Switch cabinet temperature prediction and early warning method and device, medium and equipment

Abstract

The invention discloses a switch cabinet temperature prediction and early warning method and device, a medium and equipment, and relates to the technical field of switch cabinet temperature prediction and early warning. Comprising the following steps: acquiring three groups of switch cabinet time sequence characteristics corresponding to switch cabinet current, sensor temperature and environment temperature by using Transform; splicing the three groups of switch cabinet time sequence features and then carrying out convolution operations of different scales to obtain switch cabinet local time features of different scales; the local time features of the switch cabinets with different scales are spliced and weighted, and weighted features of the switch cabinets are obtained; inputting the weighted features of the switch cabinet into the trained long-short-term memory recurrent neural network for temperature prediction, and outputting the temperature of the sensor in the future time period; and inputting the weighted features of the switch cabinet into the trained full-connection neural network to carry out temperature early warning, and carrying out temperature grade early warning on the sensor in a future time period.

Description

Technical Field

[0001] The present invention relates to the technical field of switch cabinet temperature prediction and early warning, and in particular to a switch cabinet temperature prediction and early warning method, device, medium and equipment. Background Art

[0002] Switchgear is used to open and close, control and protect electrical equipment during power generation, transmission and energy conversion in the power system to ensure safe and efficient operation of the power system. Given that the temperature of the equipment inside the switchgear may rise during long-term operation due to the thermal effect of current, equipment aging and other reasons, this temperature change is an early signal of equipment status change. Therefore, by monitoring and analyzing the temperature changes of the switchgear, its working status and potential faults can be predicted to help operators take measures before the equipment overheats and causes failures, thereby ensuring the normal operation of the power system and avoiding unexpected power outages or equipment damage.

[0003] In the prior art, Chinese patent CN117786481A discloses a method and system for early warning of temperature rise hazards in switch cabinets based on LSTM networks. The method includes: obtaining historical data of temperature rise parameters of plum blossom contacts in the switch cabinet, the temperature rise parameters at least including contact resistance, ambient temperature, current, and contact temperature data; sorting the historical data in the order of time nodes and calculating the expected temperature rise value at each time node; calculating the actual temperature rise value through the contact temperature data of two consecutive time nodes; calculating the error between the expected temperature rise value and the actual temperature rise value to obtain the first data; inputting the first data and historical data into the LSTM network for training to output a prediction model; obtaining the temperature rise parameters of the current state of the plum blossom contact, and inputting the temperature rise parameters of the current state into the prediction model to output the warning result.

[0004] However, due to the complex environment in which the switchgear is located, the corresponding time series data fluctuates unstably over a long time span, which may cause a large difference between the training data and the actual prediction data of the LSTM model. Only using the error between the expected temperature rise value and the actual temperature rise value as an additional input to assist LSTM in prediction makes it difficult to ensure the accuracy of LSTM's prediction of new switchgear data. Summary of the invention

[0005] Based on this, in order to solve the technical problems in the prior art, the present invention provides a switch cabinet temperature prediction and early warning method, device, medium and equipment.

[0006] The present invention provides a switch cabinet temperature prediction and early warning method, comprising:

[0007] Feature extraction is performed on the historical current time series data of the switch cabinet sensor node, the historical temperature time series data of the switch cabinet sensor node, and the historical ambient temperature time series data of the switch cabinet, respectively, to obtain the corresponding three groups of switch cabinet time series features;

[0008] Splice three groups of switch cabinet time series features to obtain the global time features of the switch cabinet; perform convolution operations of different scales on the global time features of the switch cabinet to extract local features of the switch cabinet at different time granularities to obtain local time features of the switch cabinet at different scales; splice the local time features of the switch cabinet at different scales to obtain multi-scale local time features of the switch cabinet; and weight the local time features of the switch cabinet whose contribution to temperature prediction is greater than the threshold contribution in the multi-scale local time features of the switch cabinet to obtain weighted features of the switch cabinet;

[0009] The weighted features of the switch cabinet are input into the trained long short-term memory recurrent neural network to obtain the temperature prediction results for the future time period; and the weighted features of the switch cabinet are input into the trained fully connected neural network to obtain the temperature level warning results for the future time period.

[0010] Furthermore, the performing convolution operations of different scales on the global time features of the switch cabinet specifically includes:

[0011] Performing a convolution operation on the switch cabinet global time feature through a 1×1 convolution layer, and fusing the switch cabinet global time feature with the output of the 1×1 convolution layer through a residual connection to obtain a first-scale switch cabinet local time feature;

[0012] Performing a convolution operation on the switch cabinet global time feature through a 1×3 convolution layer, and fusing the switch cabinet global time feature with the output of the 1×3 convolution layer through a residual connection to obtain a second-scale switch cabinet local time feature;

[0013] The global time features of the switch cabinet are convolved through two sequentially connected 1×3 convolutional layers, and the global time features of the switch cabinet are fused with the output of the second 1×3 convolutional layer through a residual connection to obtain the local time features of the switch cabinet at the third scale.

[0014] Furthermore, the weighting of the local time features of the multi-scale switch cabinet that contribute significantly to temperature prediction is achieved by adjusting the weights of the local time features of the multi-scale switch cabinet using a single-layer fully connected network with a sigmoid function.

[0015] Furthermore, the feature extraction of the historical current time series data of the switch cabinet sensor node, the historical temperature time series data of the switch cabinet sensor node and the historical ambient temperature time series data of the switch cabinet respectively includes:

[0016] The switchgear historical current time series data, switchgear historical temperature time series data, and switchgear historical ambient temperature time series data are respectively input into three trained Transformer models to capture the long-distance correlation of data points in each time series data in the time dimension and obtain the corresponding three groups of switchgear time series features.

[0017] The present invention provides a switch cabinet temperature prediction and early warning device, comprising:

[0018] A feature extraction module is used to extract features from the historical current time series data of the switch cabinet sensor node, the historical temperature time series data of the switch cabinet sensor node, and the historical ambient temperature time series data of the switch cabinet, respectively, to obtain three sets of corresponding switch cabinet time series features;

[0019] The feature fusion module is used to splice three groups of switch cabinet time series features to obtain the global time features of the switch cabinet; perform convolution operations of different scales on the global time features of the switch cabinet to extract the local features of the switch cabinet at different time granularities to obtain the local time features of the switch cabinet at different scales; splice the local time features of the switch cabinet at different scales to obtain the local time features of the switch cabinet at multiple scales; and weight the local time features of the switch cabinet whose contribution to temperature prediction is greater than the threshold contribution in the local time features of the multi-scale switch cabinet to obtain the weighted features of the switch cabinet;

[0020] The temperature prediction and warning module is used to input the weighted features of the switch cabinet into the trained long short-term memory recurrent neural network to obtain the temperature prediction results for the future time period; and input the weighted features of the switch cabinet into the trained fully connected neural network to obtain the temperature level warning results for the future time period.

[0021] The present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned switch cabinet temperature prediction and early warning method is implemented.

[0022] The present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the switch cabinet temperature prediction and early warning method when executing the program.

[0023] At least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects:

[0024] In the switch cabinet temperature prediction and early warning method provided by the present invention, after splicing different switch cabinet time series features to obtain the global information of the switch cabinet in a long span of time, the short-term change pattern of the switch cabinet in different time periods can be effectively captured through convolution operations of different scales, and then the local time features of the switch cabinet of different scales are spliced ​​and weighted, considering which short-term changes have a greater impact on temperature prediction, and can highlight those features that have a significant contribution to temperature changes, thereby improving the sensitivity of the extracted features to short-term changes. This feature extraction and integration method provides more comprehensive and detailed information for the prediction model by combining global time features and local time features, so that it can still maintain good prediction performance under changing environmental conditions, thereby improving the effect of temperature prediction on new switch cabinet data. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0026] Figure 1 A schematic diagram of a flow chart of a switch cabinet temperature prediction and early warning method provided by the present invention;

[0027] Figure 2 A schematic diagram of the Transformer model framework provided by the present invention;

[0028] Figure 3 A schematic diagram of a switch cabinet temperature prediction and early warning method framework provided by the present invention;

[0029] Figure 4 A schematic diagram of a framework for integrating features provided by the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in combination with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] Switchgear temperature prediction and early warning is an important topic, especially in power systems, where it is crucial to ensure the safe and stable operation of equipment. Ultra-long-term temperature prediction and early warning can give operators time to react and reduce power system losses. At present, most switchgear temperature warnings are based on real-time warnings of temperature measurement thresholds, and early warnings cannot be achieved; and some methods with early warnings do not have long enough early warning time and the prediction deviation is not ideal. Temperature monitoring methods have problems such as poor real-time performance and low prediction accuracy, which are difficult to meet the needs of modern power systems.

[0033] Based on this, the present invention provides a switch cabinet temperature prediction and early warning method.

[0034] Figure 1 The flow chart of the switch cabinet temperature prediction and early warning method of this embodiment is shown. Figure 1 The switch cabinet temperature prediction and early warning method is described in detail, which specifically includes the following steps:

[0035] S1: Extract features from the historical current time series data of the switch cabinet sensor node, the historical temperature time series data of the switch cabinet sensor node, and the historical ambient temperature time series data of the switch cabinet, and obtain the corresponding three groups of switch cabinet time series features. Specifically, it includes:

[0036] S101: Data processing.

[0037] The data includes three types, namely the node current of the switch cabinet, the sensor node temperature and the ambient temperature. The data needs to be processed into the following input vector format, the input data is: t =[e 1 ,e 1 ,e 1 ,…,e t ],[s 1 ,s 1 ,s 1 ,…,s t ],[a 1 ,a 1 ,a 1 ,…,a t ]; label data is y n =[e t+1 ,e t+2 ,e t+3 ,…,e t+n ]. Where e is the ambient temperature, s is the sensor temperature, a is the current size, e 1 ~e t represents the ambient temperature from the initial time to time t; s 1 ~s t represents the sensor temperature from the initial moment to the moment t; a 1 ~at Indicates the current magnitude from the initial moment to moment t; Indicates the temperature from time t+1 to time t+n.

[0038] S102: Input historical current data, historical temperature time series data, and historical switch cabinet ambient temperature data into three sets of trained Transformers respectively to capture the long-distance correlation of data points in each data in the time dimension, and obtain three sets of time series features: current time series features, sensor temperature time series features, and ambient temperature time series features.

[0039] Transformer has obvious advantages in processing long sequences and parallel computing. Its network structure is as follows: Figure 2 As shown in the figure, it includes an input layer Input, a position encoding Position Embedding and 7 multi-head self-attention mechanism modules connected in sequence. The multi-head self-attention mechanism module includes a multi-head attention Muti-headAttention and a fully connected layer FC (FullConnect).

[0040] The self-attention calculation formula is as follows:

[0041]

[0042] Among them, Q (query), K (key) and V (value) are obtained by linear transformation of the input vector, d k is the dimension of the key vector.

[0043] Transformer uses multi-head attention to enhance the representation ability of the model. In multi-head attention, the input sequence is transformed through different weight matrices separately, and then an independent attention mechanism is performed on each group.

[0044] Positional encoding:

[0045]

[0046] Where pos is the position, i is the dimension index, and d model is the dimension of the model.

[0047] By using the Transformer model's ability to extract long-term dependencies, overall features can be obtained from the data of multiple historical time periods of the switch cabinet, which makes the features more resistant to noise. The Transformer model can filter out some noise and unimportant information during the feature extraction process, making the data input to the LSTM model cleaner and more focused. Since the Transformer model has captured the deep features of the data, after fusing these features, the fusion vector contains comprehensive information from different sensors. This helps the LSTM model understand and predict the temperature changes of the switch cabinet more accurately. Since the LSTM model can more easily learn generalized representations, it can also make more accurate predictions on unseen data.

[0048] S2: splicing three groups of switchgear time series features to obtain the global time features of the switchgear; performing convolution operations of different scales on the global time features of the switchgear to extract the local features of the switchgear at different time granularities to obtain the local time features of the switchgear at different scales; splicing the local time features of the switchgear at different scales to obtain the multi-scale local time features of the switchgear; and weighting the local time features of the switchgear whose contribution to temperature prediction is greater than the threshold contribution in the multi-scale local time features of the switchgear to obtain the weighted features of the switchgear.

[0049] like Figure 3 As shown in the figure, the three groups of time series features are taken as Input_1, Input_2 and Input_3 respectively, and the three groups of time series features are spliced ​​through the concatenate operation. After splicing, they are input into three convolution branches for convolution operation extraction. The three convolution branches are respectively composed of a 1×1 convolution layer, a 1×3 convolution layer and two 1×3 convolution layers connected in sequence, and each convolution branch fuses the spliced ​​switch cabinet global time features with the corresponding convolution layer output through residual connection to obtain the scale switch cabinet local time features of the corresponding scale. After the three branches extract feature information in different time periods, they are spliced ​​for the second time through the concatenate operation, and then weighted through the fully connected layer. The fully connected layer uses the sigmoid activation function to assign weights to the extracted features and the second spliced ​​features.

[0050] S3: Input the weighted features of the switch cabinet into the trained long short-term memory recurrent neural network to obtain the temperature prediction results for the future time period; and input the weighted features of the switch cabinet into the trained fully connected neural network to obtain the temperature level warning results for the future time period.

[0051] Long short-term memory recurrent neural network (LSTM) is a deep learning model suitable for processing time series data. It can capture long-term dependencies in the data and is very suitable for temperature prediction. LSTM introduces memory cells (cellstate) and three gates (input gate, forget gate, and output gate) to control the flow of information, thereby solving the gradient vanishing problem of traditional recurrent neural networks (RNN) when dealing with long-term dependency problems. The input layer receives time series data, historical temperature data, and ambient temperature. The LSTM layer processes the input data through memory cells and gate mechanisms to capture long-term dependencies in the data. The forward calculation formula of LSTM is as follows:

[0052] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0053] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0054]

[0055] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0056] h t =o t *tanh(C t )

[0057] Among them, f t It is the forget gate, t is the input gate, is the candidate memory cell state, C t is the memory cell state, o t is the output gate, h t is the hidden state, x t is the input data, W and b are the weight and bias parameters, the subscripts of W and b represent the weight and bias parameters of the corresponding gates, σ is the sigmoid activation function, and tanh is the activation function. The prediction result finally output by the LSTM network does not need to go through the activation function.

[0058] The mathematical formula of the fully connected neural network is as follows:

[0059] h(x)=f(Wx+b)

[0060] This formula means that the input vector x undergoes a linear transformation through the weight matrix W and the bias vector b, and then undergoes a nonlinear transformation through the activation function f, and finally obtains the output vector h(x). Here, the activation function uses the softmax function.

[0061] The entire network optimization function is established as the following function:

[0062]

[0063] in is the weight, y i is the true temperature value of the i-th group of data, is the corresponding temperature prediction value; i is the true category of the warning of the i-th group of data, is the corresponding warning prediction category.

[0064] Considering the continuity of temperature change and realizing the early warning function of the switch cabinet, the network training and parameter update are carried out according to the optimization function. The steps are as follows:

[0065] (1) Input historical temperature (t1-tn), ambient temperature, and historical current data into the network to obtain the predicted temperature pt1 and calculate loss1;

[0066] (2) Add the predicted temperature pt1 to the historical temperature and remove the oldest temperature to obtain a new temperature array (t2-tn, pt1);

[0067] (3) Re-input the new temperature array into the network and repeat (1) and (2);

[0068] (4) Add the above losses;

[0069] (5) Update the network parameters based on the added loss.

[0070] based on Figure 4 A Transformer and LSTM network model is shown in the figure. The model inputs the sensor temperature, ambient temperature and current into three Transformers for time feature extraction. After fusing the extracted features and extracting local features, the extracted features are respectively input into the LSTM network and the FC network to predict the temperature and warning status of the switch cabinet. Through the designed loss function and training strategy, the ultra-long-term prediction and warning algorithm of the switch cabinet temperature is realized.

[0071] The above is a switch cabinet temperature prediction and early warning method provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding switch cabinet temperature prediction and early warning device, including:

[0072] The feature extraction module is used to extract features from the historical current time series data of the switch cabinet sensor nodes, the historical temperature time series data of the switch cabinet sensor nodes, and the historical ambient temperature time series data of the switch cabinet, respectively, to obtain the corresponding three groups of switch cabinet time series features.

[0073] The feature fusion module is used to splice three groups of switchgear time series features to obtain the global time features of the switchgear; perform convolution operations of different scales on the global time features of the switchgear to extract the local features of the switchgear at different time granularities to obtain the local time features of the switchgear at different scales; splice the local time features of the switchgear at different scales to obtain the multi-scale local time features of the switchgear; and weight the local time features of the switchgear whose contribution to temperature prediction is greater than the threshold contribution in the multi-scale local time features of the switchgear to obtain the weighted features of the switchgear.

[0074] The temperature prediction and warning module is used to input the weighted features of the switch cabinet into the trained long short-term memory recurrent neural network to obtain the temperature prediction results for the future time period; and input the weighted features of the switch cabinet into the trained fully connected neural network to obtain the temperature level warning results for the future time period.

[0075] For the specific definition of the switch cabinet temperature prediction and early warning device, please refer to the definition of the switch cabinet temperature prediction and early warning method above, which will not be repeated here. Each module in the above switch cabinet temperature prediction and early warning device can be implemented in whole or in part by software, hardware and a combination thereof. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0076] The present invention also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 Provided switch cabinet temperature prediction and early warning method.

[0077] The present invention also provides a computer device structure. At the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 Provided switch cabinet temperature prediction and early warning method.

[0078] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0079] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A switch cabinet temperature prediction and early warning method, characterized in that: include: Feature extraction is performed on the historical current time series data of the switch cabinet sensor node, the historical temperature time series data of the switch cabinet sensor node, and the historical ambient temperature time series data of the switch cabinet, respectively, to obtain the corresponding three groups of switch cabinet time series features; Splice three groups of switch cabinet time series features to obtain the global time features of the switch cabinet; Perform convolution operations of different scales on the global time features of the switch cabinet to extract local features of the switch cabinet at different time granularities, and obtain local time features of the switch cabinet at different scales; splice the local time features of the switch cabinet at different scales to obtain multi-scale local time features of the switch cabinet; and weight the local time features of the switch cabinet whose contribution to temperature prediction is greater than the threshold contribution in the multi-scale local time features of the switch cabinet to obtain weighted features of the switch cabinet; The weighted features of the switch cabinet are input into the trained long short-term memory recurrent neural network to obtain the temperature prediction results for the future time period; and the weighted features of the switch cabinet are input into the trained fully connected neural network to obtain the temperature level warning results for the future time period.

2. The switch cabinet temperature prediction and early warning method according to claim 1, characterized in that: The convolution operations of different scales are performed on the global time features of the switch cabinet, specifically including: Performing a convolution operation on the switch cabinet global time feature through a 1×1 convolution layer, and fusing the switch cabinet global time feature with the output of the 1×1 convolution layer through a residual connection to obtain a first-scale switch cabinet local time feature; Performing a convolution operation on the switch cabinet global time feature through a 1×3 convolution layer, and fusing the switch cabinet global time feature with the output of the 1×3 convolution layer through a residual connection to obtain a second-scale switch cabinet local time feature; The global time features of the switch cabinet are convolved through two sequentially connected 1×3 convolutional layers, and the global time features of the switch cabinet are fused with the output of the second 1×3 convolutional layer through a residual connection to obtain the local time features of the switch cabinet at the third scale.

3. The switch cabinet temperature prediction and early warning method according to claim 1, characterized in that: The weighting of the switch cabinet local time features that contribute significantly to temperature prediction among the multi-scale switch cabinet local time features is achieved by adjusting the weights of the multi-scale switch cabinet local time features using a single-layer fully connected network with a sigmoid function.

4. The switch cabinet temperature prediction and early warning method according to claim 1, characterized in that: The feature extraction of the historical current time series data of the switch cabinet sensor node, the historical temperature time series data of the switch cabinet sensor node and the historical ambient temperature time series data of the switch cabinet respectively includes: The switchgear historical current time series data, switchgear historical temperature time series data, and switchgear historical ambient temperature time series data are respectively input into three trained Transformer models to capture the long-distance correlation of data points in each time series data in the time dimension and obtain the corresponding three groups of switchgear time series features.

5. A switch cabinet temperature prediction and early warning device, characterized in that: include: A feature extraction module is used to extract features from the historical current time series data of the switch cabinet sensor node, the historical temperature time series data of the switch cabinet sensor node, and the historical ambient temperature time series data of the switch cabinet, respectively, to obtain three sets of corresponding switch cabinet time series features; The feature fusion module is used to combine the three sets of switch cabinet time series features to obtain the global time features of the switch cabinet; Perform convolution operations of different scales on the global time features of the switch cabinet to extract local features of the switch cabinet at different time granularities, and obtain local time features of the switch cabinet at different scales; splice the local time features of the switch cabinet at different scales to obtain multi-scale local time features of the switch cabinet; and weight the local time features of the switch cabinet whose contribution to temperature prediction is greater than the threshold contribution in the multi-scale local time features of the switch cabinet to obtain weighted features of the switch cabinet; The temperature prediction and warning module is used to input the weighted features of the switch cabinet into the trained long short-term memory recurrent neural network to obtain the temperature prediction results for the future time period; and input the weighted features of the switch cabinet into the trained fully connected neural network to obtain the temperature level warning results for the future time period.

6. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

7. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in any one of claims 1 to 4 is implemented.

Images