Switch cabinet temperature prediction method based on multi-point determination and domain incremental learning

Abstract

The invention relates to a switch cabinet temperature prediction method based on multi-point determination and domain incremental learning, and the method comprises the steps: constructing a switch cabinet temperature prediction model which comprises a feature extraction module and a prediction head module; collecting a temperature sequence sample of the switch cabinet; wherein the temperature sequence sample comprises a normal state sample and a fault state sample; training the switch cabinet temperature prediction model by using the temperature sequence sample to obtain a normal state temperature prediction model and a fault state temperature prediction model; and predicting the temperature sequence verification sets in normal and fault states, inputting the trained prediction model, and performing real-time temperature prediction of the switch cabinet by adopting a multi-point judgment method. The method realizes high-precision temperature prediction without depending on a large amount of fault state temperature sequence data, and has remarkable technical and economic advantages.

Description

Technical Field

[0001] This invention relates to the field of intelligent diagnostics and condition monitoring technology for switchgear, and in particular to a method for predicting switchgear temperature based on multi-point determination and domain incremental learning. Background Technology

[0002] Conductor corrosion, poor contact, and high ambient temperatures can all lead to increased local resistance in conductors, causing overheating. Delayed detection and repair of overheating faults can result in equipment burn-out, and in severe cases, fire, posing a significant threat to personnel and equipment safety. Furthermore, consistently high temperatures can cause insulation aging, triggering secondary faults. Therefore, under normal conditions, equipment temperature monitoring is necessary for equipment condition assessment, and under fault conditions, accurate prediction of temperature rise is crucial for timely troubleshooting. Thus, accurate temperature prediction of critical points in high-voltage switchgear under both normal and fault conditions is of paramount importance for ensuring the safe and stable operation of high-voltage switchgear.

[0003] Deep learning models are widely used in prediction due to their powerful ability to extract temporal features. Under normal conditions, there are numerous sequence samples of temperature measurement points in switchgear, and the prediction accuracy for switchgear under normal conditions has reached a high level. However, under fault conditions, because the probability of fault occurrence is low, the time series samples of switchgear temperature measurement points are small, leading to overfitting problems in the trained prediction models and extremely low prediction accuracy. Furthermore, since the temperature development trends of switchgear under normal and fault conditions are completely different, prediction models trained with a large sample size of normal temperature sequence samples cannot accurately predict fault conditions. Therefore, current high-voltage switchgear temperature prediction suffers from insufficient prediction accuracy under fault conditions.

[0004] Current solutions to this problem mainly involve feature tracking through attention mechanisms or reconstruction and compositing of complex features using mode decomposition methods. However, under fault conditions, the temporal characteristics of the temperature sequence in high-voltage switchgear differ significantly from those under normal conditions, making attention mechanisms difficult to follow. Furthermore, the temperature sequence changes under fault conditions are non-periodic, making it difficult for mode decomposition methods to effectively extract temperature change features. Both types of methods have significant room for improvement in temperature prediction accuracy. To improve the operational efficiency of high-voltage switchgear, a temperature prediction method for high-voltage switchgear that can address the small-sample prediction problem needs to be proposed. Summary of the Invention

[0005] Existing technologies face difficulties in analyzing the temporal features of switchgear temperature sequences under fault conditions. After training with a large number of normal-state temperature time-series samples, fault-state samples become overwhelming, causing the model's focus to be primarily on normal-state samples, while neglecting fault-state samples. This results in severely insufficient accuracy in predicting switchgear temperature under fault conditions. Therefore, it is necessary to train fault-state samples separately while avoiding overfitting. Domain incremental learning methods share the same feature extraction module. By incrementally training switchgear temperature sequence samples under both normal and fault conditions, this method can utilize a large number of normal-state samples to assist training, improving generalization while adapting to the temperature sequence features under fault conditions. Furthermore, since the distribution domain of the collected samples needs to be determined during prediction to select the prediction module, a sample domain determination method also needs to be proposed.

[0006] To address this, this invention proposes a switchgear temperature prediction method based on multi-point decision-making and domain incremental learning. The aim is to avoid overfitting in fault-state temperature prediction by employing multi-point probabilistic decision-making and domain incremental learning, while simultaneously determining the sample domain during actual prediction to select the appropriate prediction module. This method trains a highly generalizable prediction model for both normal and fault states using domain incremental learning. Simultaneously, it calculates the prediction error distribution of the prediction model, formulates a multi-point decision-making method, and determines the sample domain in real time to select the appropriate prediction model. This significantly improves the temperature prediction accuracy under fault states while maintaining the accuracy of high-voltage switchgear temperature prediction under normal conditions.

[0007] To achieve the above objectives, the present invention provides the following solution:

[0008] A method for predicting switchgear temperature based on multi-point decision and domain incremental learning includes:

[0009] A temperature prediction model for switchgear is constructed, wherein the temperature prediction model for switchgear includes: a feature extraction module and a prediction head module;

[0010] Collect temperature sequence samples of the switchgear; wherein, the temperature sequence samples include: normal state samples and fault state samples;

[0011] The temperature prediction model of the switchgear is trained using the temperature sequence samples to obtain a normal state temperature prediction model and a fault state temperature prediction model.

[0012] The temperature sequence validation set under normal and fault conditions is predicted and input into the trained prediction model. A multi-point decision method is used to predict the real-time temperature of the switchgear.

[0013] Optionally, the input to the switchgear temperature prediction model is the temperature value at a previous time, and the output is the temperature value at the next time.

[0014] Optionally, training the switchgear temperature prediction model using the temperature sequence samples includes:

[0015] The normal state samples are used to train the switch cabinet temperature prediction model to obtain the normal state temperature prediction model.

[0016] The fixed feature extraction module uses the fault state samples for domain incremental learning training to obtain a fault state temperature prediction model.

[0017] Optionally, when training the switchgear temperature prediction model using the normal state samples, the loss function used is:

[0018] ;

[0019] in, The loss function for training normal state samples; For sample size; These are the model's predicted values; This is the actual temperature value;

[0020] When using the aforementioned fault state samples for domain incremental learning training, the loss function employed is:

[0021] ;

[0022] in, The loss function used for training fault state samples; To train weights for domain increment, The loss function is used to train the fault state samples.

[0023] Optionally, the temperature sequence validation set under normal and fault conditions is used for prediction, and the input to the trained prediction model includes:

[0024] The trained prediction model is used to predict the temperature sequence validation set under normal and fault conditions, respectively, and the prediction error under normal and fault conditions is obtained. The probability distribution of the prediction error under normal and fault conditions is estimated by kernel density estimation method.

[0025] Based on the probability distribution of prediction error under normal and fault conditions, the upper quantile of prediction error is calculated.

[0026] Optionally, the probability distribution of prediction errors under normal and fault conditions is as follows:

[0027] ;

[0028] in, To predict the relative error probability density function, For the sample size involved in probability density estimation, To estimate the kernel function for kernel density, The sampled value of the i-th sample. Bandwidth is estimated for kernel density.

[0029] Optionally, calculating the upper quantile of the prediction error includes:

[0030] ;

[0031] ;

[0032] ;

[0033] ;

[0034] in, The probability is 1- The upper quantile, The relative error is no higher than The probability, To train weights for domain increment, To find the total number of product nodes, To find the product coefficient, To find the product nodes, For the sample size involved in probability density estimation, The sampled value of the i-th sample. For kernel density estimation bandwidth, To predict relative error Greater than The probability of the upper quantile. For weighted product nodes, To perform a linear transformation of the product nodes, Let be the probability point.

[0035] Optionally, a multi-point determination method can be used to predict the real-time temperature of the switchgear, including:

[0036] A normal-state temperature prediction model is used to predict the real-time temperature of the switchgear, and the prediction error is monitored in real time. When repeated... When the prediction error at each point is greater than that at the upper quantile, a fault state temperature prediction model is used for real-time temperature prediction. The model switching process is repeated as the state continues to change. The time window for state determination is 6 sampling points.

[0037] The beneficial effects of this invention are as follows:

[0038] This invention addresses the severe inaccuracy of temperature prediction for switchgear under fault conditions by proposing a temperature prediction method based on multi-point decision-making and domain incremental learning. Domain incremental learning solves the problem of small sample sizes under fault conditions. After training temperature prediction models for both normal and fault conditions, a multi-point decision-making method is proposed to determine the equipment status and select the appropriate prediction model. This invention achieves high-precision temperature prediction without relying on large amounts of fault condition temperature sequence data, demonstrating significant technical and economic advantages. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart and schematic diagram of a switchgear temperature prediction method based on multi-point determination and domain incremental learning, according to an embodiment of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] like Figure 1 As shown, this embodiment proposes a switchgear temperature prediction method based on multi-point decision and domain incremental learning, including:

[0044] A temperature prediction model for switchgear is constructed, wherein the temperature prediction model for switchgear includes: a feature extraction module and a prediction head module;

[0045] Collect temperature sequence samples of the switchgear; wherein, the temperature sequence samples include: normal state samples and fault state samples;

[0046] The temperature prediction model of the switchgear is trained using the temperature sequence samples to obtain a normal state temperature prediction model and a fault state temperature prediction model.

[0047] The temperature sequence validation set under normal and fault conditions is predicted and input into the trained prediction model. A multi-point decision method is used to predict the real-time temperature of the switchgear.

[0048] Specifically, this embodiment addresses the problem of low accuracy in predicting switchgear temperature under fault conditions by proposing a switchgear temperature prediction method based on multi-point determination and domain incremental learning. The implementation plan is as follows:

[0049] (1) Construct a temperature prediction model for switchgear. The model should include a feature extraction module and a prediction head module. Its input is the temperature value of the previous 5 time moments, and its output is the temperature value of the next time moment.

[0050] (2) Collect temperature sequence samples of fault state and a large number of normal state temperature sequence samples. First, use the normal state samples to train the temperature prediction model to obtain the normal state temperature prediction model. Then, fix the feature extraction module and use the fault state samples to perform domain incremental learning training to obtain the fault state temperature prediction model.

[0051] (3) Collect temperature sequence validation sets under normal and fault conditions, use the trained prediction model to make corresponding predictions on the validation sets, obtain the prediction errors under normal and fault conditions, use the kernel density estimation method to estimate the probability distribution of the prediction errors under normal and fault conditions, and calculate its upper quantile. Using the upper quantile limit of multiple points as the criterion, propose a multi-point determination method for switch cabinet status.

[0052] (4) Generally, the normal state prediction model is first used to predict the real-time temperature of the switch cabinet. When the system state changes, another model is used to predict the real-time temperature. When the state continues to change, the model switching process is repeated.

[0053] Specifically, in step (1), a temperature prediction model is constructed. The model must include a feature extraction module and a prediction head module. The feature extraction module adopts an LSTM model with the following parameters: input size: 5, hidden size: 64, and number of layers: 128. The prediction head module adopts a fully connected model with the following parameters: 64-521-1.

[0054] In step (2), a large number of normal-state temperature sequence samples are collected to train the normal-state temperature prediction model. The training loss function is:

[0055] (1);

[0056] In the formula, The loss function for training normal state samples; For sample size; These are the model's predicted values; This is the actual temperature value.

[0057] Then, the fixed feature extraction module uses fault state samples for domain incremental learning training, and the training loss function is:

[0058] (2);

[0059] In the formula, The loss function used for training fault state samples; The weights for domain increment training are set to 0.1 in this embodiment.

[0060] In step (3), after obtaining the temperature sequence validation set under normal and fault conditions, the trained prediction model is used to make corresponding predictions on the validation set to obtain the prediction errors under normal and fault conditions. The probability distribution of the prediction errors under normal and fault conditions is estimated using the kernel density estimation method. The specific calculation method is as follows:

[0061] The probability density function of the relative prediction error of the prediction model can be estimated by the following formula:

[0062] (3);

[0063] In the formula, The probability density function for predicting relative error; The sample size for probability density estimation; For kernel density estimation, this embodiment selects the Gaussian kernel function; The sample value of the i-th sample; Bandwidth is estimated for kernel density.

[0064] Under the condition of Gaussian kernel function, the optimal bandwidth is:

[0065] (4);

[0066] In the formula, This is the estimated standard deviation of the predicted relative error.

[0067] After obtaining the probability density function, based on the Gauss-Legendre numerical integration, the prediction relative error is no higher than [missing value]. The probability can be calculated using the following numerical values:

[0068] (5);

[0069] in,

[0070] (6);

[0071] In equations (5) and (6), To find the total number of product nodes; To find the product coefficient; To find the product nodes.

[0072] Since the probability distribution function of the relative prediction error is monotonically non-decreasing, the probability can be calculated as 1- Upper quantile:

[0073] (7);

[0074] in,

[0075] (8);

[0076] That is, the relative error of prediction Greater than The probability of the upper quantile is 1- .when When it is very large (0.99 in this embodiment), it can be considered that... It's a low-probability event, at the upper quantile. It can be used as a threshold for single-point judgment. The prediction errors between points are independent of each other; when the judgment point is chosen as... When all prediction errors are greater than the upper quantile, the probability of event B occurring is:

[0077] (9);

[0078] The larger the value, the closer the probability of the event occurring is to 0. When the probability is less than 1e-10 ( =5), which can be considered to have a probability close to 0, and is an impossible event.

[0079] Therefore, the multi-point determination method constructed in this embodiment is as follows: calculate the upper quantile of the prediction error, and when repeated... When the prediction error of each point is greater than the upper quantile, the event that remains unchanged can be regarded as an impossible event, that is, the state of the equipment has changed.

[0080] For step (4), specifically, firstly, a normal state prediction model is used to predict the real-time temperature of the switchgear, and the prediction error is monitored in real time. When repeated... When the prediction error at each point is greater than that at the upper quantile, another model is used for real-time temperature prediction. The model switching process is repeated as the state continues to change. The time window for state determination is 6 sampling points in this embodiment.

[0081] This embodiment addresses the problem of insufficient temperature prediction accuracy for switchgear under fault conditions by proposing a temperature prediction method based on multi-point decision-making and domain incremental learning. Domain incremental learning solves the small sample problem under fault conditions. After training temperature prediction models for normal and fault conditions respectively, a multi-point decision-making method is proposed to determine the equipment status and select the appropriate prediction model. This embodiment achieves high-precision temperature prediction without relying on a large amount of fault condition temperature sequence data, demonstrating significant technical and economic advantages.

[0082] Temperature sequences of high-voltage switchgear under normal and fault conditions were collected for training and testing. The results were compared with similar LSTM models and dual-stage attention based LSTM (DA-LSTM) models. The following two parameters were used for accuracy evaluation:

[0083] Maximum absolute percentage error (MAPE):

[0084] (10);

[0085] Root mean square error (RMSE):

[0086] (11);

[0087] In equations (10) to (11), The actual temperature value at time i; The predicted temperature value at time i; This represents the total number of time periods within the test set.

[0088] The prediction error comparison results are shown in Table 1:

[0089] Table 1 Comparison of prediction accuracy of various methods

[0090]

[0091] Observing the prediction accuracy of each method under normal conditions in Table 1, it can be seen that DA-LSTM has a better response to concentration changes due to its attention mechanism. Under normal conditions, this method has the highest prediction accuracy. LSTM has the same model and training data as the method in this embodiment, and its prediction accuracy is similar under normal conditions. The prediction accuracy of the two methods is not much different from that of DA-LSTM.

[0092] Under fault conditions, both LSTM and DA-LSTM methods exhibit poor tracking performance, resulting in high MAPE and RMSE. DA-LSTM, however, achieves higher prediction accuracy than LSTM due to its better tracking of data changes. The method proposed in this embodiment, by switching to a prediction model that addresses overfitting under fault conditions, effectively adapts to the temperature time-series distribution under fault conditions, achieving the highest prediction accuracy. Its MAPE and RMSE are significantly lower than the other two methods, with a reduction in MAPE of at least 4.153% and an RMSE reduction of at least 12.143℃ compared to the other methods.

[0093] This embodiment presents a switchgear temperature prediction method based on multi-point determination and domain incremental learning, comprising the following steps: Step 1: Construct a switchgear temperature prediction model. The model includes a feature extraction module and a prediction head module. Its input is the temperature values ​​of the previous 5 time points, and its output is the temperature value of the next time point. Step 2: Collect temperature sequence samples under fault conditions and a large number of temperature sequence samples under normal conditions. First, train the temperature prediction model using the normal condition samples to obtain a normal condition temperature prediction model. Then, fix the feature extraction module and use the fault condition samples for domain incremental learning training to obtain a fault condition temperature prediction model. Step 3: Collect a validation set of temperature sequences under normal and fault conditions. Use the trained prediction model to make corresponding predictions on the validation set to obtain the prediction errors under normal and fault conditions. Use the kernel density estimation method to estimate the probability distribution of the prediction errors under normal and fault conditions and calculate their upper quantiles. Using the upper quantile limit of multiple points as a criterion, a multi-point determination method for switchgear status is proposed. Step 4: First, use the normal state prediction model to predict the real-time temperature of the switchgear. When it is determined that the system state has changed, use another model to predict the real-time temperature. Repeat the model switching process when the state continues to change.

[0094] The technical solution of this embodiment will be further described below:

[0095] In step (1), the switch cabinet temperature prediction model adopts the LSTM+ fully connected model, the model is built using the PyTorch platform, and the same platform is used for training.

[0096] In step (2), the status of the switchgear equipment is monitored. Temperature sequence samples are collected in the background of the switchgear under normal and fault conditions. The number of temperature points under normal conditions is not less than 1000, and the number of temperature points under fault conditions is not less than 50.

[0097] In step (3), the multi-point determination method is an innovative method of the present invention. The probability density function must be estimated strictly in accordance with the method described in the present invention, and the upper quantile probability and the number of points in the multi-point determination must be kept consistent.

[0098] In step (4), this embodiment uses html+css+js+python to build the prediction platform. The historical temperature data is sent by the platform to the prediction function module through socket communication. After being processed by the background Python, it is sent to the front end for display and warning via Flask communication.

[0099] To verify the effectiveness of the proposed model, operational sample data from a 35kV switchgear was used for validation. After five rounds of online learning, the model's prediction error MAPE was no higher than 1.883%, and the prediction error RMSE was no higher than 1.995℃, significantly improving the accuracy of switchgear temperature prediction. Furthermore, the model maintained stable performance during 120 days of continuous operation in the field. This method can significantly improve the intelligence level of switchgear operation and maintenance and the reliability of power supply.

[0100] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A switchgear temperature prediction method based on multi-point judgment and domain incremental learning, characterized in that, The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet.

2. The switchgear temperature prediction method based on multi-point decision and domain incremental learning according to claim 1, characterized in that, The application relates to a temperature prediction model of a switch cabinet. 3.The switchgear temperature prediction method based on multi-point decision and domain incremental learning according to claim 1, characterized in that, The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet.

4. The switchgear temperature prediction method based on multi-point decision and domain incremental learning according to claim 3, characterized in that, The application relates to a temperature prediction model of a switch cabinet. ； wherein, the loss function trained for normal state samples; is the sample size; is the model predicted value; is the temperature true value; The application relates to a temperature prediction model of a switch cabinet. ； wherein, a loss function for training on fault condition samples; a domain-incremental training weight, a loss function for training on fault condition samples.

5. The switchgear temperature prediction method based on multi-point decision and domain incremental learning according to claim 1, characterized in that, The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet.

6. The switchgear temperature prediction method based on multi-point decision and domain incremental learning according to claim 5, characterized in that, The application relates to a temperature prediction model of a switch cabinet. ； wherein, is the relative error probability density function, is the sample size participating in the probability density estimation, is the kernel density estimation kernel function, is the sample sampling value of the i-th sample, is the kernel density estimation bandwidth.

7. The switchgear temperature prediction method based on multi-point decision and domain incremental learning according to claim 5, characterized in that, The application relates to a temperature prediction model of a switch cabinet. ； ； ； ； wherein, is the upper quantile for a probability of 1- is the probability of a relative error not higher than is the domain increment training weight, is the total number of quadrature nodes, is the quadrature coefficient, is the quadrature node, is the sample capacity participating in the probability density estimation, is the sample sampling value of the i-th sample, is the kernel density estimation bandwidth, is the probability of a prediction relative error greater than the upper quantile, is the weighted quadrature node, is the quadrature node linear transformation, is the probability point.​​ 8. The switchgear temperature prediction method based on multi-point decision and domain incremental learning according to claim 7, characterized in that, The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction model of a switch cabinet. The application relates to a temperature prediction The normal state temperature prediction model is used for real-time temperature prediction of the switch cabinet, and real-time monitoring of prediction errors is performed. When the prediction errors of multiple points are greater than the upper quantile points, the fault state temperature prediction model is used for real-time temperature prediction, and the model switching process is repeated when the state continues to change. is the length of the time window for state determination.

Images