DC / DC equipment abnormity early warning method based on alberttiny model and terminal

Through the deep learning model, the feature sequence is constructed and fine-tuned to the input power, power efficiency and capacitance temperature of DC/DC devices, and an abnormal warning and determination model is generated, which solves the problem of high fault warning false alarm rate of traditional DC/DC devices, and realizes advance The effect of detecting failures and reducing economic losses.

CN119066564AActive Publication Date: 2024-12-03CONTEMPORARY NEBULA TECH ENERGY CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411018580.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-05-29
Publication Date
2024-12-03
Estimated Expiration
2044-05-29

AI Technical Summary

Technical Problem

The traditional DC/DC equipment fault warning method has a high false alarm rate, resulting in unnecessary downtime and increasing operating costs, making it difficult to accurately detect equipment failures in advance.

Method used

A deep learning model is used to construct a feature sequence by obtaining the input power, power efficiency and capacitance temperature of DC/DC devices, and train and fine-tune using the ALBERT_Tiny model to generate an abnormal warning and determination model to determine whether an early warning is issued.

Benefits of technology

Effectively detect potential equipment failures in advance, reduce economic losses and safety accidents, and improve the accuracy and reliability of early warnings.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066564A_ABST
    Figure CN119066564A_ABST
Patent Text Reader

Abstract

The invention discloses a DC / DC equipment abnormity early warning method and terminal, and the method comprises the steps: obtaining the input power, power efficiency and capacitance temperature of DC / DC equipment at a certain moment as an acquisition unit, integrating the acquisition units containing a first preset number, and generating a feature sequence as a sample; constructing an operation sequence data set by using a second preset number of samples, and pre-training a deep learning model by using an operation sequence data solution; constructing an operation sequence data set with an abnormal data label to perform fine adjustment on the deep learning model, and generating an abnormal early warning judgment model; and obtaining the sample composed of the input power, the power efficiency and the capacitance temperature of the current DC / DC equipment as the input of the abnormity early warning judgment model, and judging whether early warning needs to be sent out according to the output result of the abnormity early warning judgment model.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] This case is a divisional application based on the invention patent with application date of 2024-5-29, application number 202410675781.8, and name “A method and terminal for abnormal warning of DC / DC equipment” as the parent case. Technical Field

[0002] The present invention relates to the field of energy storage, and in particular to a method and a terminal for abnormal warning of DC / DC equipment. Background Art

[0003] With the increasing share of new energy vehicles and the popularization of supercharging and fast charging, the application of solar-storage charging inspection stations has gradually become popular. Among them, DC / DC (direct current converter components) as key equipment in solar-storage charging inspection stations need to be frequently maintained and replaced. Sudden damage will seriously affect the normal charging order of the station.

[0004] Traditional DC / DC fault warning alarms mainly monitor DC / DC voltage and current and set a voltage and current threshold. During daily operation, if the threshold is exceeded, an alarm is issued. This solution has a high probability of false alarms, which can easily cause unnecessary downtime and increase operating costs. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method and a terminal for abnormal early warning of a DC / DC device, so as to accurately realize early discovery of faults of the DC / DC device.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] A method for abnormal warning of DC / DC equipment, comprising the steps of:

[0008] S1, obtaining the input power, power efficiency and capacitor temperature of the DC / DC device at a certain moment as a collection unit, integrating the collection units including a first preset number and generating a feature sequence as a sample; the sample is a three-dimensional feature sequence composed of the input power, power efficiency and capacitor temperature in the first preset number of collection units;

[0009] S2. Constructing a running sequence data set using the second preset number of samples; training the albert_tiny model using the running sequence data set, and returning a deep learning model capable of characterizing sequence relationships and sequence states; using a fully connected layer in the deep learning model to replace the original input Embedding layer;

[0010] S3. Construct a running sequence data set with abnormal data annotations to fine-tune the deep learning model and generate an abnormal warning judgment model;

[0011] S4. Obtain the sample consisting of the input power, power efficiency and capacitor temperature of the current DC / DC device as the input of the abnormal warning determination model, and determine whether it is necessary to issue a warning according to the output result of the abnormal warning determination model.

[0012] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0013] A terminal for abnormal warning of DC / DC equipment includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are performed:

[0014] S1, obtaining the input power, power efficiency and capacitor temperature of the DC / DC device at a certain moment as a collection unit, integrating the collection units including a first preset number and generating a feature sequence as a sample; the sample is a three-dimensional feature sequence composed of the input power, power efficiency and capacitor temperature in the first preset number of collection units;

[0015] S2. Constructing a running sequence data set using the second preset number of samples; training the albert_tiny model using the running sequence data set, and returning a deep learning model capable of characterizing sequence relationships and sequence states; using a fully connected layer in the deep learning model to replace the original input Embedding layer;

[0016] S3. Construct a running sequence data set with abnormal data annotations to fine-tune the deep learning model and generate an abnormal warning judgment model;

[0017] S4. Obtain the sample consisting of the input power, power efficiency and capacitor temperature of the current DC / DC device as the input of the abnormal warning determination model, and determine whether it is necessary to issue a warning according to the output result of the abnormal warning determination model.

[0018] The beneficial effects of the present invention are: providing a method and terminal for abnormal warning of DC / DC equipment, obtaining the input power, power efficiency and capacitor temperature of the DC / DC equipment during operation as the basic acquisition unit for constructing samples, effectively capturing the operating status of the equipment, forming meaningful samples, and using these samples to pre-train and fine-tune the deep learning model to construct an efficient abnormal warning judgment model, thereby discovering potential equipment failures in advance and avoiding greater economic losses and safety accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of a method for abnormal warning of a DC / DC device in an embodiment of the present invention;

[0020] Figure 2 A specific flow chart of a method for abnormal warning of a DC / DC device in an embodiment of the present invention;

[0021] Figure 3 Schematic diagram of the architecture of an abnormality warning determination model in a method for abnormality warning of a DC / DC device in an embodiment of the present invention;

[0022] Figure 4 A schematic diagram of a terminal for abnormal warning of a DC / DC device in an embodiment of the present invention;

[0023] Figure 5 A topological diagram of the ALBERT_Tiny model in an embodiment of the present invention;

[0024] Description of labels:

[0025] 1. A terminal for abnormal warning of DC / DC equipment; 2. Memory; 3. Processor; DETAILED DESCRIPTION

[0026] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in conjunction with the implementation modes and the accompanying drawings.

[0027] The relevant terms appearing in the present invention are explained as follows:

[0028] DC / DC: A DC converter used to convert the voltage of an input DC power supply into a different output DC voltage.

[0029] Pre-trained model: A pre-trained model refers to a neural network model that has been pre-trained on large-scale data through unsupervised or semi-supervised learning. These models are usually trained using powerful computing resources and large amounts of text, image, or audio data to learn the statistical structure and feature representation of the data. The purpose of a pre-trained model is to extract universal feature representations by learning patterns and information from large-scale data so that it can be fine-tuned or transferred in subsequent tasks, thereby improving the performance of the model on specific tasks.

[0030] ALBERT_Tiny: A small version of the ALBERT (ALite BERT) model. ALBERT is an improved pre-trained model based on BERT (Bidirectional Encoder Representations from Transformers) developed by the Google Brain team.

[0031] Embedding: Embedding is an important concept in computer science and machine learning. It usually refers to the process of mapping high-dimensional data to low-dimensional space, where each data point (such as words, images, users, etc.) is represented as a vector, which usually contains fewer dimensions; it is the process of mapping discrete symbolic representations (such as words, characters, categories, etc.) to continuous real vector space. Embedding is usually used to convert high-dimensional, sparse discrete features into low-dimensional, dense continuous features, so as to better express the semantic relationship between features.

[0032] Fully connected layer: A fully connected layer, also called a densely connected layer or a fully connected layer, is a common layer type in deep learning neural networks. In a fully connected layer, each node (neuron) is connected to all nodes in the previous layer, so each input feature is multiplied by the weight of each neuron and output through an activation function.

[0033] Layernorm: Layer Normalization is a normalization technique used in neural networks to solve problems such as gradient vanishing and gradient exploding when training deep neural networks. In layer normalization, the features of each sample are normalized so that the mean of each feature in the sample dimension is 0 and the standard deviation is 1.

[0034] Dropout: Dropout is a regularization technique used in deep neural networks to prevent overfitting. During training, Dropout randomly sets some neurons in the network to 0, that is, "discards" them, and the activation values ​​of these neurons are set to 0 during forward propagation and backpropagation. This means that in each training iteration, the model is randomly pruned to reduce the complexity of the model to a certain extent.

[0035] Transformer: Transformer is a deep learning model for processing sequence data. The Transformer model consists of an encoder and a decoder, and is often used to process sequence-to-sequence (Seq2Seq) tasks.

[0036] Softmax: Softmax is a commonly used activation function, mainly used for the output of the last layer in multi-classification problems.

[0037] Please refer to Figure 1 , a method for abnormal warning of DC / DC equipment, comprising the steps of:

[0038] S1, obtaining the input power, power efficiency and capacitor temperature of the DC / DC device at a certain moment as a collection unit, integrating the collection units including a first preset number and generating a feature sequence as a sample; the sample is a three-dimensional feature sequence composed of the input power, power efficiency and capacitor temperature in the first preset number of collection units;

[0039] S2. Constructing a running sequence data set using the second preset number of samples; training the albert_tiny model using the running sequence data set, and returning a deep learning model capable of characterizing sequence relationships and sequence states; using a fully connected layer in the deep learning model to replace the original input Embedding layer;

[0040] S3. Construct a running sequence data set with abnormal data annotations to fine-tune the deep learning model and generate an abnormal warning judgment model;

[0041] S4. Obtain the sample consisting of the input power, power efficiency and capacitor temperature of the current DC / DC device as the input of the abnormal warning determination model, and determine whether it is necessary to issue a warning according to the output result of the abnormal warning determination model.

[0042] It can be understood that the operating series data set with abnormal data annotations is a data set formed by the corresponding input power, power efficiency and capacitor temperature when an abnormal situation occurs in the application background of the DC / DC device.

[0043] From the above description, it can be seen that the beneficial effect of the present invention is that the input power, power efficiency and capacitor temperature of the DC / DC device during operation are obtained as the basic acquisition unit for constructing samples, effectively capturing the operating status of the device and forming meaningful samples. These samples are used to pre-train and fine-tune the deep learning model to build an efficient abnormal warning judgment model, thereby discovering potential equipment failures in advance and avoiding greater economic losses and safety accidents.

[0044] Please refer to Figures 1 to 3 In an embodiment of the present invention, the deep learning model is an ALBERT_Tiny model, and a fully connected layer is used to replace the original input Embedding layer;

[0045] The step S2 is specifically as follows:

[0046] constructing a running sequence data set using a second preset number of the samples;

[0047] The ALBERT_Tiny model is trained using the running sequence dataset, and a deep learning model capable of characterizing sequence relationships and sequence states is returned.

[0048] From the above description, we can see that using the ALBERT_Tiny model and replacing the original Embedding layer with a fully connected layer can reduce the complexity of the model and improve computational efficiency while maintaining the model's high sensitivity to sequence relationships and states. This improvement helps improve the model's performance when processing time series data, especially in identifying abnormal device behavior.

[0049] In an embodiment of the present invention, step S3 is specifically:

[0050] Freezing some parameters of the deep learning model;

[0051] Construct a run sequence dataset with annotated abnormal data;

[0052] The deep learning model is fine-tuned using the running sequence data set with abnormal data annotations to generate an abnormal warning determination model.

[0053] From the above description, we can see that by freezing some parameters of the deep learning model and fine-tuning it using data with anomaly annotations, the model can be more focused on learning the features that distinguish normal and abnormal states. This method improves the generalization ability and accuracy of the model in practical applications and ensures the reliability of anomaly warning.

[0054] In an embodiment of the present invention, the sample is a three-dimensional feature sequence consisting of input power, power efficiency and capacitor temperature in a first preset number of acquisition units.

[0055] From the above description, it can be seen that the three-dimensional feature sequence of the sample consisting of input power, power efficiency and capacitor temperature enables the model to evaluate the operating status of the equipment from a comprehensive perspective, thereby enhancing the comprehensive judgment ability of the early warning system.

[0056] Specifically, the sample structure is as follows:

[0057] [[P1,P2,…,P 64 ],[E1,E2,…,E 64 ],[T1,T2,…,T 64 ]]

[0058] Among them, the input power is recorded as P n , power efficiency is recorded as E n , the capacitance temperature is recorded as T n , 64 points are taken in the order of sampling time to form a 64*3-dimensional feature sequence.

[0059] In an embodiment of the present invention, the first preset number is equal to the second preset number and has a value range of 50-100.

[0060] From the above description, it can be seen that based on the application background of the optical storage and charging inspection station, it is concluded from experience that when the first preset number and the second preset number are equal and the value range is 50-100, the model training effect is best, preferably 60 and 64.

[0061] Please refer to Figure 4 A terminal for abnormal warning of DC / DC equipment includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned method for abnormal warning of DC / DC equipment are completed.

[0062] From the above description, it can be seen that an execution carrier of a method for abnormal warning of DC / DC equipment is provided. When executing the above method, the input power, power efficiency and capacitor temperature of the DC / DC equipment during operation are obtained as the basic acquisition unit for constructing samples, effectively capturing the operating status of the equipment and forming meaningful samples. These samples are used to pre-train and fine-tune the deep learning model to build an efficient abnormal warning judgment model, so as to discover potential equipment failures in advance and avoid greater economic losses and safety accidents.

[0063] The present invention provides a method and terminal for warning abnormality of a DC / DC device, which is mainly used for accurately warning abnormality of a DC / DC device in a photovoltaic storage and charging inspection station. The following is a specific description in conjunction with an embodiment:

[0064] Please refer to Figure 1 , Embodiment 1 of the present invention is:

[0065] A method for abnormal warning of DC / DC equipment, comprising the steps of:

[0066] S1, obtaining the input power, power efficiency and capacitor temperature of the DC / DC device at a certain moment as a collection unit, integrating a first preset number of collection units and generating a feature sequence as a sample;

[0067] S2. constructing a running sequence data set using a second preset number of samples and pre-training a deep learning model using the running sequence data set;

[0068] S3. Construct a running sequence data set with abnormal data annotations to fine-tune the deep learning model and generate an abnormal warning judgment model;

[0069] S4. Obtain samples of the input power, power efficiency and capacitor temperature of the current DC / DC device as inputs to the abnormal warning determination model, and determine whether a warning needs to be issued based on the output results of the abnormal warning determination model.

[0070] That is, in this embodiment, the input power, power efficiency and capacitor temperature of the DC / DC device during operation are obtained as the basic acquisition unit for constructing samples, effectively capturing the operating status of the device and forming meaningful samples. These samples are used to pre-train and fine-tune the deep learning model to build an efficient abnormal warning judgment model, thereby discovering potential equipment failures in advance and avoiding greater economic losses and safety accidents.

[0071] Specifically, before step S1, step S0 is also included:

[0072] Obtain the input power and output power of the DC / DC device at a certain moment, and calculate the power efficiency.

[0073] Specifically, the fine-tuning principle and process in step S3 are as follows:

[0074] Use a pre-trained-fine-tuned model where:

[0075] Pre-training: In this phase, a large-scale unlabeled dataset (usually text data) is used to train a model. This model is usually a deep neural network, such as a Transformer model. During the training process, the model captures the abstract representation of language by learning the language patterns, grammatical structures, semantic information, etc. of the text data. The purpose of pre-training is to provide the model with a general understanding of the language world.

[0076] Fine-tuning: After pre-training, the pre-trained model is used for a specific task (such as text classification, named entity recognition, sentiment analysis, etc.). In the fine-tuning stage, the model is additionally trained on a small amount of labeled data to adjust its parameters to make it more suitable for a specific task. The fine-tuning process usually fixes some or all of the parameters of the pre-trained model, and then updates the remaining parameters on the new labeled data.

[0077] The advantage of the pre-training-fine-tuning model training method is that it can use large-scale unlabeled data for pre-training, so that the model has better generalization ability, and only relatively small amounts of labeled data are needed for fine-tuning on specific tasks, thus saving the cost of data annotation. In addition, since the pre-trained model has learned the general laws of language, it can usually achieve better performance on specific tasks.

[0078] Please refer to Figures 1 to 3 , Embodiment 2 of the present invention is:

[0079] Based on Example 1, the deep learning model is an ALBERT_Tiny model, and a fully connected layer is used to replace the original input Embedding layer; using the ALBERT_Tiny model and replacing the original Embedding layer with a fully connected layer can reduce the complexity of the model and improve computational efficiency, while maintaining the model's high sensitivity to sequence relationships and states.

[0080] For details, please refer to Figure 5 , the ALBERT_Tiny model includes Timedistributed(dense), which means that a dense operation is performed on each of the 64 time steps in the sequence. Dense, namely the dense connection layer (DenseLayer), is also called the fully connected layer (Fully Connected Layer) or affine layer (Affine Layer), and the specific expression is:

[0081] Y = f(W*X+b);

[0082] Among them, X represents the input data, the shape is (n, m), n is the number of samples, and m is the input feature dimension;

[0083] W represents the connection weight matrix, with a shape of (m, k), W ij Represents the connection weight between the i-th input feature and the j-th neuron;

[0084] b is the bias vector, shape is (1, k);

[0085] f is the activation function applied on XW, such as ReLU, Sigmoid, etc.;

[0086] Y represents output;

[0087] Specifically, the ALBERT_Tiny model runs as follows:

[0088] (1) Input data:

[0089] Input power sequence (1*64)

[0090] Power efficiency sequence (1*64)

[0091] Capacitor temperature sequence (1*64)

[0092] These input data represent the input power, power efficiency and capacitor temperature of the DCDC device, respectively, and are input into the model in the form of time series, with each series containing 64 time-step values.

[0093] (2) Merge sequence:

[0094] The three input sequences are merged into an overall sequence (3*64), that is, the data of each time step are spliced ​​together to form a time series containing 3 features.

[0095] (3)Position Embedding:

[0096] Add position embedding information (128*64) to each time step of the merged sequence. Through position embedding, the model can identify the position of each data in the time series, so as to better understand the sequential information of the time series.

[0097] (4) TimeDistributed(Dense)Layer (time-distributed fully connected layer):

[0098] Perform full connection (128*64) on each time step of the merged sequence, that is, perform full connection operation on the data of each time step, extract features and increase feature dimensions.

[0099] (5) Plus (addition of positional embedded elements):

[0100] The time series output by the above fully connected layer is added to the position embedding vector to further integrate the position information of the time step.

[0101] (6) LayerNorm and Dropout:

[0102] The data is standardized through layer normalization, and then the Dropout operation is performed to randomly discard some neurons to prevent overfitting and improve the generalization ability of the model.

[0103] (7) Transformer (transformer module):

[0104] A 12-layer transformer module is used, the hidden size of each layer is 768, and num_hidden_layers is 12. The transformer module performs feature extraction and sequence modeling on time series through a multi-head attention mechanism and a feedforward neural network.

[0105] Step S2 is specifically as follows:

[0106] constructing a run sequence data set using a second preset number of samples;

[0107] The ALBERT_Tiny model is trained using the running sequence dataset and a deep learning model that can characterize sequence relationships and sequence states is returned. The above steps help improve the performance of the model when processing time series data, especially in identifying abnormal device behavior.

[0108] Specifically, refer to Figure 3 , step S3 is specifically as follows:

[0109] Freeze some parameters of the deep learning model; by freezing some parameters of the deep learning model and fine-tuning it using data with abnormal annotations, the model can be more focused on learning features that distinguish normal from abnormal states.

[0110] Construct a run sequence dataset with annotated abnormal data;

[0111] The deep learning model is fine-tuned using the running sequence data set with anomaly data annotations to generate an anomaly warning judgment model. The above steps improve the generalization ability and accuracy of the model in practical applications and ensure the reliability of anomaly warning.

[0112] Specifically, in this example, the early warning of DC / DC equipment is divided into two parts: online and offline. The online part uses the trained abnormal early warning judgment model to provide real-time early warning for DC / DC equipment. The offline part generates an abnormal early warning judgment model based on historical data according to the above method. The abnormal early warning judgment model is corrected by continuously replacing samples according to the daily data of DC / DC equipment to improve the accuracy of model judgment.

[0113] Embodiment 3 of the present invention is:

[0114] Based on the second embodiment, the sample structure is as follows:

[0115] [[P1,P2,…,P 64 ],[E1,E2,…,E 64 ],[T1,T2,…,T 64 ]]

[0116] Among them, the input power is recorded as P n , power efficiency is recorded as E n , the capacitance temperature is recorded as T n , 64 points are taken in the order of sampling time to form a 64*3-dimensional feature sequence.

[0117] Please refer to Figure 4 , Embodiment 4 of the present invention is:

[0118] A terminal 1 for abnormal warning of a DC / DC device comprises a memory 2, a processor 3 and a computer program stored in the memory 2 and executable on the processor 3. When the processor 3 executes the computer program, the steps of any one of the methods for abnormal warning of a DC / DC device in the first to third embodiments are completed.

[0119] In summary, the present invention provides a method and terminal for abnormal warning of DC / DC equipment, which obtains the input power, power efficiency and capacitor temperature of the DC / DC equipment during operation as the basic acquisition units for constructing samples, effectively captures the operating status of the equipment, forms meaningful samples, and uses these samples to pre-train and fine-tune the deep learning model to construct an efficient abnormal warning judgment model, thereby discovering potential equipment failures in advance and avoiding greater economic losses and safety accidents.

[0120] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for abnormal warning of DC / DC equipment based on the albert_tiny model, characterized in that: Includes steps: S1, obtaining the input power, power efficiency and capacitor temperature of the DC / DC device at a certain moment as a collection unit, integrating the collection units including a first preset number and generating a feature sequence as a sample; the sample is a three-dimensional feature sequence composed of the input power, power efficiency and capacitor temperature in the first preset number of collection units; S2. Constructing a running sequence data set using the second preset number of samples; training the albert_tiny model using the running sequence data set, and returning a deep learning model capable of characterizing sequence relationships and sequence states; using a fully connected layer in the deep learning model to replace the original input Embedding layer; S3. Construct a running sequence data set with abnormal data annotations to fine-tune the deep learning model and generate an abnormal warning judgment model; S4, obtaining the sample consisting of the input power, power efficiency and capacitor temperature of the current DC / DC device as the input of the abnormal warning determination model, and determining whether it is necessary to issue an early warning according to the output result of the abnormal warning determination model; The albert_tiny model operation includes the following steps: Input power series, power efficiency series and capacitor temperature series of the input DCDC device; Merge the three input sequences into one overall sequence; Add position embedding information to each time series of the merged sequence; Perform a full connection on each time series of the merged sequence; Add the time series output by the fully connected layer to the position embedding vector; The data is standardized through layer normalization, followed by dropout operation; Utilize Transformer units for complex feature extraction.

2. The method for abnormal warning of DC / DC equipment based on the albert_tiny model according to claim 1, characterized in that: The albert_tiny model specifically includes: Input unit: used to receive a power sequence, a power efficiency sequence and a capacitor temperature sequence, each sequence having a dimension of 1×n; n is a positive integer; A data merging unit, used for merging the power sequence, the power efficiency sequence and the capacitor temperature sequence into a 3×n matrix, wherein n is a positive integer; A position embedding unit, used to add a position label to each data point in the merged data sequence; The time series distribution unit includes a fully connected layer, which is used to transform the features of each time point of the merged sequence and map the data of each time point into a 128-dimensional space; An adding unit, used to add the data after feature transformation and the position embedding information to fuse the position information; Normalization and dropout units are used to normalize the data and randomly remove some data to prevent overfitting; Transformer unit, which consists of 12 layers, each containing 768 hidden units, used to capture the relationship between distant data points in the sequence through the self-attention mechanism and perform complex feature extraction; The step S3 is specifically as follows: Freezing some parameters of the deep learning model; Construct a run sequence dataset with annotated abnormal data; The deep learning model is fine-tuned using the running sequence data set with abnormal data annotations to generate an abnormal warning determination model.

3. The method for abnormal warning of DC / DC equipment based on the albert_tiny model according to claim 1, characterized in that: The structure of the sample is: [[P1,P2,…,P n ],[E1,E2,…,E n ],[T1,T2,…,T n ]]; Among them, the input power is recorded as P n , power efficiency is recorded as E n , the capacitance temperature is recorded as T n , wherein n is a positive integer.

4. The method for abnormal warning of DC / DC equipment based on the albert_tiny model according to claim 1, characterized in that: The time series distribution unit includes a time-distributed fully connected layer for performing independent feature transformation on the data at each time point of the merged sequence; The Transformer unit performs feature extraction through a self-attention mechanism to capture the relationship between distant data points in the sequence; The normalization and dropout unit is used to standardize the data after feature transformation and position fusion, and randomly remove a part of the data.

5. A terminal for abnormal warning of DC / DC equipment based on the albert_tiny model, characterized in that: The invention comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the following steps are performed: S1, obtaining the input power, power efficiency and capacitor temperature of the DC / DC device at a certain moment as a collection unit, integrating the collection units including a first preset number and generating a feature sequence as a sample; the sample is a three-dimensional feature sequence composed of the input power, power efficiency and capacitor temperature in the first preset number of collection units; S2. Constructing a running sequence data set using the second preset number of samples; training the albert_tiny model using the running sequence data set, and returning a deep learning model capable of characterizing sequence relationships and sequence states; using a fully connected layer in the deep learning model to replace the original input Embedding layer; S3. Construct a running sequence data set with abnormal data annotations to fine-tune the deep learning model and generate an abnormal warning judgment model; S4, obtaining the sample consisting of the input power, power efficiency and capacitor temperature of the current DC / DC device as the input of the abnormal warning determination model, and determining whether it is necessary to issue an early warning according to the output result of the abnormal warning determination model; The albert_tiny model operation includes the following steps: Input power series, power efficiency series and capacitor temperature series of the input DCDC device; Merge the three input sequences into one overall sequence; Add position embedding information to each time series of the merged sequence; Perform a full connection on each time series of the merged sequence; Add the time series output by the fully connected layer to the position embedding vector; The data is standardized through layer normalization, followed by dropout operation; Utilize Transformer units for complex feature extraction.

6. The terminal for abnormal warning of DC / DC equipment based on the albert_tiny model according to claim 1, characterized in that: The albert_tiny model specifically includes: Input unit: used to receive power sequence, power efficiency sequence and capacitance temperature sequence, each sequence has a dimension of 1×n, where n is a positive integer A data merging unit, used to merge the power sequence, power efficiency sequence and capacitor temperature sequence into a 3×n matrix, where n is a positive integer A position embedding unit, used to add a position label to each data point in the merged data sequence; The time series distribution unit includes a fully connected layer, which is used to transform the features of each time point of the merged sequence and map the data of each time point into a 128-dimensional space; An adding unit, used to add the data after feature transformation and the position embedding information to fuse the position information; Normalization and dropout units are used to normalize the data and randomly remove some data to prevent overfitting; Transformer unit, which consists of 12 layers, each containing 768 hidden units, used to capture the relationship between distant data points in the sequence through the self-attention mechanism and perform complex feature extraction; The step S3 is specifically as follows: Freezing some parameters of the deep learning model; Construct a run sequence dataset with annotated abnormal data; The deep learning model is fine-tuned using the running sequence data set with abnormal data annotations to generate an abnormal warning determination model.

7. The terminal for abnormal warning of DC / DC equipment based on the albert_tiny model according to claim 5, characterized in that: The structure of the sample is: [[P1,P2,…,P n ],[E1,E2,…,E n ],[T1,T2,…,T n ]]; Among them, the input power is recorded as P n , power efficiency is recorded as E n , the capacitance temperature is recorded as T n , wherein n is a positive integer.

8. The terminal for abnormal warning of DC / DC equipment based on the albert_tiny model according to claim 5, characterized in that: The time series distribution unit includes a time-distributed fully connected layer for performing independent feature transformation on the data at each time point of the merged sequence; The Transformer unit performs feature extraction through a self-attention mechanism to capture the relationship between distant data points in the sequence; The normalization and dropout unit is used to standardize the data after feature transformation and position fusion, and randomly remove a part of the data.

Citation Information

Patent Citations

  • Health state monitoring method of DC-DC power supply based on machine learning

    CN112560892A

  • Construction method and application of hybrid capacitor power state online estimation model

    CN114186522A

  • Early warning method and system for operation state of power grid

    CN116881704A

  • KR20230036776A