CN117370847B

CN117370847B - Deep learning-based disconnecting switch detection method and device

Info

Publication number: CN117370847B
Application number: CN202311674593.5A
Authority: CN
Inventors: 刘路江; 翟伟强; 杜肖悦
Original assignee: Shenzhen Yuyi Technology Co ltd
Current assignee: Shenzhen Yuyi Technology Co ltd
Priority date: 2023-12-08
Filing date: 2023-12-08
Publication date: 2024-02-13
Anticipated expiration: 2043-12-08
Also published as: CN117370847A

Abstract

The invention relates to the field of deep learning, and discloses a method and a device for detecting an isolating switch based on deep learning, which are used for improving the accuracy of detecting the running state of the isolating switch. The method comprises the following steps: acquiring multidimensional data to obtain initial multidimensional sensing signal data; performing modal decomposition and feature classification to obtain noise aliasing component data; performing signal filtering processing to obtain standard multidimensional sensing signal data; performing sensing data conversion and classification to obtain target current data, target pressure data and target temperature data; performing distribution mapping and feature extraction to obtain a current waveform feature set, a pressure change feature set and a temperature rise and fall feature set; performing feature coding and vector fusion to obtain a target fusion feature vector; and inputting the target fusion feature vector into a disconnector life prediction model to predict the disconnector life, obtaining disconnector life prediction data, and creating a target running state parameter control scheme.

Description

Deep learning-based disconnecting switch detection method and device

Technical Field

The invention relates to the field of deep learning, in particular to a method and a device for detecting an isolating switch based on deep learning.

Background

An isolating switch is an important component in a power system, and accurate detection of the running state of the isolating switch is important to the stability and safety of a power grid.

Traditional isolating switch state detection methods often rely on direct measurement of sensors, are easily affected by noise interference and environmental changes, and further lead to lower accuracy of the existing schemes.

Disclosure of Invention

The invention provides a method and a device for detecting an isolating switch based on deep learning, which are used for improving the accuracy of detecting the running state of the isolating switch.

The first aspect of the invention provides a method for detecting a disconnecting switch based on deep learning, which comprises the following steps: the operation state of the target isolating switch is subjected to multidimensional data acquisition through a preset sensor group, so that initial multidimensional sensing signal data are obtained; performing modal decomposition on the initial multidimensional sensing signal data to obtain a plurality of intrinsic modal component data, and respectively performing feature extraction and feature classification on the plurality of intrinsic modal component data through a preset multiscale entropy algorithm to obtain noise aliasing component data; performing signal filtering processing on the initial multi-dimensional sensing signal data according to the noise aliasing component data through a preset back propagation neural network and a Kalman filter to obtain standard multi-dimensional sensing signal data; performing sensing data conversion and classification on the standard multidimensional sensing signal data to obtain target current data, target pressure data and target temperature data; respectively carrying out distribution mapping and feature extraction on the target current data, the target pressure data and the target temperature data to obtain a current waveform feature set, a pressure change feature set and a temperature rise and fall feature set; performing feature coding and vector fusion on the current waveform feature set, the pressure change feature set and the temperature rise and fall feature set to obtain a target fusion feature vector; and inputting the target fusion feature vector into a preset isolating switch life prediction model to predict the life of the isolating switch, obtaining isolating switch life prediction data, and creating a target running state parameter control scheme of the target isolating switch according to the isolating switch life prediction data.

With reference to the first aspect, in a first implementation manner of the first aspect of the present invention, performing modal decomposition on the initial multidimensional sensing signal data to obtain a plurality of intrinsic mode component data, and performing feature extraction and feature classification on the plurality of intrinsic mode component data through a preset multiscale entropy algorithm to obtain noise aliasing component data, where the method includes: inputting the initial multidimensional sensing signal into a preset set empirical mode decomposition analysis algorithm, and performing parameter setting on the set empirical mode decomposition analysis algorithm to obtain a plurality of target decomposition parameters; performing modal decomposition on the initial multidimensional sensing signal data according to the target decomposition parameters to obtain a plurality of intrinsic modal component data; respectively extracting the characteristics of the plurality of intrinsic mode component data through a preset multi-scale entropy algorithm to obtain multi-scale entropy characteristics of each intrinsic mode component data; and carrying out feature classification on the multi-scale entropy value features of each eigenvector component data through a preset support vector machine model to obtain noise aliasing component data.

With reference to the first aspect, in a second implementation manner of the first aspect of the present invention, the performing signal filtering processing on the initial multi-dimensional sensing signal data according to the noise aliasing component data through a preset back propagation neural network and a kalman filter to obtain standard multi-dimensional sensing signal data includes: taking the noise aliasing component data as input data of a preset back propagation neural network, and simultaneously taking the initial multidimensional sensing signal data as target output of the back propagation neural network; training the back propagation neural network, and adjusting the weight and bias parameters of the back propagation neural network to obtain a trained back propagation neural network; predicting the noise aliasing component data through the trained back propagation neural network to obtain new noise aliasing component data; and inputting the new noise aliasing component data into a preset Kalman filter, and performing signal filtering processing on the initial multi-dimensional sensing signal data to obtain standard multi-dimensional sensing signal data.

With reference to the first aspect, in a third implementation manner of the first aspect of the present invention, the performing sensing data conversion and classification on the standard multidimensional sensing signal data to obtain target current data, target pressure data and target temperature data includes: performing sensing data conversion on the standard multidimensional sensing signal data to obtain target multidimensional attribute data; inputting the target multidimensional attribute data into a preset cluster analysis model, and performing cluster center calculation on the target multidimensional attribute data through the cluster analysis model to obtain a corresponding initial current cluster center, an initial voltage cluster center and an initial temperature cluster center; carrying out data distance calculation on the target multidimensional attribute data, the initial current clustering center, the initial voltage clustering center and the initial temperature clustering center to obtain a data distance calculation result; according to the data distance calculation result, carrying out cluster center correction on the initial current cluster center, the initial voltage cluster center and the initial temperature cluster center to obtain a target current cluster center, a target voltage cluster center and a target temperature cluster center; and performing secondary clustering on the target multidimensional attribute data according to the target current clustering center, the target voltage clustering center and the target temperature clustering center to obtain target current data, target pressure data and target temperature data.

With reference to the first aspect, in a fourth implementation manner of the first aspect of the present invention, the performing distribution mapping and feature extraction on the target current data, the target pressure data, and the target temperature data to obtain a current waveform feature set, a pressure change feature set, and a temperature rise feature set respectively includes: performing current spectrogram mapping on the target current data to obtain a current distribution spectrogram, performing pressure curve conversion on the target pressure data to obtain a pressure distribution curve, and performing temperature distribution mapping on the target temperature data to obtain a temperature distribution box diagram; carrying out current waveform characteristic identification on the current distribution spectrogram to obtain a plurality of initial current waveform characteristics, and carrying out characteristic screening and integrated conversion on the plurality of initial current waveform characteristics to obtain a current waveform characteristic set; detecting the pressure change characteristics of the pressure distribution curve to obtain a plurality of initial pressure change characteristics, and carrying out characteristic screening and integrated conversion on the plurality of initial pressure change characteristics to obtain a pressure change characteristic set; and extracting temperature lifting features of the temperature distribution box diagram to obtain a plurality of initial temperature lifting features, and carrying out feature screening and integrated conversion on the plurality of initial temperature lifting features to obtain a temperature lifting feature set.

With reference to the first aspect, in a fifth implementation manner of the first aspect of the present invention, performing feature encoding and vector fusion on the current waveform feature set, the pressure change feature set, and the temperature rise feature set to obtain a target fusion feature vector includes: carrying out characteristic standardization processing on the current waveform characteristic set, the pressure change characteristic set and the temperature rise and fall characteristic set in a distribution manner to obtain a plurality of standard current waveform characteristics, a plurality of standard pressure change characteristics and a plurality of standard temperature rise and fall characteristics; performing characteristic normalization processing on the plurality of standard current waveform characteristics, the plurality of standard pressure change characteristics and the plurality of standard temperature rise and fall characteristics respectively to obtain a plurality of normalized current waveform characteristics, a plurality of normalized pressure change characteristics and a plurality of normalized temperature rise and fall characteristics; performing feature coding on the plurality of normalized current waveform features, the plurality of normalized pressure change features and the plurality of normalized temperature rise and fall features respectively to obtain a plurality of current waveform feature codes, a plurality of pressure change feature codes and a plurality of temperature rise and fall feature codes; vector conversion is carried out on the plurality of current waveform feature codes, the plurality of pressure change feature codes and the plurality of temperature rise and fall feature codes respectively to obtain a current waveform feature vector, a pressure change feature vector and a temperature rise and fall feature vector; performing weight analysis on the target current data, the target pressure data and the target temperature data to obtain target weight data; and carrying out vector fusion on the current waveform feature vector, the pressure change feature vector and the temperature rise and fall feature vector according to the target weight data to obtain a target fusion feature vector.

With reference to the first aspect, in a sixth implementation manner of the first aspect of the present invention, inputting the target fusion feature vector into a preset life prediction model of the isolating switch to predict the life of the isolating switch, to obtain life prediction data of the isolating switch, and creating a target running state parameter control scheme of the target isolating switch according to the life prediction data of the isolating switch, where the method includes: inputting the target fusion feature vector into a preset isolating switch life prediction model, wherein the isolating switch life prediction model comprises a decision tree model, a GRU model and an XGBoost model; performing isolating switch life prediction on the target fusion feature vector through the decision tree model to obtain first life prediction data, performing isolating switch life prediction on the target fusion feature vector through the GRU model to obtain second life prediction data, and performing isolating switch life prediction on the target fusion feature vector through the XGBoost model to obtain third life prediction data; according to a preset model weight proportion, carrying out weighted average processing on the first life prediction data, the second life prediction data and the third life prediction data to obtain isolating switch life prediction data; according to the life prediction data of the isolating switch, carrying out state parameter control scheme analysis on the target isolating switch to obtain an initial running state parameter control scheme; and carrying out parameter control optimization on the initial running state parameter control scheme through a preset particle swarm algorithm to generate a target running state parameter control scheme.

The second aspect of the present invention provides a deep learning-based disconnecting switch detection device, which includes: the acquisition module is used for acquiring multidimensional data of the running state of the target isolating switch through a preset sensor group to obtain initial multidimensional sensing signal data; the decomposition module is used for carrying out modal decomposition on the initial multidimensional sensing signal data to obtain a plurality of intrinsic modal component data, and carrying out feature extraction and feature classification on the plurality of intrinsic modal component data through a preset multiscale entropy algorithm to obtain noise aliasing component data; the processing module is used for carrying out signal filtering processing on the initial multi-dimensional sensing signal data according to the noise aliasing component data through a preset counter-propagation neural network and a Kalman filter to obtain standard multi-dimensional sensing signal data; the classification module is used for carrying out sensing data conversion and classification on the standard multidimensional sensing signal data to obtain target current data, target pressure data and target temperature data; the extraction module is used for carrying out distribution mapping and feature extraction on the target current data, the target pressure data and the target temperature data respectively to obtain a current waveform feature set, a pressure change feature set and a temperature rise feature set; the fusion module is used for carrying out feature coding and vector fusion on the current waveform feature set, the pressure change feature set and the temperature rise and fall feature set to obtain a target fusion feature vector; the creation module is used for inputting the target fusion feature vector into a preset isolating switch life prediction model to predict the life of the isolating switch, obtaining isolating switch life prediction data, and creating a target running state parameter control scheme of the target isolating switch according to the isolating switch life prediction data.

A third aspect of the present invention provides a deep learning-based disconnecting switch detection apparatus, including: a memory and at least one processor, the memory having instructions stored therein; the at least one processor invokes the instructions in the memory to cause the deep learning based isolation switch detection device to perform the deep learning based isolation switch detection method described above.

A fourth aspect of the present invention provides a computer readable storage medium having instructions stored therein that, when run on a computer, cause the computer to perform the deep learning based method of isolating switch detection described above.

In the technical scheme provided by the invention, multi-dimensional data acquisition is carried out to obtain initial multi-dimensional sensing signal data; performing modal decomposition and feature classification to obtain noise aliasing component data; performing signal filtering processing to obtain standard multidimensional sensing signal data; performing sensing data conversion and classification to obtain target current data, target pressure data and target temperature data; performing distribution mapping and feature extraction to obtain a current waveform feature set, a pressure change feature set and a temperature rise and fall feature set; performing feature coding and vector fusion to obtain a target fusion feature vector; the method comprises the steps of inputting the target fusion feature vector into a life prediction model of the isolating switch to predict the life of the isolating switch, obtaining life prediction data of the isolating switch, and creating a target running state parameter control scheme. By using a combined empirical mode decomposition analysis (CEEMDAN) technology, a plurality of intrinsic mode components can be effectively extracted, complex state changes of the isolating switch can be better reflected, and accuracy and reliability of signal decomposition are enhanced. The characteristic extraction and classification are carried out on the intrinsic mode components by utilizing a multi-scale entropy algorithm, so that the accurate distinguishing capability of noise aliasing components is improved, and the noise components can be accurately identified and filtered. And by combining the counter propagation neural network and the Kalman filter, the noise aliasing component is learned and predicted through the deep learning network, and then the signal filtering processing is performed through the Kalman filter, so that the accurate filtering and recovering capacity of the initial multidimensional sensing signal is improved. The standard multidimensional sensing signal data is subjected to sensing data conversion and classification, so that key information such as target current, target pressure and target temperature can be better extracted, and more targeted data can be provided for subsequent analysis. Through multi-level processing such as distribution mapping, feature extraction, coding and vector fusion, the characteristics of multiple aspects such as current waveform, pressure change and temperature rise and fall are successfully extracted and fused, and the state of the isolating switch is more comprehensively represented. The target fusion feature vector is input into a preset isolating switch life prediction model, the multi-model comprehensive effects of decision trees, GRU, XGBoost and the like are fully utilized, and the accuracy and the robustness of life prediction are improved. The life prediction result is optimized through a preset particle swarm algorithm, so that the robustness and the reliability of a control scheme are improved, the operation state parameters of the target isolating switch are controlled more accurately, and the accuracy of the operation state detection of the isolating switch is improved.

Drawings

FIG. 1 is a schematic diagram of an embodiment of a deep learning based method for detecting an isolating switch according to an embodiment of the present invention;

FIG. 2 is a flowchart of a signal filtering process according to an embodiment of the present invention;

FIG. 3 is a flow chart of the conversion and classification of sensor data in an embodiment of the invention;

FIG. 4 is a flow chart of distribution mapping and feature extraction in an embodiment of the invention;

FIG. 5 is a schematic diagram of an embodiment of a deep learning based isolation switch detection device according to an embodiment of the present invention;

fig. 6 is a schematic diagram of an embodiment of a deep learning-based disconnecting switch detection device in an embodiment of the invention.

Detailed Description

The embodiment of the invention provides a method and a device for detecting an isolating switch based on deep learning, which are used for improving the accuracy of detecting the running state of the isolating switch. The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.

For easy understanding, the following describes a specific flow of an embodiment of the present invention, referring to fig. 1, and one embodiment of a method for detecting an isolating switch based on deep learning in the embodiment of the present invention includes:

s101, performing multidimensional data acquisition on the running state of a target isolating switch through a preset sensor group to obtain initial multidimensional sensing signal data;

it can be appreciated that the execution subject of the present invention may be a deep learning based disconnecting switch detection device, and may also be a terminal or a system, which is not limited herein. The embodiment of the invention is described by taking a server as an execution main body as an example.

Specifically, a suitable set of sensors is selected. The sensor is capable of monitoring a multi-dimensional signal, such as current, pressure, temperature, etc., associated with the operating state of the disconnector. Different sensors require different operating principles and techniques and therefore the sensor type needs to be selected according to specific requirements. For example, a current sensor may be used to monitor current data, a pressure sensor to monitor pressure data, a temperature sensor to monitor temperature data, etc. The position where the sensor is installed should be able to accurately reflect the state of the disconnector. For example, in terms of current monitoring, sensors may be mounted on the cable connection points or the switchgear to ensure that current changes are accurately captured. Pressure sensors may be installed on the gas or liquid lines to monitor pressure changes. Temperature sensors may be mounted at critical locations to measure temperature data in real time. The sensor will continuously collect multidimensional signal data including current, pressure, temperature, etc. These data will be transmitted to the data acquisition system in the form of digital or analog signals. The data acquisition system stores the data in a database or cloud platform for later processing and analysis. In the data acquisition process, accuracy and continuity of data need to be ensured. Calibration and maintenance of the sensor is necessary to ensure proper operation. In addition, the data acquisition system should have fault detection and alarm functions, as well as data redundancy mechanisms to avoid data loss.

S102, performing modal decomposition on initial multidimensional sensing signal data to obtain a plurality of intrinsic modal component data, and respectively performing feature extraction and feature classification on the plurality of intrinsic modal component data through a preset multiscale entropy algorithm to obtain noise aliasing component data;

specifically, the system performs modal decomposition. The multi-dimensional sensing signal data is decomposed into eigenmode component data. The system uses a collective empirical mode decomposition analysis algorithm. This algorithm is a mathematical method that can decompose the signal into eigenmodes of different frequencies. Parameter settings are critical to the performance of the algorithm. The system selects appropriate parameters to ensure accuracy and effectiveness of the modal decomposition. For example, the system sets algorithm parameters such as sampling frequency and decomposition scale. And then, according to the set parameters, carrying out modal decomposition on the initial multidimensional sensing signal data by the system to obtain a plurality of intrinsic modal component data. These component data represent the vibrational modes of the signal at different frequencies. The system analyzes these eigenmode component data using multi-scale entropy. Multiscale entropy is a feature analysis method that can be used to measure the complexity and randomness of signals. The system applies a multi-scale entropy algorithm to each eigenmode component data to obtain multi-scale entropy value features. The multi-scale entropy value features reflect the information entropy of the signal at different time scales. With these features, the system better understands the complexity and randomness of the signal. In the feature extraction process, each intrinsic mode component data will obtain a corresponding multi-scale entropy feature. The system uses a support vector machine model to perform feature classification to obtain noise aliasing component data. The support vector machine is a machine learning algorithm that can be used to classify problems. The system takes as input multi-scale entropy features, and the support vector machine model will classify each eigenmode component data as noise aliased component data or non-noise aliased component data according to these features. For example, assume that the system is monitoring an isolation switch in a power system, and a current signal is collected by a current sensor. The system uses an aggregate empirical mode decomposition analysis algorithm to perform mode decomposition on the current signal to obtain intrinsic mode component data of different frequencies. The system applies a multi-scale entropy algorithm to each eigenmode component data to calculate a multi-scale entropy value characteristic. These features will reflect the entropy of the current signal at different time scales. The system inputs the multi-scale entropy characteristics into a support vector machine model for classification. The model will classify each eigenmode component data and determine if they are noise aliased component data. This helps the system identify and isolate the noise aliasing components to better understand the operating state and detection performance of the disconnector. By this feature, the system analyses the sensed signal of the disconnector deeper, extracts features, classifies noise aliasing components, and thereby provides more accurate data and information for operational state detection and life prediction.

S103, performing signal filtering processing on the initial multi-dimensional sensing signal data according to the noise aliasing component data through a preset back propagation neural network and a Kalman filter to obtain standard multi-dimensional sensing signal data;

it should be noted that the system constructs and trains the back propagation neural network. The purpose of this neural network is to predict from the noise aliasing component data, thereby helping to filter out noise aliasing. The noise aliasing component data is taken as input data, and the initial multidimensional sensing signal data is taken as target output of the neural network. In this way, the neural network will learn how to recover the original signal from the noise aliasing component data. The structure of the back propagation neural network is designed to comprise an input layer, a hidden layer and an output layer. An appropriate neural network architecture is selected to achieve data reduction. The back propagation neural network is trained using existing data sets. The weight and the bias parameters of the neural network are continuously adjusted through a back propagation algorithm, so that the neural network can accurately restore multidimensional sensing signal data. Subsequently, another independent data set is used to verify the performance of the neural network, ensuring that it accurately restores the multi-dimensional sensor signal data. Once the system training is completed, the system uses it to predict new noise aliasing component data. The neural network will generate new estimated data from the input noise aliasing component data. The system inputs the new estimated data into a preset kalman filter to perform signal filtering processing. The kalman filter is a recursive filter for estimating the state of the system and filtering out noise. The kalman filter requires an initial state estimate, which may be new noise aliasing component data generated by a neural network. Subsequently, an observation model and a state transition model of the kalman filter are defined to describe the dynamic variation of the signal and the measurement noise. The new estimation data is combined with the observation model and the state transition model using a kalman filter algorithm to estimate the state of the system. This process will filter out noise resulting in standard multidimensional sensing signal data.

S104, carrying out sensing data conversion and classification on the standard multidimensional sensing signal data to obtain target current data, target pressure data and target temperature data;

specifically, the standard multidimensional sensing signal data is converted into target multidimensional attribute data. Important features are extracted from the raw signal to better describe the different sensing properties. For example, in a power system, the system extracts frequency components by performing frequency domain analysis on a current signal; performing waveform characteristic analysis on the pressure signal, and extracting the amplitude of pressure change; and carrying out trend analysis on the temperature signal, and extracting the characteristic of temperature rise and fall. The system inputs the obtained target multidimensional attribute data into a preset cluster analysis model. This model will help to divide the data into different clusters, each cluster representing a state or characteristic. At this stage, the system selects an appropriate clustering algorithm, such as K-means clustering, and determines the number of clusters. Cluster analysis helps identify patterns and clusters of different states. The running of the cluster model will produce a cluster center for each cluster. These cluster centers represent the average features of each cluster. For example, for a power system, the initial current cluster center represents the average characteristics of the different current modes, the initial pressure cluster center represents the average characteristics of the different pressure modes, and so on. The system calculates the distance between the target multidimensional attribute data and the initial cluster center. This is typically done using euclidean distance or other distance measures. The data points will be assigned to cluster centers closest to them, thereby determining the clusters to which they belong. The system performs a correction of the cluster center by comparing the distance of the data point from the initial cluster center. This helps to more accurately reflect the distribution of the data, improving the accuracy of clustering. The corrected cluster centers will better represent the average features of the different states or characteristics. And the system performs secondary clustering on the target multidimensional attribute data by using the target current clustering center, the target pressure clustering center and the target temperature clustering center again. This step will divide the data more finely, resulting in target current data, target pressure data, and target temperature data. These data represent more detailed information of different states or characteristics.

S105, respectively carrying out distribution mapping and feature extraction on target current data, target pressure data and target temperature data to obtain a current waveform feature set, a pressure change feature set and a temperature rise and fall feature set;

specifically, for target current data, the system applies a current spectrogram map, converting the time domain signal to the frequency domain. This will produce a current profile containing information about the distribution of the current signal at different frequencies; for target pressure data, the system performs pressure curve conversion to better understand the dynamic changes in pressure. This will generate a pressure profile reflecting the pressure over time; and for the target temperature data, the system performs temperature distribution mapping to generate a temperature distribution box diagram. Such a graph can display the distribution range of the temperature data and the abnormal situation. From the current distribution spectrogram, the system identifies the current waveform characteristics. This may include extracting information of the spectral content, frequency, amplitude, etc. of the waveform. These features help describe the ripple and periodicity of the current signal; from the pressure profile, the system detects a pressure change characteristic. This may include extracting features of slope, fluctuation amplitude, peak, etc. of the pressure to describe dynamic changes in pressure; and extracting the temperature lifting characteristic from the temperature distribution box diagram by the system. This includes extracting information of upper and lower quartiles, outliers, etc. of the box map to reflect the rise and fall of the temperature. Meanwhile, the system screens and integrates the extracted initial features. This is to reduce redundancy of features and improve the efficiency of the deep learning model. Typically, the system uses statistical methods or domain knowledge to select the most relevant features.

S106, carrying out feature coding and vector fusion on the current waveform feature set, the pressure change feature set and the temperature rise and fall feature set to obtain a target fusion feature vector;

specifically, the system performs feature normalization processing on the current waveform feature set, the pressure change feature set and the temperature rise and fall feature set. This is to ensure that the different features have similar dimensions and ranges so that they are easier to compare and combine. The system performs feature normalization, mapping the value of each feature to a standard range, typically [0,1]. This helps to eliminate scale differences between different features, ensuring that their contributions during model training are equal. For each feature, the system performs feature encoding to convert the raw feature values into a more expressive representation. This may include the use of one-hot coding, tag coding, or other coding methods so that the deep learning model can better understand these features. The system then converts the single feature codes into feature vectors to combine the different features into one vector. This vector will contain a plurality of features, the encoded value of each feature being incorporated into an overall vector to provide more comprehensive information. Meanwhile, the system performs weight analysis on the target current data, the target pressure data and the target temperature data. The relative importance of the different attributes for the running state detection is determined. This may be based on a priori knowledge, feature importance analysis, or other methods. The system uses the target weight data to vector fuse the current waveform feature vector, the pressure change feature vector, and the temperature rise feature vector. This involves linearly combining the different feature vectors according to their weights to obtain the target fusion feature vector.

S107, inputting the target fusion feature vector into a preset isolating switch life prediction model to predict the life of the isolating switch, obtaining isolating switch life prediction data, and creating a target running state parameter control scheme of the target isolating switch according to the isolating switch life prediction data.

Specifically, the system inputs the target fusion feature vector into a preset isolating switch life prediction model. This model includes different sub-models, such as decision tree models, GRU models, and XGBoost models. Each of these sub-models has its own unique capabilities and can analyze different types of data and features. For the decision tree model, it predicts the life of the disconnector by building a tree. The decision tree model takes as input the eigenvalues in the target fusion eigenvector and makes a series of decisions based on these eigenvalues, ultimately yielding first life prediction data regarding the disconnector life. The GRU model is a recurrent neural network, and is particularly suitable for processing time series data. It provides more comprehensive information by analyzing the temporal correlation and sequence patterns in the target fusion feature vector to generate second life prediction data. The XGBoost model is an integrated learning algorithm and combines the strengths of multiple decision tree models. By considering the information of multiple aspects, the third life prediction data is generated, and the accuracy is improved. The system assigns weights to each sub-model to perform a weighted average process on their prediction results. These weights may be adjusted according to the performance and reliability of the model to reflect their relative contribution in life prediction. In this way, the system obtains predictive data of the disconnector lifetime. The disconnector lifetime prediction data will be used to create an operating state parameter control scheme for the target disconnector. This includes analyzing the current state of the disconnector, taking into account life prediction data, and determining an optimal operational control scheme. The operating state parameters include key parameters such as operating time, current value, etc., which are determined by an algorithm. To further optimize the operating state parameter control scheme, the system uses a particle swarm algorithm. This algorithm searches for the optimal parameter configuration by moving the virtual particles in the parameter space. Through this process, the system generates a final target operating state parameter control scheme to ensure safe operation of the disconnector and reliability of the power system.

In the embodiment of the invention, the target isolating switch is subjected to multidimensional data acquisition through the preset sensor group, so that the information from different sensors is fully integrated, and the overall perception of the isolating switch state is improved. By using a combined empirical mode decomposition analysis (CEEMDAN) technology, a plurality of intrinsic mode components can be effectively extracted, complex state changes of the isolating switch can be better reflected, and accuracy and reliability of signal decomposition are enhanced. The characteristic extraction and classification are carried out on the intrinsic mode components by utilizing a multi-scale entropy algorithm, so that the accurate distinguishing capability of noise aliasing components is improved, and the noise components can be accurately identified and filtered. And by combining the counter propagation neural network and the Kalman filter, the noise aliasing component is learned and predicted through the deep learning network, and then the signal filtering processing is performed through the Kalman filter, so that the accurate filtering and recovering capacity of the initial multidimensional sensing signal is improved. The standard multidimensional sensing signal data is subjected to sensing data conversion and classification, so that key information such as target current, target pressure and target temperature can be better extracted, and more targeted data can be provided for subsequent analysis. Through multi-level processing such as distribution mapping, feature extraction, coding and vector fusion, the characteristics of multiple aspects such as current waveform, pressure change and temperature rise and fall are successfully extracted and fused, and the state of the isolating switch is more comprehensively represented. The target fusion feature vector is input into a preset isolating switch life prediction model, the multi-model comprehensive effects of decision trees, GRU, XGBoost and the like are fully utilized, and the accuracy and the robustness of life prediction are improved. The life prediction result is optimized through a preset particle swarm algorithm, so that the robustness and the reliability of a control scheme are improved, the operation state parameters of the target isolating switch are controlled more accurately, and the accuracy of the operation state detection of the isolating switch is improved.

In a specific embodiment, the process of executing step S101 may specifically include the following steps:

(1) Inputting the initial multidimensional sensing signal into a preset integrated empirical mode decomposition analysis algorithm, and performing parameter setting on the integrated empirical mode decomposition analysis algorithm to obtain a plurality of target decomposition parameters;

(2) Performing modal decomposition on the initial multidimensional sensing signal data according to a plurality of target decomposition parameters to obtain a plurality of intrinsic modal component data;

(3) Respectively extracting characteristics of a plurality of intrinsic mode component data through a preset multi-scale entropy algorithm to obtain multi-scale entropy value characteristics of each intrinsic mode component data;

(4) And carrying out feature classification on the multi-scale entropy value features of each eigenvector component data through a preset support vector machine model to obtain noise aliasing component data.

Specifically, the system inputs the initial multidimensional sensing signal into a preset aggregate empirical mode decomposition analysis algorithm, and sets parameters of the algorithm. This algorithm is a key tool for processing multidimensional sensing signal data, and it can implement signal decomposition and feature extraction according to parameter setting. Through the parameter setting, the system obtains a plurality of target decomposition parameters which are used for carrying out modal decomposition on the initial multi-dimensional sensing signal data. Modal decomposition is the decomposition of a signal into different modal components in order to better understand the characteristics and structure of the signal. These eigenmode component data will become the basis for subsequent analysis. The system uses a preset multi-scale entropy algorithm to perform feature extraction on the plurality of intrinsic mode component data. The multi-scale entropy is an index for measuring the complexity of a signal, and can reflect the information entropy of the signal under different scales. Through multi-scale entropy analysis, the system obtains multi-scale entropy value features of each eigenmode component data, which help describe the multi-scale characteristics of the signal. The system uses a preset Support Vector Machine (SVM) model to conduct feature classification on the multi-scale entropy value features of each eigenvector component data so as to obtain noise aliasing component data. SVM is a supervised learning algorithm that classifies data and identifies different classes of features. Here, the SVM is used to distinguish a useful signal component from a noise aliasing component.

In a specific embodiment, as shown in fig. 2, the process of executing step S102 may specifically include the following steps:

s201, taking noise aliasing component data as input data of a preset back propagation neural network, and simultaneously taking initial multidimensional sensing signal data as target output of the back propagation neural network;

s202, training a neural network of the back propagation neural network, and adjusting the weight and the bias parameters of the back propagation neural network to obtain the trained back propagation neural network;

s203, predicting the noise aliasing component data through the trained back propagation neural network to obtain new noise aliasing component data;

s204, inputting the new noise aliasing component data into a preset Kalman filter, and performing signal filtering processing on the initial multi-dimensional sensing signal data to obtain standard multi-dimensional sensing signal data.

Specifically, the system takes noise aliasing component data as input data of a preset back propagation neural network, and simultaneously takes initial multidimensional sensing signal data as target output of the back propagation neural network. This is to train the neural network to learn how to convert the noise aliasing component data into standard multi-dimensional sensing signal data. The system trains the neural network of the back propagation neural network and adjusts the weight and the bias parameters thereof. Through a large amount of training data, the neural network will learn how to map the noise aliasing component data to the standard multidimensional sensing signal data. Once trained, the system has a neural network model that predicts noise aliasing component data. And predicting the noise aliasing component data by using the trained back propagation neural network to obtain new noise aliasing component data. These new data will be closer to standard multidimensional sensing signal data because neural networks have learned how to eliminate noise and aliasing effects. The new noise aliasing component data is input into a preset Kalman filter, and signal filtering processing is carried out on the initial multidimensional sensing signal data. The kalman filter is a filter for estimating a signal state and removing noise, which can further improve accuracy and reliability of data. Through this step, the system obtains standard multidimensional sensing signal data, which has been minimized by noise and aliasing effects, and can be used for subsequent state detection and analysis.

In a specific embodiment, as shown in fig. 3, the process of executing step S103 may specifically include the following steps:

s301, carrying out sensing data conversion on standard multidimensional sensing signal data to obtain target multidimensional attribute data;

s302, inputting the target multidimensional attribute data into a preset cluster analysis model, and carrying out cluster center calculation on the target multidimensional attribute data through the cluster analysis model to obtain a corresponding initial current cluster center, an initial voltage cluster center and an initial temperature cluster center;

s303, performing data distance calculation on the target multidimensional attribute data, the initial current clustering center, the initial voltage clustering center and the initial temperature clustering center to obtain a data distance calculation result;

s304, carrying out cluster center correction on the initial current cluster center, the initial voltage cluster center and the initial temperature cluster center according to the data distance calculation result to obtain a target current cluster center, a target voltage cluster center and a target temperature cluster center;

s305, performing secondary clustering on the target multidimensional attribute data according to the target current clustering center, the target voltage clustering center and the target temperature clustering center to obtain target current data, target pressure data and target temperature data.

Specifically, the system converts the standard multidimensional sensing signal data into sensing data so as to obtain target multidimensional attribute data. This process includes processing and feature extraction of raw sensor signal data to obtain attribute information describing the state of the disconnector. These properties cover multidimensional information of current, voltage, temperature, etc. The target multidimensional attribute data is input into a preset cluster analysis model, and a clustering algorithm such as K-means clustering, hierarchical clustering and the like is generally adopted. The models group the target attribute data and calculate an initial current cluster center, an initial voltage cluster center and an initial temperature cluster center. These cluster centers represent typical eigenvalues of the different attribute data. And carrying out data distance calculation on the target multidimensional attribute data, the initial current clustering center, the initial voltage clustering center and the initial temperature clustering center to obtain a data distance calculation result. This step is to quantify the similarity or distance between the target data and the cluster center, so that the features of the data are better understood. And carrying out cluster center correction on the initial current cluster center, the initial voltage cluster center and the initial temperature cluster center based on the data distance calculation result. The system is better adapted to the target data by optimizing the clustering center, and ensures that the clustering center can accurately represent the characteristics of the target attribute data. And performing secondary clustering on the target multidimensional attribute data based on the target current clustering center, the target voltage clustering center and the target temperature clustering center. This may help the system to further group the attribute data to obtain target current data, target pressure data, and target temperature data. These target data will be used for detection and analysis of the disconnector state.

In a specific embodiment, as shown in fig. 4, the process of executing step S104 may specifically include the following steps:

s401, performing current spectrogram mapping on target current data to obtain a current distribution spectrogram, performing pressure curve conversion on target pressure data to obtain a pressure distribution curve, and performing temperature distribution mapping on target temperature data to obtain a temperature distribution box diagram;

s402, carrying out current waveform characteristic identification on a current distribution spectrogram to obtain a plurality of initial current waveform characteristics, and carrying out characteristic screening and integrated conversion on the plurality of initial current waveform characteristics to obtain a current waveform characteristic set;

s403, detecting pressure change characteristics of the pressure distribution curve to obtain a plurality of initial pressure change characteristics, and carrying out characteristic screening and aggregation conversion on the plurality of initial pressure change characteristics to obtain a pressure change characteristic set;

s404, extracting temperature lifting features of the temperature distribution box diagram to obtain a plurality of initial temperature lifting features, and carrying out feature screening and collection conversion on the plurality of initial temperature lifting features to obtain a temperature lifting feature set.

Specifically, the system performs current spectrogram mapping on target current data, which is a signal processing process, and obtains a current distribution spectrogram by converting a time domain signal into a frequency domain signal. Meanwhile, the system performs pressure curve conversion on the target pressure data, and converts the target pressure data into a pressure distribution curve. In addition, temperature distribution mapping is performed on the target temperature data, and a temperature distribution box diagram is generated. These operations help to convert raw sensor signal data into a more informative form for subsequent feature extraction. And the system performs current waveform characteristic identification on the current distribution spectrogram. The objective is to extract key waveform features, such as frequency components, harmonics, etc., from the frequency domain representation of the current signal. These features help to better understand the nature of the current signal. Through the current waveform characteristic recognition, the system obtains a plurality of initial current waveform characteristics. For pressure data, the system performs pressure change feature detection. This involves analyzing the change in pressure profile to detect the increasing or decreasing trend and amplitude change in pressure. These pressure change features help to understand the operating state of the disconnector, as they reflect pressure fluctuations. Similarly, the system obtains a plurality of initial pressure change characteristics. And extracting temperature lifting characteristics of the system according to the temperature data. The purpose is to analyze a temperature distribution box diagram to identify the rise and fall of temperature. The change in temperature may reflect the operating state and stability of the device, and thus these features may facilitate detection of the state of the disconnector. From which the system obtains a plurality of initial temperature ramp features.

In a specific embodiment, the process of executing step S105 may specifically include the following steps:

(1) Carrying out characteristic standardization processing on the current waveform characteristic set, the pressure change characteristic set and the temperature rise and fall characteristic set in a distribution manner to obtain a plurality of standard current waveform characteristics, a plurality of standard pressure change characteristics and a plurality of standard temperature rise and fall characteristics;

(2) Respectively carrying out characteristic normalization processing on the plurality of standard current waveform characteristics, the plurality of standard pressure change characteristics and the plurality of standard temperature rise and fall characteristics to obtain a plurality of normalized current waveform characteristics, a plurality of normalized pressure change characteristics and a plurality of normalized temperature rise and fall characteristics;

(3) Respectively carrying out feature coding on a plurality of normalized current waveform features, a plurality of normalized pressure change features and a plurality of normalized temperature rise and fall features to obtain a plurality of current waveform feature codes, a plurality of pressure change feature codes and a plurality of temperature rise and fall feature codes;

(4) Vector conversion is carried out on the plurality of current waveform feature codes, the plurality of pressure change feature codes and the plurality of temperature rise feature codes respectively to obtain a current waveform feature vector, a pressure change feature vector and a temperature rise feature vector;

(5) Performing weight analysis on the target current data, the target pressure data and the target temperature data to obtain target weight data;

(6) And carrying out vector fusion on the current waveform feature vector, the pressure change feature vector and the temperature rise feature vector according to the target weight data to obtain a target fusion feature vector.

Specifically, the system performs feature normalization processing. The system performs mean value decentralization and standard deviation scaling on the current waveform feature set, the pressure change feature set and the temperature rise and fall feature set. This helps ensure that the feature data has zero mean and unit variance to reduce scale differences between different features. And carrying out feature normalization processing. The system scales the normalized feature values to a specific range, typically [0,1] or [ -1,1]. This helps to ensure that all feature values are within similar ranges, avoiding some features from having an excessive impact on the model. Subsequently, feature encoding is performed. The system processes the current waveform signature code, the pressure change signature code, and the temperature rise and fall signature code to convert them into a numerical form that can be used by the model. Vector conversion is performed. The feature codes are converted into feature vectors, which are suitable for the deep learning model. Vector conversion involves concatenating feature codes together or combining them into one feature vector using other methods. Vector fusion is performed. The system uses the target weight data to weight and fuse the current waveform feature vector, the pressure change feature vector and the temperature rise and fall feature vector in consideration of the importance of each feature. This helps determine which features are more influential to the final state detection task and how to combine them into a comprehensive target fusion feature vector.

In a specific embodiment, the process of executing step S106 may specifically include the following steps:

(1) Inputting the target fusion feature vector into a preset isolating switch life prediction model, wherein the isolating switch life prediction model comprises a decision tree model, a GRU model and an XGBoost model;

(2) Performing isolating switch life prediction on the target fusion feature vector through a decision tree model to obtain first life prediction data, performing isolating switch life prediction on the target fusion feature vector through a GRU model to obtain second life prediction data, and performing isolating switch life prediction on the target fusion feature vector through an XGBoost model to obtain third life prediction data;

(3) According to a preset model weight proportion, carrying out weighted average processing on the first life prediction data, the second life prediction data and the third life prediction data to obtain the life prediction data of the isolating switch;

(4) According to the life prediction data of the isolating switch, carrying out state parameter control scheme analysis on the target isolating switch to obtain an initial running state parameter control scheme;

(5) And carrying out parameter control optimization on the initial running state parameter control scheme through a preset particle swarm algorithm to generate a target running state parameter control scheme.

Specifically, the system prepares a disconnector life prediction model that includes a decision tree model, a GRU model, and an XGBoost model. These models have different characteristics and applicable scenarios to provide accurate life predictions in different situations. The system inputs the target fusion feature vector into these models. And predicting the service life of the isolating switch by using the decision tree model to the target fusion feature vector, and obtaining first service life prediction data by using the system. Decision tree models are generally applicable where the data has discrete features. And predicting the service life of the isolating switch by using the GRU model to obtain second service life prediction data. The GRU model is a cyclic neural network suitable for use with sequence data and can process time series information, wherein the GRU model comprises two layers of GRU neural networks, each layer of GRU neural network comprising 256 GRU units. And then, performing isolating switch life prediction on the target fusion feature vector through an XGBoost model to obtain third life prediction data. XGBoost is a powerful integrated learning model and is suitable for various data types. The system obtains life prediction results of three different models. But different models have different accuracies in different situations. Therefore, the system carries out weighted average processing on the first life prediction data, the second life prediction data and the third life prediction data through the preset model weight proportion, and final life prediction data of the isolating switch is obtained. This step ensures that the advantages of each model are taken into account, thereby improving the accuracy of the prediction. The system uses the predicted data of the service life of the isolating switch to analyze the state parameter control scheme of the target isolating switch. This step helps determine how to manage and maintain the disconnector to extend its service life, improve performance and reduce failure rate. The system will formulate an initial operating state parameter control scheme based on the life prediction data. And (3) carrying out parameter control optimization on the initial running state parameter control scheme by the system through a preset particle swarm algorithm. The particle swarm algorithm is a heuristic optimization method for finding the best combination of parameters to run the disconnector over a predicted lifetime and maintain its performance. This optimization process takes into account various factors including lifetime, performance, and resource utilization, etc., to generate a final target operating state parameter control scheme.

The method for detecting the disconnecting switch based on the deep learning in the embodiment of the present invention is described above, and the device for detecting the disconnecting switch based on the deep learning in the embodiment of the present invention is described below, referring to fig. 5, one embodiment of the device for detecting the disconnecting switch based on the deep learning in the embodiment of the present invention includes:

the acquisition module 501 is used for acquiring multidimensional data of the running state of the target isolating switch through a preset sensor group to obtain initial multidimensional sensing signal data;

the decomposition module 502 is configured to perform modal decomposition on the initial multidimensional sensing signal data to obtain a plurality of intrinsic modal component data, and perform feature extraction and feature classification on the plurality of intrinsic modal component data through a preset multi-scale entropy algorithm to obtain noise aliasing component data;

the processing module 503 is configured to perform signal filtering processing on the initial multi-dimensional sensing signal data according to the noise aliasing component data through a preset counter-propagating neural network and a kalman filter, so as to obtain standard multi-dimensional sensing signal data;

the classification module 504 is configured to perform sensing data conversion and classification on the standard multidimensional sensing signal data to obtain target current data, target pressure data and target temperature data;

The extracting module 505 is configured to perform distribution mapping and feature extraction on the target current data, the target pressure data, and the target temperature data, respectively, to obtain a current waveform feature set, a pressure change feature set, and a temperature rise feature set;

the fusion module 506 is configured to perform feature encoding and vector fusion on the current waveform feature set, the pressure change feature set, and the temperature rise and fall feature set to obtain a target fusion feature vector;

the creating module 507 is configured to input the target fusion feature vector into a preset isolating switch life prediction model to perform isolating switch life prediction, obtain isolating switch life prediction data, and create a target operating state parameter control scheme of the target isolating switch according to the isolating switch life prediction data.

Through the cooperation of the components, the target isolating switch is subjected to multidimensional data acquisition through the preset sensor group, so that information from different sensors is fully integrated, and the overall perception of the state of the isolating switch is improved. By using a combined empirical mode decomposition analysis (CEEMDAN) technology, a plurality of intrinsic mode components can be effectively extracted, complex state changes of the isolating switch can be better reflected, and accuracy and reliability of signal decomposition are enhanced. The characteristic extraction and classification are carried out on the intrinsic mode components by utilizing a multi-scale entropy algorithm, so that the accurate distinguishing capability of noise aliasing components is improved, and the noise components can be accurately identified and filtered. And by combining the counter propagation neural network and the Kalman filter, the noise aliasing component is learned and predicted through the deep learning network, and then the signal filtering processing is performed through the Kalman filter, so that the accurate filtering and recovering capacity of the initial multidimensional sensing signal is improved. The standard multidimensional sensing signal data is subjected to sensing data conversion and classification, so that key information such as target current, target pressure and target temperature can be better extracted, and more targeted data can be provided for subsequent analysis. Through multi-level processing such as distribution mapping, feature extraction, coding and vector fusion, the characteristics of multiple aspects such as current waveform, pressure change and temperature rise and fall are successfully extracted and fused, and the state of the isolating switch is more comprehensively represented. The target fusion feature vector is input into a preset isolating switch life prediction model, the multi-model comprehensive effects of decision trees, GRU, XGBoost and the like are fully utilized, and the accuracy and the robustness of life prediction are improved. The life prediction result is optimized through a preset particle swarm algorithm, so that the robustness and the reliability of a control scheme are improved, the operation state parameters of the target isolating switch are controlled more accurately, and the accuracy of the operation state detection of the isolating switch is improved.

Fig. 5 above describes the deep learning-based disconnecting switch detection device in the embodiment of the present invention in detail from the point of view of the modularized functional entity, and the deep learning-based disconnecting switch detection device in the embodiment of the present invention is described in detail from the point of view of hardware processing.

Fig. 6 is a schematic structural diagram of a deep learning-based isolation switch detection device 600 according to an embodiment of the present invention, where the deep learning-based isolation switch detection device 600 may have a relatively large difference due to configuration or performance, and may include one or more processors (central processing units, CPU) 610 (e.g., one or more processors) and a memory 620, and one or more storage media 630 (e.g., one or more mass storage devices) storing application programs 633 or data 632. Wherein the memory 620 and the storage medium 630 may be transitory or persistent storage. The program stored on the storage medium 630 may include one or more modules (not shown), each of which may include a series of instruction operations on the deep learning based on the isolation switch detection device 600. Still further, the processor 610 may be configured to communicate with the storage medium 630 to execute a series of instruction operations in the storage medium 630 on the deep learning based on the isolation switch detection device 600.

The deep learning based isolation switch detection device 600 may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input/output interfaces 660, and/or one or more operating systems 631, such as Windows Server, mac OS X, unix, linux, freeBSD, and the like. It will be appreciated by those skilled in the art that the deep learning based isolator detection device structure shown in fig. 6 is not limiting of the deep learning based isolator detection device and may include more or fewer components than shown, or may combine certain components, or a different arrangement of components.

The invention also provides a deep learning-based disconnecting switch detection device, which comprises a memory and a processor, wherein the memory stores computer readable instructions, and the computer readable instructions, when executed by the processor, cause the processor to execute the steps of the deep learning-based disconnecting switch detection method in the above embodiments.

The present invention also provides a computer readable storage medium, which may be a non-volatile computer readable storage medium, and may also be a volatile computer readable storage medium, where instructions are stored in the computer readable storage medium, when the instructions are executed on a computer, cause the computer to perform the steps of the deep learning-based method for detecting a disconnector.

It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. The isolating switch detection method based on the deep learning is characterized by comprising the following steps of:

the operation state of the target isolating switch is subjected to multidimensional data acquisition through a preset sensor group, so that initial multidimensional sensing signal data are obtained;

performing modal decomposition on the initial multidimensional sensing signal data to obtain a plurality of intrinsic modal component data, and respectively performing feature extraction and feature classification on the plurality of intrinsic modal component data through a preset multiscale entropy algorithm to obtain noise aliasing component data;

performing signal filtering processing on the initial multi-dimensional sensing signal data according to the noise aliasing component data through a preset back propagation neural network and a Kalman filter to obtain standard multi-dimensional sensing signal data;

Performing sensing data conversion and classification on the standard multidimensional sensing signal data to obtain target current data, target pressure data and target temperature data;

respectively carrying out distribution mapping and feature extraction on the target current data, the target pressure data and the target temperature data to obtain a current waveform feature set, a pressure change feature set and a temperature rise and fall feature set;

performing feature coding and vector fusion on the current waveform feature set, the pressure change feature set and the temperature rise and fall feature set to obtain a target fusion feature vector;

and inputting the target fusion feature vector into a preset isolating switch life prediction model to predict the life of the isolating switch, obtaining isolating switch life prediction data, and creating a target running state parameter control scheme of the target isolating switch according to the isolating switch life prediction data.

2. The method for detecting the isolating switch based on deep learning according to claim 1, wherein the performing modal decomposition on the initial multidimensional sensing signal data to obtain a plurality of intrinsic modal component data, and performing feature extraction and feature classification on the plurality of intrinsic modal component data through a preset multiscale entropy algorithm to obtain noise aliasing component data respectively, comprises:

Inputting the initial multidimensional sensing signal into a preset set empirical mode decomposition analysis algorithm, and performing parameter setting on the set empirical mode decomposition analysis algorithm to obtain a plurality of target decomposition parameters;

performing modal decomposition on the initial multidimensional sensing signal data according to the target decomposition parameters to obtain a plurality of intrinsic modal component data;

respectively extracting the characteristics of the plurality of intrinsic mode component data through a preset multi-scale entropy algorithm to obtain multi-scale entropy characteristics of each intrinsic mode component data;

and carrying out feature classification on the multi-scale entropy value features of each eigenvector component data through a preset support vector machine model to obtain noise aliasing component data.

3. The method for detecting a deep learning-based disconnecting switch according to claim 1, wherein the performing signal filtering processing on the initial multi-dimensional sensing signal data according to the noise aliasing component data through a preset back propagation neural network and a kalman filter to obtain standard multi-dimensional sensing signal data comprises:

taking the noise aliasing component data as input data of a preset back propagation neural network, and simultaneously taking the initial multidimensional sensing signal data as target output of the back propagation neural network;

Training the back propagation neural network, and adjusting the weight and bias parameters of the back propagation neural network to obtain a trained back propagation neural network;

predicting the noise aliasing component data through the trained back propagation neural network to obtain new noise aliasing component data;

and inputting the new noise aliasing component data into a preset Kalman filter, and performing signal filtering processing on the initial multi-dimensional sensing signal data to obtain standard multi-dimensional sensing signal data.

4. The method for deep learning based on the isolating switch detection of claim 1, wherein said performing the sensor data conversion and classification on the standard multidimensional sensor signal data to obtain the target current data, the target pressure data and the target temperature data comprises:

performing sensing data conversion on the standard multidimensional sensing signal data to obtain target multidimensional attribute data;

inputting the target multidimensional attribute data into a preset cluster analysis model, and performing cluster center calculation on the target multidimensional attribute data through the cluster analysis model to obtain a corresponding initial current cluster center, an initial voltage cluster center and an initial temperature cluster center;

Carrying out data distance calculation on the target multidimensional attribute data, the initial current clustering center, the initial voltage clustering center and the initial temperature clustering center to obtain a data distance calculation result;

according to the data distance calculation result, carrying out cluster center correction on the initial current cluster center, the initial voltage cluster center and the initial temperature cluster center to obtain a target current cluster center, a target voltage cluster center and a target temperature cluster center;

and performing secondary clustering on the target multidimensional attribute data according to the target current clustering center, the target voltage clustering center and the target temperature clustering center to obtain target current data, target pressure data and target temperature data.

5. The deep learning-based disconnecting switch detection method according to claim 1, wherein the performing distribution mapping and feature extraction on the target current data, the target pressure data, and the target temperature data to obtain a current waveform feature set, a pressure change feature set, and a temperature rise feature set respectively includes:

performing current spectrogram mapping on the target current data to obtain a current distribution spectrogram, performing pressure curve conversion on the target pressure data to obtain a pressure distribution curve, and performing temperature distribution mapping on the target temperature data to obtain a temperature distribution box diagram;

Carrying out current waveform characteristic identification on the current distribution spectrogram to obtain a plurality of initial current waveform characteristics, and carrying out characteristic screening and integrated conversion on the plurality of initial current waveform characteristics to obtain a current waveform characteristic set;

detecting the pressure change characteristics of the pressure distribution curve to obtain a plurality of initial pressure change characteristics, and carrying out characteristic screening and integrated conversion on the plurality of initial pressure change characteristics to obtain a pressure change characteristic set;

and extracting temperature lifting features of the temperature distribution box diagram to obtain a plurality of initial temperature lifting features, and carrying out feature screening and integrated conversion on the plurality of initial temperature lifting features to obtain a temperature lifting feature set.

6. The method for detecting a deep learning-based isolating switch according to claim 1, wherein the performing feature encoding and vector fusion on the current waveform feature set, the pressure change feature set and the temperature rise and fall feature set to obtain a target fusion feature vector comprises:

carrying out characteristic standardization processing on the current waveform characteristic set, the pressure change characteristic set and the temperature rise and fall characteristic set in a distribution manner to obtain a plurality of standard current waveform characteristics, a plurality of standard pressure change characteristics and a plurality of standard temperature rise and fall characteristics;

Performing characteristic normalization processing on the plurality of standard current waveform characteristics, the plurality of standard pressure change characteristics and the plurality of standard temperature rise and fall characteristics respectively to obtain a plurality of normalized current waveform characteristics, a plurality of normalized pressure change characteristics and a plurality of normalized temperature rise and fall characteristics;

performing feature coding on the plurality of normalized current waveform features, the plurality of normalized pressure change features and the plurality of normalized temperature rise and fall features respectively to obtain a plurality of current waveform feature codes, a plurality of pressure change feature codes and a plurality of temperature rise and fall feature codes;

vector conversion is carried out on the plurality of current waveform feature codes, the plurality of pressure change feature codes and the plurality of temperature rise and fall feature codes respectively to obtain a current waveform feature vector, a pressure change feature vector and a temperature rise and fall feature vector;

performing weight analysis on the target current data, the target pressure data and the target temperature data to obtain target weight data;

and carrying out vector fusion on the current waveform feature vector, the pressure change feature vector and the temperature rise and fall feature vector according to the target weight data to obtain a target fusion feature vector.

7. The deep learning-based disconnecting switch detection method according to claim 1, wherein the inputting the target fusion feature vector into a preset disconnecting switch life prediction model to perform disconnecting switch life prediction to obtain disconnecting switch life prediction data, and creating a target operation state parameter control scheme of the target disconnecting switch according to the disconnecting switch life prediction data comprises:

inputting the target fusion feature vector into a preset isolating switch life prediction model, wherein the isolating switch life prediction model comprises a decision tree model, a GRU model and an XGBoost model;

performing isolating switch life prediction on the target fusion feature vector through the decision tree model to obtain first life prediction data, performing isolating switch life prediction on the target fusion feature vector through the GRU model to obtain second life prediction data, and performing isolating switch life prediction on the target fusion feature vector through the XGBoost model to obtain third life prediction data;

according to a preset model weight proportion, carrying out weighted average processing on the first life prediction data, the second life prediction data and the third life prediction data to obtain isolating switch life prediction data;

According to the life prediction data of the isolating switch, carrying out state parameter control scheme analysis on the target isolating switch to obtain an initial running state parameter control scheme;

and carrying out parameter control optimization on the initial running state parameter control scheme through a preset particle swarm algorithm to generate a target running state parameter control scheme.

8. The utility model provides a isolator detection device based on degree of depth study which characterized in that, isolator detection device based on degree of depth study includes:

the acquisition module is used for acquiring multidimensional data of the running state of the target isolating switch through a preset sensor group to obtain initial multidimensional sensing signal data;

the decomposition module is used for carrying out modal decomposition on the initial multidimensional sensing signal data to obtain a plurality of intrinsic modal component data, and carrying out feature extraction and feature classification on the plurality of intrinsic modal component data through a preset multiscale entropy algorithm to obtain noise aliasing component data;

the processing module is used for carrying out signal filtering processing on the initial multi-dimensional sensing signal data according to the noise aliasing component data through a preset counter-propagation neural network and a Kalman filter to obtain standard multi-dimensional sensing signal data;

The classification module is used for carrying out sensing data conversion and classification on the standard multidimensional sensing signal data to obtain target current data, target pressure data and target temperature data;

the extraction module is used for carrying out distribution mapping and feature extraction on the target current data, the target pressure data and the target temperature data respectively to obtain a current waveform feature set, a pressure change feature set and a temperature rise feature set;

the fusion module is used for carrying out feature coding and vector fusion on the current waveform feature set, the pressure change feature set and the temperature rise and fall feature set to obtain a target fusion feature vector;

the creation module is used for inputting the target fusion feature vector into a preset isolating switch life prediction model to predict the life of the isolating switch, obtaining isolating switch life prediction data, and creating a target running state parameter control scheme of the target isolating switch according to the isolating switch life prediction data.

9. Deep learning-based disconnecting switch detection device, characterized in that the deep learning-based disconnecting switch detection device comprises: a memory and at least one processor, the memory having instructions stored therein;

The at least one processor invokes the instructions in the memory to cause the deep learning based isolation switch detection device to perform the deep learning based isolation switch detection method of any of claims 1-7.

10. A computer readable storage medium having instructions stored thereon, which when executed by a processor, implement the deep learning based method of isolation switch detection of any of claims 1-7.