temperature and humidity prediction model involves multiple environmental variable factors, and these factors have nonlinear, time-varying, coupled, and hysteresis characteristics, temperature and humidity prediction is currently often performed through neural network models such as BP neural network, extreme learning machine, support vector machine, or random forest. These temperature and humidity prediction methods can not only derive the mapping relationship between the external environmental factors and internal heating factors of the switch cabinet and the internal temperature and humidity of the switch cabinet, but also reduce the prediction time and improve the prediction accuracy. However, this method of using a single neural network model to predict temperature and humidity cannot handle large data sets, and it is easy to have phenomena such as gradient disappearance. The prediction accuracy is difficult to meet the switch cabinet temperature and humidity prediction requirements, and the operation safety of the switch cabinet cannot be guaranteed. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art that a single neural network model is used to predict temperature and humidity, which cannot process large data sets, is prone to gradient disappearance and other phenomena, and has low prediction accuracy. A switch cabinet temperature and humidity monitoring method based on a multivariate neural network is provided, and the feature extraction function of a convolutional neural network is utilized to combine a convolutional neural network with a long short-term memory neural network to construct a CNN-LSTM temperature and humidity prediction model, thereby adapting to temperature and humidity prediction scenarios with large data volumes, and further applying a sparrow optimization algorithm to the training process of the constructed CNN-LSTM temperature and humidity prediction model to avoid the constructed CNN-LSTM temperature and humidity prediction model from fitting transition or underfitting during the prediction process, thereby improving the prediction accuracy, and thereby improving the accuracy of subsequent switch cabinet condensation abnormality judgment, thereby ensuring the safe operation of the switch cabinet.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] The switch cabinet temperature and humidity monitoring method based on multivariate neural network includes:

[0007] Collect the temperature and humidity data of the switch cabinet, pre-process the temperature and humidity data, and build the corresponding training data set;

[0008] Construct a CNN-LSTM temperature and humidity prediction model, and train the CNN-LSTM temperature and humidity prediction model by combining the sparrow optimization algorithm and training data set;

[0009] Obtain the current switch cabinet temperature and humidity data, input the trained CNN-LSTM temperature and humidity prediction model, and obtain the switch cabinet temperature and humidity prediction data;

[0010] The switch cabinet condensation abnormality is judged according to the temperature and humidity prediction data, and the switch cabinet adjustment strategy is set according to the switch cabinet condensation abnormality judgment result, and the temperature and humidity in the switch cabinet are adjusted according to the switch cabinet adjustment strategy.

[0011] Furthermore, the CNN-LSTM temperature and humidity prediction model is constructed, and the CNN-LSTM temperature and humidity prediction model is trained by combining the sparrow optimization algorithm and the training data set, including:

[0012] Initialize the CNN-LSTM temperature and humidity prediction model;

[0013] The parameter configuration of the CNN-LSTM temperature and humidity prediction model was optimized by using the sparrow optimization algorithm, and the CNN-LSTM temperature and humidity prediction model with updated parameters was retrained according to the training data set;

[0014] Obtain the model performance parameters of the CNN-LSTM temperature and humidity prediction model before and after parameter update, and compare the model performance parameters before and after the update;

[0015] If the updated model performance parameters are greater than the model performance parameters before the update, the parameter configuration of the currently updated CNN-LSTM temperature and humidity prediction model will continue to be optimized through the sparrow optimization algorithm until the updated model performance parameters are less than or equal to the model performance parameters before the update;

[0016] If the updated model performance parameters are less than or equal to the model performance parameters before the update, the parameter configuration optimization of the CNN-LSTM temperature and humidity prediction model is terminated, and the trained CNN-LSTM temperature and humidity prediction model is obtained.

[0017] Furthermore, the CNN-LSTM temperature and humidity prediction model is initialized, and parameter configuration optimization of the CNN-LSTM temperature and humidity prediction model is performed by using a sparrow optimization algorithm, including:

[0018] Set the initial parameters of the CNN-LSTM temperature and humidity prediction model, randomly select part of the data in the training data set to train the CNN-LSTM temperature and humidity prediction model, and calculate the corresponding model performance parameters before updating;

[0019] The model parameters of the CNN-LSTM temperature and humidity prediction model are used as the initial population of the sparrow optimization algorithm to update the model parameters of the CNN-LSTM temperature and humidity prediction model;

[0020] Reselect some data in the training data set to train the updated CNN-LSTM temperature and humidity prediction model, and calculate the corresponding updated model performance parameters.

[0021] Furthermore, the switch cabinet condensation abnormality judgment based on the temperature and humidity prediction data includes:

[0022] Set the unit time and calculate the corresponding dew point temperature based on the temperature and humidity forecast data within each unit time;

[0023] Compare the temperature component in the temperature and humidity prediction data in each unit time with the dew point temperature of the switch cabinet to determine the dew point temperature difference in each unit time;

[0024] The corresponding humidity change rate is determined according to the maximum and minimum values ​​of the humidity component in the temperature and humidity prediction data within each unit time, and the corresponding condensation risk level is calculated according to the humidity change rate and dew point temperature difference corresponding to each unit time.

[0025] Furthermore, the switch cabinet adjustment strategy is set according to the switch cabinet condensation abnormality judgment result, and the temperature and humidity in the switch cabinet are adjusted according to the switch cabinet adjustment strategy, including:

[0026] Compare the condensation risk levels of every two adjacent unit times, and select the adjustment nodes according to the comparison results and the corresponding condensation risk levels;

[0027] Determine the temperature component and humidity component in the temperature and humidity prediction data of each adjustment node, and select the adjustment execution object according to the temperature component and the humidity component;

[0028] The temperature adjustment value or humidity adjustment value of the adjustment execution object is set according to the humidity change rate and the dew point temperature difference corresponding to each adjustment node.

[0029] Furthermore, the step of comparing the condensation risk levels of every two adjacent unit times and selecting the adjustment node according to the comparison result and the corresponding condensation risk level includes:

[0030] Starting from the first unit time, the condensation risk levels of adjacent unit times are compared in sequence;

[0031] When the level difference of the condensation risk level between adjacent unit times does not exceed the preset threshold, the higher condensation risk level in the adjacent unit time is selected for comparison with the preset condensation risk level;

[0032] If the display exceeds the preset condensation risk level, the corresponding unit time is selected as the adjustment node, and the next unit time is selected to compare the condensation risk levels of adjacent unit times and select the adjustment nodes;

[0033] If the display does not exceed the preset condensation risk level, the lower condensation risk level in the adjacent unit time is used as the condensation risk level of the two adjacent unit time, and the condensation risk level of the adjacent unit time is compared with the next unit time and the adjustment node is selected;

[0034] When the level difference of condensation risk levels between adjacent unit times exceeds a preset threshold, the unit time with a higher condensation risk level in the adjacent unit times is selected as the adjustment node.

[0035] Furthermore, the current switch cabinet temperature and humidity data is obtained, and the trained CNN-LSTM temperature and humidity prediction model is input to obtain the switch cabinet temperature and humidity prediction data, including:

[0036] Obtain the current switch cabinet temperature and humidity data, pre-process the switch cabinet temperature and humidity data, and input the trained CNN-LSTM temperature and humidity prediction model;

[0037] The output prediction data of the CNN-LSTM temperature and humidity prediction model is denormalized to obtain the temperature and humidity prediction data of the switch cabinet.

[0038] Furthermore, the temperature and humidity data of the switch cabinet include external environment temperature, external environment humidity, external environment wind speed, external environment pressure, heat generation inside the switch cabinet and time node data.

[0039] The beneficial effects of the present invention are:

[0040] The feature extraction function of convolutional neural network is utilized, and the convolutional neural network and long short-term memory neural network are combined to construct a CNN-LSTM temperature and humidity prediction model, so as to adapt to the temperature and humidity prediction scenarios with large data volumes. The global exploration and local convergence capabilities of the sparrow optimization algorithm are further utilized and applied to the training process of the constructed temperature and humidity prediction model to avoid the fitting transition or under-fitting phenomenon of the constructed CNN-LSTM temperature and humidity prediction model in the prediction process, improve the prediction accuracy, and then improve the accuracy of the subsequent switch cabinet condensation abnormality judgment, and ensure the safe operation of the switch cabinet, so as to avoid the fitting transition or under-fitting phenomenon of the constructed temperature and humidity prediction model in the prediction process, improve the prediction accuracy, and then improve the accuracy of the subsequent switch cabinet condensation abnormality judgment.

[0041] After the switch cabinet condensation abnormality is determined, in order to optimize the temperature and humidity adjustment efficiency and adjustment effect, a refined adjustment strategy is formulated based on the temperature and humidity forecast data at each time point, and the specific time node and adjustment value for adjustment are determined to avoid condensation in the switch cabinet and ensure the safe operation of the switch cabinet. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of a process of the present invention;

[0043] Figure 2 It is a schematic diagram of the network structure of the convolutional neural network part of a CNN-LSTM temperature and humidity prediction model according to an embodiment of the present invention;

[0044] Figure 3 It is a schematic diagram of the network structure of the long short-term memory neural network part in a CNN-LSTM temperature and humidity prediction model in an embodiment of the present invention;

[0045] Figure 4 It is a model parameter optimization flow chart of a sparrow optimization algorithm according to an embodiment of the present invention;

[0046] Figure 5 This is a comparison chart of the temperature prediction value and the actual value output by a CNN-LSTM temperature and humidity prediction model in an embodiment of the present invention;

[0047] Figure 6 It is a comparison chart of humidity prediction value and actual value output by a CNN-LSTM temperature and humidity prediction model in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0049] Example:

[0050] Switch cabinet temperature and humidity monitoring method based on multivariate neural network, such as Figure 1 As shown, including:

[0051] Collect the temperature and humidity data of the switch cabinet, pre-process the temperature and humidity data, and build the corresponding training data set;

[0052] Construct a CNN-LSTM temperature and humidity prediction model, and train the CNN-LSTM temperature and humidity prediction model by combining the sparrow optimization algorithm and training data set;

[0053] Obtain the current switch cabinet temperature and humidity data, input the trained CNN-LSTM temperature and humidity prediction model, and obtain the switch cabinet temperature and humidity prediction data;

[0054] The switch cabinet condensation abnormality is judged according to the temperature and humidity prediction data, and the switch cabinet adjustment strategy is set according to the switch cabinet condensation abnormality judgment result, and the temperature and humidity in the switch cabinet are adjusted according to the switch cabinet adjustment strategy.

[0055] The temperature and humidity data of the switch cabinet include external environment temperature, external environment humidity, external environment wind speed, external environment pressure, heat generation inside the switch cabinet and time node data.

[0056] The switch cabinet can be divided into four compartments: terminal room, trolley room, busbar room and cable room. Each compartment is equipped with corresponding temperature and humidity sensors and temperature and humidity adjustment objects.

[0057] The objects for executing the temperature and humidity adjustment include controllable heaters, dehumidifiers, fans and other devices that can adjust the temperature or humidity.

[0058] When collecting the temperature and humidity data of the switch cabinet, first collect the external environment data of the switch cabinet through a small meteorological station, such as external environment temperature, external environment humidity, external environment wind speed, external environment pressure and other data, and then use four temperature and humidity sensors to collect the internal temperature and humidity data of the corresponding compartments of the switch cabinet.

[0059] The CNN-LSTM temperature and humidity prediction model is constructed, and the CNN-LSTM temperature and humidity prediction model is trained by combining the sparrow optimization algorithm and the training data set, including:

[0060] Initialize the CNN-LSTM temperature and humidity prediction model;

[0061] The parameter configuration of the CNN-LSTM temperature and humidity prediction model was optimized by using the sparrow optimization algorithm, and the CNN-LSTM temperature and humidity prediction model with updated parameters was retrained according to the training data set;

[0062] Obtain the model performance parameters of the CNN-LSTM temperature and humidity prediction model before and after parameter update, and compare the model performance parameters before and after the update;

[0063] If the updated model performance parameters are greater than the model performance parameters before the update, the parameter configuration of the currently updated CNN-LSTM temperature and humidity prediction model will continue to be optimized through the sparrow optimization algorithm until the updated model performance parameters are less than or equal to the model performance parameters before the update;

[0064] If the updated model performance parameters are less than or equal to the model performance parameters before the update, the parameter configuration optimization of the CNN-LSTM temperature and humidity prediction model is terminated, and the trained CNN-LSTM temperature and humidity prediction model is obtained.

[0065] Utilizing the feature extraction function of convolutional neural network, the convolutional neural network CNN and the long short-term memory neural network LSTM are combined to construct a CNN-LSTM temperature and humidity prediction model.

[0066] The convolutional neural network mainly includes a convolutional layer, a pooling layer and a fully connected layer. The network structure of the convolutional neural network part of the CNN-LSTM temperature and humidity prediction model in this embodiment is as follows: Figure 2 shown.

[0067] Among them, the convolution layer can extract the nonlinear temperature and humidity prediction input features in the original input data, and mine the internal connections of the data, reduce the network complexity, and reduce the number of connections between convolution layers. The temperature and humidity prediction input features are then reduced in dimension and secondary features are extracted through the pooling layer. Finally, the local features extracted by the convolution unit are summarized through the fully connected layer.

[0068] The long short-term memory neural network consists of three gates, namely the forget gate, input gate, and output gate. Through the above three gates, the input data feature information such as external ambient temperature, external ambient humidity, external ambient wind speed, external ambient pressure, and heat generation in the switch cabinet extracted by the convolutional neural network can be retained or discarded according to certain calculation rules, thereby characterizing the sequence data with time and space correlation to achieve the purpose of temperature and humidity prediction.

[0069] The forget gate of the long short-term memory neural network determines whether to retain or discard feature information, which can be expressed as:

[0070] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0071] Among them, f t is the forget gate state matrix at time t, h t-1 is the hidden state at the previous moment of time t, σ is the sigmoid activation function, W f is the forget gate weight matrix, b f is the forget gate constant matrix.

[0072] The input gate is used to update the cell state and determine which feature information needs to be retained according to certain calculation rules. Specifically, the calculation rules of the input gate are as follows:

[0073] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0074]

[0075]

[0076] Among them, i t , and C t are the input gate, candidate cell state and new cell state matrix at time t, tanh is the activation function, W i and W c are the weight matrices of the input gate and candidate cell states, respectively, and b i and b c are all input gate parameter matrices.

[0077] The output gate controls the feature information of the final output, and its calculation rules are as follows:

[0078] O t =σ(W o ·[h t-1 ,x t ]+b o );

[0079] h t =O t tanh(C t );

[0080] Among them, O t is the state matrix of the output gate at time t, W o is the weight matrix of the output gate, b o is the output gate parameter matrix.

[0081] The network structure of the long short-term memory neural network part in the CNN-LSTM temperature and humidity prediction model in this embodiment is as follows: Figure 3 As shown, where h t is the hidden state at time t.

[0082] In order to avoid overfitting or underfitting caused by setting model parameters by experience or grid optimization, the parameter configuration of the CNN-LSTM temperature and humidity prediction model is optimized by the sparrow optimization algorithm.

[0083] The sparrow optimization algorithm has good global exploration and local development capabilities. It can take into account all factors in the population and move the sparrows in the population toward the global optimal value, thereby quickly converging near the optimal value.

[0084] There are essentially two steps in the parameter configuration optimization of this CNN-LSTM temperature and humidity prediction model. One is the initialization stage, which is to set the initial parameters and conduct preliminary training. The model performance obtained in this stage may not be ideal.

[0085] The other is the optimization and training stage. For the CNN-LSTM temperature and humidity prediction model trained in the initialization stage, the sparrow optimization algorithm is used to search for a better model parameter configuration of the CNN-LSTM temperature and humidity prediction model. In this process, the sparrow optimization algorithm will continuously update the model parameters and retrain the model using the updated parameters.

[0086] The model performance parameters may be prediction accuracy, loss function value, etc., and the model performance parameters may be directly used as the fitness function of the sparrow optimization algorithm. The comparison of the model performance parameters before and after the update described in this embodiment is the step of using the fitness function to perform optimization iteration in the sparrow optimization algorithm.

[0087] In addition to stopping the model parameter optimization iteration by comparing the model performance parameters, the sparrow optimization algorithm can also be converged by setting the maximum number of iterations.

[0088] The initialization of the CNN-LSTM temperature and humidity prediction model and the optimization of the parameter configuration of the CNN-LSTM temperature and humidity prediction model by the sparrow optimization algorithm include:

[0089] Set the initial parameters of the CNN-LSTM temperature and humidity prediction model, randomly select part of the data in the training data set to train the CNN-LSTM temperature and humidity prediction model, and calculate the corresponding model performance parameters before updating;

[0090] The model parameters of the CNN-LSTM temperature and humidity prediction model are used as the initial population of the sparrow optimization algorithm to update the model parameters of the CNN-LSTM temperature and humidity prediction model;

[0091] Reselect some data in the training data set to train the updated CNN-LSTM temperature and humidity prediction model, and calculate the corresponding updated model performance parameters.

[0092] Among them, when using the sparrow optimization algorithm to optimize model parameters, it is necessary to first clarify the model parameters to be optimized, such as weights, biases, learning rates, etc., and then use these model parameters as the initial population of the sparrow optimization algorithm.

[0093] Part of the data in the training data set is randomly selected to train the CNN-LSTM temperature and humidity prediction model, and the corresponding model performance parameters before updating are calculated to provide a data basis for the optimization iteration of the sparrow optimization algorithm.

[0094] In order to further avoid the overfitting phenomenon of the CNN-LSTM temperature and humidity prediction model, it is necessary to ensure the use of independent validation sets and test sets to evaluate the performance of the model. Therefore, in the process of optimizing the model parameters, part of the data in the training data set is randomly selected to train the CNN-LSTM temperature and humidity prediction model.

[0095] The flowchart of model parameter optimization using the sparrow optimization algorithm is as follows Figure 4 shown.

[0096] Specifically, after selecting the model parameters as the initial population of the sparrow optimization algorithm, the population is initialized first. Each sparrow in the population represents a possible solution to the model parameters. Parameters such as population size, maximum number of iterations, number of discoverers, and number of joiners are set.

[0097] Each sparrow is randomly generated, that is, the initial position of the model parameter solution, and then the fitness value of each sparrow is calculated. The fitness value in the sparrow optimization algorithm in this embodiment is the model performance parameter.

[0098] Then, the population is sorted through foraging behavior, grouping behavior, jumping behavior and escape behavior to determine the optimal position of individual sparrows and the corresponding fitness value, and the corresponding discoverer position, joiner position and vigilant position are refreshed, thereby refreshing the sparrow position and determining the optimal fitness value.

[0099] Finally, it is determined whether the comparison result reaches the stopping condition, that is, whether the maximum number of iterations is reached, or the optimal fitness value before the update is less than or equal to the optimal fitness value after the update. If the stopping condition is reached, the iteration of the sparrow optimization algorithm is terminated. If not, it returns to the population classification step to re-determine the individual optimal position and corresponding fitness value of each sparrow.

[0100] The switch cabinet condensation abnormality judgment based on the temperature and humidity prediction data includes:

[0101] Set the unit time and calculate the corresponding dew point temperature based on the temperature and humidity forecast data within each unit time;

[0102] Compare the temperature component in the temperature and humidity prediction data in each unit time with the dew point temperature of the switch cabinet to determine the dew point temperature difference in each unit time;

[0103] The corresponding humidity change rate is determined according to the maximum and minimum values ​​of the humidity component in the temperature and humidity prediction data within each unit time, and the corresponding condensation risk level is calculated according to the humidity change rate and dew point temperature difference corresponding to each unit time.

[0104] Since the dew point temperature is not a fixed value, it is affected by two factors, temperature and humidity. Different dew point temperatures may appear in different time periods. Therefore, in order to ensure the accuracy of condensation abnormality judgment, the unit time is set to calculate the dew point temperature corresponding to each unit time based on the predicted temperature and humidity forecast data.

[0105] In this embodiment, the dew point temperature can be calculated specifically by using the Magras formula or the like.

[0106] Since the occurrence of condensation is related to both temperature and humidity, in addition to the temperature difference, the rapid increase in humidity will also increase the possibility of condensation. Therefore, the humidity change rate is also used as a basis for judging the condensation risk level.

[0107] When calculating the corresponding condensation risk level according to the humidity change rate and dew point temperature difference corresponding to each unit time, the corresponding condensation risk value can be calculated by weighted summation, and then the corresponding condensation risk level can be determined according to the condensation risk value.

[0108] Specifically, the condensation risk level per unit time may be determined by setting the upper and lower limits of the condensation risk value corresponding to each condensation risk level, and then matching the condensation risk value per unit time with each condensation risk level.

[0109] The step of setting a switch cabinet adjustment strategy according to the switch cabinet condensation abnormality judgment result and adjusting the temperature and humidity in the switch cabinet according to the switch cabinet adjustment strategy includes:

[0110] Compare the condensation risk levels of every two adjacent unit times, and select the adjustment nodes according to the comparison results and the corresponding condensation risk levels;

[0111] Determine the temperature component and humidity component in the temperature and humidity prediction data of each adjustment node, and select the adjustment execution object according to the temperature component and the humidity component;

[0112] The temperature adjustment value or humidity adjustment value of the adjustment execution object is set according to the humidity change rate and the dew point temperature difference corresponding to each adjustment node.

[0113] In order to ensure the temperature and humidity adjustment efficiency of the switch cabinet, the adjustment nodes are screened according to the corresponding condensation risk level, and then the temperature and humidity conditions corresponding to the adjustment nodes are determined to determine the appropriate adjustment execution objects.

[0114] If one of the adjustment nodes has a higher condensation risk level due to a lower temperature, then an adjustment execution object such as a heater that adjusts the temperature is selected for adjustment.

[0115] If the condensation risk level is high due to excessive humidity, select a dehumidifier or other humidity control execution object for adjustment.

[0116] The specific adjustment value can be set according to the adjustment dimension of the adjustment execution object through the corresponding humidity change rate or dew point temperature difference.

[0117] The step of comparing the condensation risk levels of every two adjacent unit times and selecting the adjustment node according to the comparison result and the corresponding condensation risk level includes:

[0118] Starting from the first unit time, the condensation risk levels of adjacent unit times are compared in sequence;

[0119] When the level difference of the condensation risk level between adjacent unit times does not exceed the preset threshold, the higher condensation risk level in the adjacent unit time is selected for comparison with the preset condensation risk level;

[0120] If the display exceeds the preset condensation risk level, the corresponding unit time is selected as the adjustment node, and the next unit time is selected to compare the condensation risk levels of adjacent unit times and select the adjustment nodes;

[0121] If the display does not exceed the preset condensation risk level, the lower condensation risk level in the adjacent unit time is used as the condensation risk level of the two adjacent unit time, and the condensation risk level of the adjacent unit time is compared with the next unit time and the adjustment node is selected;

[0122] When the level difference of condensation risk levels between adjacent unit times exceeds a preset threshold, the unit time with a higher condensation risk level in the adjacent unit times is selected as the adjustment node.

[0123] Whether to merge unit times is determined based on the condensation risk levels of adjacent unit times, thereby determining the corresponding adjustment nodes to optimize the adjustment efficiency of the temperature and humidity of the switch cabinet.

[0124] Specifically, starting from the first unit time, the condensation risk levels of adjacent unit times are compared in sequence to determine whether to merge the adjacent unit times.

[0125] Taking the comparison of condensation risk levels of unit time A, unit time B, unit time C, unit time D and unit time E as an example, the condensation risk level of unit time A and unit time B is level 1, the condensation risk level of unit time C is level 4, the condensation risk level of unit time D is level 5, and the condensation risk level of unit time E is level 3. The preset level difference is 2 levels, and the preset condensation risk level is level 3.

[0126] First, compare the condensation risk levels of unit time A and unit time B. Since the condensation risk levels of unit time A and unit time B are the same and neither exceeds level 3, unit time A and unit time B are merged.

[0127] After the merger, the condensation risk level of the merged time period is determined to be level 1, which is compared with the condensation risk level of unit time C. The condensation risk level exceeds level 2, so unit time C is set as one of the adjustment nodes.

[0128] Starting from unit time D, the condensation risk level is compared with that of unit time E. The level difference between the two is 2 levels, which does not exceed the preset top level difference. Therefore, the condensation risk level of the higher unit time D is compared with the preset condensation risk level. Since the condensation risk level of unit time D exceeds level 4, unit time D is set as one of the adjustment nodes.

[0129] The current switch cabinet temperature and humidity data is obtained, and the trained CNN-LSTM temperature and humidity prediction model is input to obtain the switch cabinet temperature and humidity prediction data, including:

[0130] Obtain the current switch cabinet temperature and humidity data, pre-process the switch cabinet temperature and humidity data, and input the trained CNN-LSTM temperature and humidity prediction model;

[0131] The output prediction data of the CNN-LSTM temperature and humidity prediction model is denormalized to obtain the temperature and humidity prediction data of the switch cabinet.

[0132] Since the current switch cabinet temperature and humidity data contains data of different dimensions, it is necessary to preprocess the switch cabinet temperature and humidity data, such as removing, filling and normalizing abnormal data, and then input it into the CNN-LSTM temperature and humidity prediction model.

[0133] Therefore, after the CNN-LSTM temperature and humidity prediction model outputs the prediction data, the output prediction data needs to be denormalized to obtain specific temperature and humidity prediction data.

[0134] Among them, the normalization process can be performed through the minimum-maximum normalization method or the standardization method, and when performing the subsequent denormalization process, the parameters used in the normalization process must be kept consistent to ensure the accuracy of the results.

[0135] Taking the temperature and humidity prediction results within a period of time as an example, the comparison diagrams of the temperature and humidity prediction values ​​and the actual values ​​output by the CNN-LSTM temperature and humidity prediction model established in this embodiment are as follows: Figure 5 and Figure 6 shown.

[0136] Among them, combined Figure 5 and Figure 6 It can be seen that the CNN-LSTM temperature and humidity prediction model has the following evaluation indicators for temperature prediction: RMSE = 1.3746, MSE = 1.8897, MAE = 0.94949, MAPE = 0.066133; the optimal number of neurons is 64, the optimal initial learning rate is 0.0030592, and the optimal L2 regularization coefficient is 0.00062006.

[0137] The CNN-LSTM temperature and humidity prediction model has the following evaluation indicators for humidity prediction: RMSE = 7.1548, MSE = 51.1912, MAE = 4.9513, MAPE = 0.068431, the optimal number of neurons is 108, the optimal initial learning rate is 0.0067007, and the optimal L2 regularization coefficient is 0.0027243.

[0138] It can be seen that the CNN-LSTM temperature and humidity prediction model established in this embodiment has good temperature and humidity prediction accuracy, and can effectively improve the efficiency of subsequent switch cabinet condensation prevention.

[0139] The above-described embodiment is only a preferred solution of the present invention and does not limit the present invention in any form. There are other variations and modifications without exceeding the technical solution described in the claims.

Claims

1. A switch cabinet temperature and humidity monitoring method based on a multivariate neural network, characterized in that: include: Collect the temperature and humidity data of the switch cabinet, pre-process the temperature and humidity data, and build the corresponding training data set; Construct a CNN-LSTM temperature and humidity prediction model, and train the CNN-LSTM temperature and humidity prediction model by combining the sparrow optimization algorithm and training data set; Obtain the current switch cabinet temperature and humidity data, input the trained CNN-LSTM temperature and humidity prediction model, and obtain the switch cabinet temperature and humidity prediction data; The switch cabinet condensation abnormality is judged according to the temperature and humidity prediction data, and the switch cabinet adjustment strategy is set according to the switch cabinet condensation abnormality judgment result, and the temperature and humidity in the switch cabinet are adjusted according to the switch cabinet adjustment strategy.

2. The switch cabinet temperature and humidity monitoring method based on multivariate neural network according to claim 1 is characterized in that: The CNN-LSTM temperature and humidity prediction model is constructed, and the CNN-LSTM temperature and humidity prediction model is trained by combining the sparrow optimization algorithm and the training data set, including: Initialize the CNN-LSTM temperature and humidity prediction model; The parameter configuration of the CNN-LSTM temperature and humidity prediction model was optimized by using the sparrow optimization algorithm, and the CNN-LSTM temperature and humidity prediction model with updated parameters was retrained according to the training data set; Obtain the model performance parameters of the CNN-LSTM temperature and humidity prediction model before and after parameter update, and compare the model performance parameters before and after the update; If the updated model performance parameters are greater than the model performance parameters before the update, the parameter configuration of the currently updated CNN-LSTM temperature and humidity prediction model will continue to be optimized through the sparrow optimization algorithm until the updated model performance parameters are less than or equal to the model performance parameters before the update; If the updated model performance parameters are less than or equal to the model performance parameters before the update, the parameter configuration optimization of the CNN-LSTM temperature and humidity prediction model is terminated, and the trained CNN-LSTM temperature and humidity prediction model is obtained.

3. The switch cabinet temperature and humidity monitoring method based on multivariate neural network according to claim 2 is characterized in that: The initialization of the CNN-LSTM temperature and humidity prediction model and the optimization of the parameter configuration of the CNN-LSTM temperature and humidity prediction model by the sparrow optimization algorithm include: Set the initial parameters of the CNN-LSTM temperature and humidity prediction model, randomly select part of the data in the training data set to train the CNN-LSTM temperature and humidity prediction model, and calculate the corresponding model performance parameters before updating; The model parameters of the CNN-LSTM temperature and humidity prediction model are used as the initial population of the sparrow optimization algorithm to update the model parameters of the CNN-LSTM temperature and humidity prediction model; Reselect some data in the training data set to train the updated CNN-LSTM temperature and humidity prediction model, and calculate the corresponding updated model performance parameters.

4. The switch cabinet temperature and humidity monitoring method based on multivariate neural network according to claim 1 is characterized in that: The judgment of abnormal condensation of the switch cabinet according to the temperature and humidity prediction data includes setting a unit time and calculating the corresponding dew point temperature according to the temperature and humidity prediction data within each unit time; Compare the temperature component in the temperature and humidity prediction data in each unit time with the dew point temperature of the switch cabinet to determine the dew point temperature difference in each unit time; The corresponding humidity change rate is determined according to the maximum and minimum values ​​of the humidity component in the temperature and humidity prediction data within each unit time, and the corresponding condensation risk level is calculated according to the humidity change rate and dew point temperature difference corresponding to each unit time.

5. The switch cabinet temperature and humidity monitoring method based on multivariate neural network according to claim 4 is characterized in that: The switch cabinet adjustment strategy is set according to the switch cabinet condensation abnormality judgment result, and the temperature and humidity in the switch cabinet are adjusted according to the switch cabinet adjustment strategy. include, Compare the condensation risk levels of every two adjacent unit times, and select the adjustment nodes according to the comparison results and the corresponding condensation risk levels; Determine the temperature component and humidity component in the temperature and humidity prediction data of each adjustment node, and select the adjustment execution object according to the temperature component and the humidity component; The temperature adjustment value or humidity adjustment value of the adjustment execution object is set according to the humidity change rate and the dew point temperature difference corresponding to each adjustment node.

6. The switch cabinet temperature and humidity monitoring method based on multivariate neural network according to claim 5 is characterized in that: The condensation risk levels of every two adjacent unit times are compared, and the adjustment nodes are selected according to the comparison results and the corresponding condensation risk levels. include, Starting from the first unit time, the condensation risk levels of adjacent unit times are compared in sequence; When the level difference of the condensation risk level between adjacent unit times does not exceed the preset threshold, the higher condensation risk level in the adjacent unit time is selected for comparison with the preset condensation risk level; If the display exceeds the preset condensation risk level, the corresponding unit time is selected as the adjustment node, and the next unit time is selected to compare the condensation risk levels of adjacent unit times and select the adjustment nodes; If the display does not exceed the preset condensation risk level, the lower condensation risk level in the adjacent unit time is used as the condensation risk level of the two adjacent unit time, and the condensation risk level of the adjacent unit time is compared with the next unit time and the adjustment node is selected; When the level difference of condensation risk levels between adjacent unit times exceeds a preset threshold, the unit time with a higher condensation risk level in the adjacent unit times is selected as the adjustment node.

7. The switch cabinet temperature and humidity monitoring method based on multivariate neural network according to claim 1 is characterized in that: The current switch cabinet temperature and humidity data is obtained, and the trained CNN-LSTM temperature and humidity prediction model is input to obtain the switch cabinet temperature and humidity prediction data, including: Obtain the current switch cabinet temperature and humidity data, pre-process the switch cabinet temperature and humidity data, and input the trained CNN-LSTM temperature and humidity prediction model; The output prediction data of the CNN-LSTM temperature and humidity prediction model is denormalized to obtain the temperature and humidity prediction data of the switch cabinet.

8. The switch cabinet temperature and humidity monitoring method based on multivariate neural network according to claim 1 is characterized in that: The switch cabinet temperature and humidity data includes external environment temperature, external environment humidity, external environment wind speed, external environment pressure, heat generation inside the switch cabinet and time node data.

Images