GIS equipment state research and judgment method and device based on IMF energy entropy

Abstract

The application discloses a GIS equipment state research and judgment method and device based on IMF energy entropy, and the method comprises the following steps: acquiring GIS switch state data and corresponding state label values, and constructing a training set; constructing a time-frequency memory recurrent neural network based on IMF energy entropy; designing a cross-entropy loss function by considering a classification result and a real state label value; training the neural network, stopping the training when the training round reaches a maximum training round or the value of the cross-entropy loss function reaches a minimum, so that a trained neural network is obtained; inputting real-time collected GIS switch state data into the trained neural network to obtain a GIS equipment state research and judgment result; and the application has the advantages that the accuracy of GIS equipment state research and judgment is improved.

Description

Technical Field

[0001] This invention relates to the field of GIS equipment status assessment, and more specifically to a method for assessing the status of GIS equipment based on IMF energy entropy. Background Technology

[0002] Gas-Insulated Metal-Enclosed Switchgear (GIS) is a metal-enclosed switchgear that uses gas, or at atmospheric pressure, as the insulating medium, either entirely or partially. Compared to traditional open-type power distribution equipment, GIS has advantages such as small footprint, completely sealed components that are not affected by the environment, convenient operation, low maintenance workload, easy equipment installation, and short construction period.

[0003] A disconnector switch is a switching device primarily used to connect and disconnect small-current circuits, lacking arc-extinguishing functionality. Because disconnectors in GIS equipment are sealed in a metal casing, their exact location cannot be directly observed with the naked eye. Reliable opening and closing of GIS disconnectors is crucial for the safe operation of the equipment and the power grid. According to the 2017 Guangdong Power Grid GIS Disconnector Switch X-ray Special Survey Report, the company took 6976 photographs of GIS equipment disconnectors. Of these, 1013 photographs provided by the manufacturers showed defects, and 1002 of them exhibited defects in the opening and closing of disconnectors, accounting for 99% of all defect types. Although all GIS manufacturers have maintenance cycle limits for their products, due to power outages and construction constraints, most GIS equipment cannot be overhauled during power outages. After prolonged operation and break-in or severe external influences, the opening and closing positions of GIS equipment disconnectors may change. Incomplete opening and closing of disconnectors can cause serious damage to the equipment and the power grid; therefore, accurate assessment of the opening and closing status of GIS equipment disconnectors is necessary.

[0004] Chinese Patent Publication No. CN114781458A discloses a method for identifying initial faults in distribution networks based on a time-frequency memory neural network using adaptive wavelet transform and an attention mechanism. The method includes: 1. Preprocessing and dividing the collected current and voltage data into datasets; 2. Designing the core structure of a time-frequency memory recurrent neural network based on adaptive wavelet transform; 3. Constructing a time-frequency memory recurrent neural network based on adaptive wavelet transform and an attention mechanism; 4. Obtaining the output results of the test set samples using the time-frequency memory recurrent neural network based on adaptive wavelet transform and an attention mechanism. By providing fine-grained analysis of time series data using a time-frequency memory recurrent neural network based on adaptive wavelet transform and an attention mechanism, the method can dynamically capture the features of data in both the time and frequency domains, thereby improving the accuracy of initial fault identification and meeting the practical needs for both accuracy and speed. This patent application does not assess the status of GIS equipment. Even if it were used to assess the status of GIS equipment, the data processing would be affected by the use of wavelet transform. Wavelet transform requires the selection of a suitable mother wavelet and the setting of a feasible number of decomposition layers, which depends on the experience of the data processor. If the mother wavelet and the number of decomposition layers are not set reasonably, the time-frequency decomposition results will not be accurate, thus affecting the final equipment status assessment results and making it difficult to guarantee the accuracy of GIS equipment status assessment. Summary of the Invention

[0005] The technical problem to be solved by this invention is that the accuracy of the status assessment of GIS equipment in the prior art is difficult to guarantee.

[0006] This invention solves the above-mentioned technical problems through the following technical means: a GIS equipment status assessment method based on IMF energy entropy, the method comprising:

[0007] Step 1: Obtain GIS switch status data and their corresponding status label values, and construct a training set;

[0008] Step 2: Construct a time-frequency memory recurrent neural network based on IMF energy entropy. Input GIS switch state data into the neural network to obtain the hidden state of GIS switch state information for all time steps, and use the mean of these hidden states as the GIS switch state feature vector. Use the GIS switch state feature vector to classify GIS switch states, and design a cross-entropy loss function considering the classification results and the true state label values.

[0009] Step 3: Train the neural network. Stop training when the maximum number of training rounds is reached or the value of the cross-entropy loss function reaches its minimum, thus obtaining a trained neural network.

[0010] Step 4: Input the real-time collected GIS switch status data into the trained neural network to obtain the GIS equipment status assessment results.

[0011] Beneficial effects: This invention constructs a time-frequency memory recurrent neural network based on IMF energy entropy, and performs subsequent feature vector extraction and GIS switch status classification based on IMF energy entropy. There are no special requirements for sample selection, no need for human experience to set relevant parameters, and the accuracy of the output results will not be affected by human factors. The method is simple and has a small computational load. The cross-entropy loss function is designed to train the neural network, and the trained neural network is used to judge the status of GIS equipment, which greatly improves the accuracy of the results.

[0012] Further, step one includes:

[0013] Construct datasets of grounding current, surge arrester leakage current, and enclosure vibration for GIS switch maintenance, cold standby, and hot standby states. ,in, This represents a dataset of ground currents. This represents the surge arrester leakage current dataset. This represents a dataset of vibration data for GIS equipment casings.

[0014] Furthermore, step two includes:

[0015] Step 2.1: Construct a time-frequency memory recurrent neural network based on IMF energy entropy. This neural network includes a time-frequency memory recurrent layer and a fully connected layer. The time-frequency memory recurrent layer includes a time-frequency memory unit, a time-frequency memory forgetting part, and a cell renewal unit.

[0016] Step 2.2: Input the GIS switch status data into the time-frequency memory recursive layer to obtain the first... The GIS switch status data in the time-frequency memory unit GIS switch status selection information at time steps The time-frequency decomposition and filtering are performed to obtain each IMF component, and the energy entropy of each IMF component is calculated; the energy entropy is input into the time-frequency memory-forgetting part to obtain the first IMF component. The first GIS switch status data Output set at each time step ;

[0017] Step 2.3, the The first GIS switch status data Data at each time step and the Hidden status values ​​of GIS switch status data at each time step The input to the cell renewal unit yields the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status information to be updated The time-frequency decomposition and filtering are performed to obtain each IMF component, and the energy entropy of each IMF component is calculated. ;

[0018] Step 2.4, the cell renewal unit will... The first GIS switch status data Output set at each time step With energy entropy Adding them together gives the result of the first addition. The first GIS switch status data Memory state at each time step After processing, the first output is... GIS switch status information for all time steps, including hidden status. For hidden state Take the mean value as the GIS switch status feature vector ;

[0019] Step 2.5, GIS switch status feature vector In the fully connected input layer, the cross-entropy loss function is designed to consider both the classification result and the true state label value.

[0020] Furthermore, step 2.2 includes:

[0021] Step 2.2.1, the first The first GIS switch status data Data at each time step In the input time-frequency memory recursive layer, obtain the first... The GIS switch status data in the time-frequency memory unit GIS switch status selection information at time steps :

[0022] (1)

[0023] (2)

[0024] (3)

[0025] in Indicates the first The first GIS switch status data Hidden status values ​​of GIS switch status information over a time step; This represents an activation function that maps variables to values ​​between 0 and 1. , and , All are weight matrices.

[0026] Furthermore, step 2.2 also includes:

[0027] Step 2.2.2: Select information for GIS switch status. Time-frequency decomposition and filtering are performed to obtain each IMF component:

[0028] (4)

[0029] (5)

[0030] (6)

[0031] in For the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status selection information. This refers to the frequency domain information obtained after time-frequency decomposition. For decomposing sequence number and , Angular velocity, This is the output value of this filter. and These are the filter coefficients. and These are two sampled values, The previous sample value, for New information obtained from time-frequency conversion.

[0032] Furthermore, step 2.2 also includes:

[0033] Step 2.2.3: Calculate the energy entropy of each component using equations (8)-(9).

[0034] (7)

[0035] (8)

[0036] (9)

[0037] (10)

[0038] in, This represents the energy value of each IMF component. This represents the total energy value of the IMF components. This indicates the proportion of energy in each component to the total energy. This represents the energy entropy value of each IMF component;

[0039] The forgetting part of the time-frequency memory is obtained using equation (11). The first GIS switch status data Output set at each time step :

[0040] (11)

[0041] in, This represents the outer product operation. It is the first The GIS switch status data in the time-frequency memory unit The memory information of the time step.

[0042] Furthermore, step 2.3 includes:

[0043] Step 2.3.1: The cell renewal unit uses equations (12)-(14) to obtain the first... The first GIS switch status data GIS switch status input information at each time step and GIS switch status modulation information :

[0044] (12)

[0045] (13)

[0046] (14)

[0047] in, For the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status information awaiting update; and Both are input weight matrices. Represents the input bias vector; and All are modulation weight matrices. Represents the modulation deviation vector; This represents element-wise multiplication. For activation functions;

[0048] Step 2.3.2, regarding the first... The GIS switch status data in the time-frequency memory unit Time step GIS switch status information to be updated Perform time-frequency decomposition and filtering to obtain the first... The first GIS switch status data Each IMF component of the GIS switch status information output at each time step ;

[0049] Step 2.3.3: Calculate each IMF component. Energy entropy .

[0050] Furthermore, step 2.4 includes:

[0051] Step 2.4.1: The output gate of the cell renewal unit is used to obtain the first... The first GIS switch status data Memory state at each time step :

[0052] (twenty two)

[0053] The output gate obtains the first... using equation (23). The first GIS switch status data A composite signal with a time step :

[0054] (twenty three)

[0055] in, and Both are weight matrices. This represents the bias vector of the output gate;

[0056] Based on equations (22) and (23), we can obtain the first equation using equations (24) and (25). GIS switch status information for all time steps; hidden status. :

[0057] (twenty four)

[0058] (25)

[0059] Step 2.4.2, take the first... GIS switch status information for all time steps; hidden status. The mean value is used as the feature vector of GIS switch status. .

[0060] Furthermore, step 2.5 includes:

[0061] Step 2.5.1, the GIS switch state feature vector Input into the fully connected layer, and use equation (26) to obtain the first... GIS switch status data with tags The probability value corresponding to all GIS switch state types N :

[0062] (26)

[0063] in, and These represent the weighted average GIS switch state feature vectors, respectively. The weight matrix and bias vector;

[0064] Step 2.5.2: Construct the cross-entropy loss function using equation (27). :

[0065] (27)

[0066] in, Indicates the first The first GIS switch status data The probability of predicting a GIS switch status label at each time step. Indicates the first The first GIS switch status data The true label value for each time step.

[0067] This invention also provides a GIS equipment status assessment device based on IMF energy entropy, the device comprising:

[0068] The training set construction module is used to obtain GIS switch status data and their corresponding status label values ​​to construct a training set.

[0069] The neural network construction module is used to construct a time-frequency memory recurrent neural network based on IMF energy entropy. GIS switch state data is input into the neural network to obtain the hidden state of GIS switch state information at all time steps, and the mean of these hidden states is used as the GIS switch state feature vector. The GIS switch state feature vector is used to classify GIS switch states. The cross-entropy loss function is designed considering the classification results and the true state label values.

[0070] The neural network training module is used to train the neural network. When the maximum number of training rounds is reached, or the value of the cross-entropy loss function reaches its minimum, the training stops, thus obtaining a trained neural network.

[0071] The analysis result output module is used to input the real-time collected GIS switch status data into the trained neural network to obtain the GIS equipment status analysis results.

[0072] Furthermore, the training set construction module is also used for:

[0073] Construct datasets of grounding current, surge arrester leakage current, and enclosure vibration for GIS switch maintenance, cold standby, and hot standby states. ,in, This represents a dataset of ground currents. This represents the surge arrester leakage current dataset. This represents a dataset of vibration data for GIS equipment casings.

[0074] Furthermore, the neural network building module is also used for:

[0075] Step 2.1: Construct a time-frequency memory recurrent neural network based on IMF energy entropy. This neural network includes a time-frequency memory recurrent layer and a fully connected layer. The time-frequency memory recurrent layer includes a time-frequency memory unit, a time-frequency memory forgetting part, and a cell renewal unit.

[0076] Step 2.2: Input the GIS switch status data into the time-frequency memory recursive layer to obtain the first... The GIS switch status data in the time-frequency memory unit GIS switch status selection information at time steps The time-frequency decomposition and filtering are performed to obtain each IMF component, and the energy entropy of each IMF component is calculated; the energy entropy is input into the time-frequency memory-forgetting part to obtain the first IMF component. The first GIS switch status data Output set at each time step ;

[0077] Step 2.3, the The first GIS switch status data Data at each time step and the Hidden status values ​​of GIS switch status data at each time step The input to the cell renewal unit yields the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status information to be updated The time-frequency decomposition and filtering are performed to obtain each IMF component, and the energy entropy of each IMF component is calculated. ;

[0078] Step 2.4, the cell renewal unit will... The first GIS switch status data Output set at each time step With energy entropy Adding them together gives the result of the first addition. The first GIS switch status data Memory state at each time step After processing, the first output is... GIS switch status information for all time steps, including hidden status. For hidden state Take the mean value as the GIS switch status feature vector ;

[0079] Step 2.5, GIS switch status feature vector In the fully connected input layer, the cross-entropy loss function is designed to consider both the classification result and the true state label value.

[0080] Furthermore, step 2.2 includes:

[0081] Step 2.2.1, the first The first GIS switch status data Data at each time step In the input time-frequency memory recursive layer, obtain the first... The GIS switch status data in the time-frequency memory unit GIS switch status selection information at time steps :

[0082] (1)

[0083] (2)

[0084] (3)

[0085] in Indicates the first The first GIS switch status data Hidden status values ​​of GIS switch status information over a time step; This represents an activation function that maps variables to values ​​between 0 and 1. , and , All are weight matrices.

[0086] Furthermore, step 2.2 also includes:

[0087] Step 2.2.2: Select information for GIS switch status. Time-frequency decomposition and filtering are performed to obtain each IMF component:

[0088] (4)

[0089] (5)

[0090] (6)

[0091] in For the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status selection information. This refers to the frequency domain information obtained after time-frequency decomposition. For decomposing sequence number and , Angular velocity, This is the output value of this filter. and These are the filter coefficients. and These are two sampled values, The previous sample value, for New information obtained from time-frequency conversion.

[0092] Furthermore, step 2.2 also includes:

[0093] Step 2.2.3: Calculate the energy entropy of each component using equations (8)-(9).

[0094] (7)

[0095] (8)

[0096] (9)

[0097] (10)

[0098] in, This represents the energy value of each IMF component. This represents the total energy value of the IMF components. This indicates the proportion of energy in each component to the total energy. This represents the energy entropy value of each IMF component;

[0099] The forgetting part of the time-frequency memory is obtained using equation (11). The first GIS switch status data Output set at each time step :

[0100] (11)

[0101] in, This represents the outer product operation. It is the first The GIS switch status data in the time-frequency memory unit The memory information of the time step.

[0102] Furthermore, step 2.3 includes:

[0103] Step 2.3.1: The cell renewal unit uses equations (12)-(14) to obtain the first... The first GIS switch status data GIS switch status input information at each time step and GIS switch status modulation information :

[0104] (12)

[0105] (13)

[0106] (14)

[0107] in, For the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status information awaiting update; and Both are input weight matrices. Represents the input bias vector; and All are modulation weight matrices. Represents the modulation deviation vector; This represents element-wise multiplication. For activation functions;

[0108] Step 2.3.2, regarding the first... The GIS switch status data in the time-frequency memory unit Time step GIS switch status information to be updated Perform time-frequency decomposition and filtering to obtain the first... The first GIS switch status data Each IMF component of the GIS switch status information output at each time step ;

[0109] Step 2.3.3: Calculate each IMF component. Energy entropy .

[0110] Furthermore, step 2.4 includes:

[0111] Step 2.4.1: The output gate of the cell renewal unit is used to obtain the first... The first GIS switch status data Memory state at each time step :

[0112] (twenty two)

[0113] The output gate obtains the first... using equation (23). The first GIS switch status data A composite signal with a time step :

[0114] (twenty three)

[0115] in, and Both are weight matrices. This represents the bias vector of the output gate;

[0116] Based on equations (22) and (23), we can obtain the first equation using equations (24) and (25). GIS switch status information for all time steps; hidden status. :

[0117] (twenty four)

[0118] (25)

[0119] Step 2.4.2, take the first... GIS switch status information for all time steps; hidden status. The mean value is used as the feature vector of GIS switch status. .

[0120] Furthermore, step 2.5 includes:

[0121] Step 2.5.1, the GIS switch state feature vector Input into the fully connected layer, and use equation (26) to obtain the first... GIS switch status data with tags The probability value corresponding to all GIS switch state types N :

[0122] (26)

[0123] in, and These represent the weighted average GIS switch state feature vectors, respectively. The weight matrix and bias vector;

[0124] Step 2.5.2: Construct the cross-entropy loss function using equation (27). :

[0125] (27)

[0126] in, Indicates the first The first GIS switch status data The probability of predicting a GIS switch status label at each time step. Indicates the first The first GIS switch status data The true label value for each time step.

[0127] The advantages of this invention are:

[0128] (1) This invention constructs a time-frequency memory recurrent neural network based on IMF energy entropy, and performs subsequent feature vector extraction and GIS switch status classification based on IMF energy entropy. There are no special requirements for sample selection, no need for human experience to set relevant parameters, and the accuracy of the output results will not be affected by human factors. The method is simple and has a small amount of computation. The cross-entropy loss function is designed to train the neural network, and the trained neural network is used to judge the status of GIS equipment, which greatly improves the accuracy of the results.

[0129] (2) The present invention uses the time-frequency decomposition method based on IMF energy entropy to decompose the input signal. Compared with the wavelet transform method, it is simple to calculate, fast and efficient. It also filters out external influences such as noise and removes IMF components that have little relation to the status judgment of GIS equipment, which greatly improves the accuracy of the results. Attached Figure Description

[0130] Figure 1 This is a flowchart of the GIS equipment status assessment method based on IMF energy entropy disclosed in Embodiment 1 of the present invention. Detailed Implementation

[0131] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0132] Example 1

[0133] like Figure 1As shown, this invention provides a method for assessing the status of GIS equipment based on IMF energy entropy, the method comprising the following steps:

[0134] Step 1: GIS switch status data processing

[0135] Step 1.1: Construct a dataset of grounding current, surge arrester leakage current, and enclosure vibration for GIS switch maintenance status, cold standby status, and hot standby status. ,in, Represents the ground current dataset, and , Indicates the first Grounding current data of the GIS switch status data. , Indicates the first The first GIS switch status data Ground current data at the sampling time; This represents the surge arrester leakage current dataset, and , Indicates the first Surge arrester leakage current data from the GIS switch status data. , Indicates the first The first GIS switch status data Surge arrester leakage current data at the sampling time; This represents a dataset of vibration data from a GIS equipment casing. , Indicates the first Vibration data of the casing of the GIS switch status data. , Indicates the first The first GIS switch status data Vibration data of the outer shell at the sampling time; , Indicates the total number of GIS switch states; , Show the total sampling time.

[0136] Step 1.2: Construct a set of GIS switch status data labels, denoted as... ,in, Indicates the first Label values ​​for each GIS switch status data;

[0137] Step 1.3: Construct the training set for a time-frequency memory recurrent neural network based on IMF energy entropy. ;and , Indicates the first GIS switch status data, and , Indicates the first The first GIS switch status data GIS switch status data at the sampling time. Indicates the first Label values ​​for each GIS switch status data.

[0138] Step 2: Construct a time-frequency memory recurrent neural network based on IMF energy entropy. Input GIS switch state data into this neural network to obtain the hidden states of GIS switch state information for all time steps, and use the average of these hidden states as the GIS switch state feature vector. Use the GIS switch state feature vector to classify GIS switch states, and design a cross-entropy loss function considering the classification results and the true state label values. The process is described in detail below.

[0139] Step 2.1: Construct a time-frequency memory recurrent neural network based on IMF energy entropy, including: a time-frequency memory recurrent layer based on IMF energy entropy and a fully connected layer; the time-frequency memory recurrent layer includes a time-frequency memory unit, a time-frequency memory forgetting part, and a cell renewal unit; wherein, the time-frequency memory forgetting part includes: a time-domain forgetting gate, a frequency-domain forgetting gate, and a state forgetting gate; the cell renewal unit includes: an input gate, an IMF decomposition unit, and an output gate.

[0140] The specific process of step 2.2 is as follows:

[0141] Step 2.2.1: Training set of the GIS switch status judgment network The Middle The first GIS switch status data Data at each time step In the input time-frequency memory recursive layer, the forgotten part of the time-frequency memory is selectively discarded using equations (1)-(3) to obtain the first... The GIS switch status data in the time-frequency memory unit GIS switch status selection information at time steps :

[0142] (1)

[0143] (2)

[0144] (3)

[0145] In equations (1)-(3) Indicates the first The first GIS switch status data The GIS switch status information hides the status value over a time step; when At that time, set the hidden state value to zero; This represents an activation function that maps variables to values ​​between 0 and 1. , and , They represent the first The first GIS switch status data Sampling time data and the The GIS switch status data in the time-frequency memory unit Time step GIS switch status data hidden status The forget gate weight matrix;

[0146] Step 2.2.2: Use equations (4)-(6) to apply the first... The GIS switch status data in the time-frequency memory unit GIS switch status selection information at each time step Perform time-frequency decomposition and filtering to obtain the first... The first GIS switch status data Each IMF component of the GIS switch status information output at each time step:

[0147] (4)

[0148] (5)

[0149] (6)

[0150] In equations (4)-(6) For the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status selection information Frequency domain information obtained after time-frequency decomposition For decomposing sequence number and , Angular velocity, This is the output value of this filter. and These are the filter coefficients. and These are two sampled values, The previous sample value, for New information obtained from time-frequency conversion;

[0151] Step 2.2.3: Calculate the energy entropy of each component using equations (8)-(9).

[0152] (7)

[0153] (8)

[0154] (9)

[0155] (10)

[0156] In equations (7)-(10), This represents the energy value of each IMF component. This represents the total energy value of the IMF components. This indicates the proportion of energy in each component to the total energy. This represents the energy entropy value of each IMF component.

[0157] The forgetting part of the time-frequency memory is obtained using equation (11). The first GIS switch status data Output set at each time step :

[0158] (11)

[0159] In equation (11), This represents the outer product operation. It is the first The GIS switch status data in the time-frequency memory unit The memory information of the time step.

[0160] The specific process of step 2.3 is as follows:

[0161] Step 2.3.1, the first The first GIS switch status data Data at each time step and the Hidden status values ​​of GIS switch status data at each time step In the cell renewal unit, the input gate uses equations (12)-(14) to obtain the first... The first GIS switch status data GIS switch status input information at each time step and GIS switch status modulation information ;

[0162] (12)

[0163] (13)

[0164] (14)

[0165] In equations (12)-(14), For the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status information awaiting update; and They represent the first The first GIS switch status Sampling time data and the The GIS switch status data in the time-frequency memory unit Hidden status information of GIS switch at time step The input weight matrix, Indicates the first Input deviation vector of GIS switch status data; and They represent the first The first GIS switch status Sampling time data and the The GIS switch status data in the time-frequency memory unit Information Hiding State of Time Step The modulation weight matrix, Indicates the first Modulation deviation vector of GIS switch status data; This represents element-wise multiplication. This is the activation function.

[0166] Step 2.3.2: The IMF unit uses equations (15)-(17) to process the first... The GIS switch status data in the time-frequency memory unit Time step GIS switch status information to be updated Perform time-frequency decomposition and filtering to obtain the first... The first GIS switch status data Each IMF component of the GIS switch status information output at each time step:

[0167] (15)

[0168] (16)

[0169] (17)

[0170] In equations (15)-(17) For the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status information awaiting update; This refers to the frequency domain information obtained after time-frequency decomposition. For decomposing sequence number and , Angular velocity, This is the output value of this filter. and These are the filter coefficients. and These are two sampled values, for New information obtained from time-frequency conversion;

[0171] Step 2.3.3: Calculate the energy entropy of each component using equations (19)-(21).

[0172] (18)

[0173] (19)

[0174] (20)

[0175] (twenty one)

[0176] In equations (18)-(21), This represents the energy value of each IMF component. This represents the total energy value of the IMF components. This indicates the proportion of energy in each component to the total energy. This represents the energy entropy value of each IMF component.

[0177] The specific process of step 2.4 is as follows:

[0178] Step 2.4.1: The output gate uses (22) to obtain the first... The first GIS switch status data Memory state at each time step :

[0179] (twenty two)

[0180] In equation (14), Indicates the first The first GIS switch status data The memory state over a time step; when season Zero;

[0181] The output gate uses equation (23) to... The first GIS switch status data Data at each time step and the The first GIS switch status data Hidden status values ​​of GIS switch status information at each time step Process and obtain the first The first GIS switch status data A composite signal with a time step :

[0182] (twenty three)

[0183] In equation (23), and These represent the output gate pairs of the first and second generations. The first GIS switch status data Data at each time step and the The first GIS switch status data Hidden status values ​​of GIS switch status information at each time step The weight matrix at output. This represents the bias vector of the output gate.

[0184] According to equations (22) and (23), the output gate obtains the first... The first GIS switch status data Hidden status values ​​of GIS switch status information at each time step Thus, the first GIS switch status information for all time steps; hidden status.

[0185] (twenty four)

[0186] (25)

[0187] Step 2.4.2, take the first... GIS switch status information for all time steps; hidden status. The mean value is used as the feature vector of GIS switch status. ,use Perform GIS switch status classification.

[0188] The specific process of step 2.5 is as follows:

[0189] Step 2.5.1, the GIS switch state feature vector Input into the fully connected layer, and use equation (26) to obtain the first... GIS switch status data with tags The probability value corresponding to all GIS switch state types N ;

[0190] (26)

[0191] In equation (26), and These represent the weighted average GIS switch state feature vectors, respectively. The weight matrix and bias vector; Indicates the activation function;

[0192] Step 2.5.2: Construct the cross-entropy loss function using equation (27). :

[0193] (27)

[0194] In equation (27), Indicates the first The first GIS switch status data The probability of predicting a GIS switch status label at each time step. Indicates the first The first GIS switch status data The true label value for each time step.

[0195] Step 3: Based on the training set The loss function is then processed using backpropagation and gradient descent. Training continues until the maximum number of training epochs (Epochs) is reached, or the loss function... When the minimum value is reached, training stops, resulting in a trained time-frequency memory recurrent neural network based on IMF energy entropy, which is used to map the input GIS switch state data set to the corresponding GIS switch state category label.

[0196] Step 4: Input the real-time collected GIS switch status data into the trained time-frequency memory recurrent neural network based on IMF energy entropy, and output the GIS switch status.

[0197] Example 2

[0198] Based on Embodiment 1, Embodiment 2 of the present invention also provides a GIS equipment status assessment device based on IMF energy entropy, the device comprising:

[0199] The training set construction module is used to obtain GIS switch status data and their corresponding status label values ​​to construct a training set.

[0200] The neural network construction module is used to construct a time-frequency memory recurrent neural network based on IMF energy entropy. GIS switch state data is input into the neural network to obtain the hidden state of GIS switch state information at all time steps, and the mean of these hidden states is used as the GIS switch state feature vector. The GIS switch state feature vector is used to classify GIS switch states. The cross-entropy loss function is designed considering the classification results and the true state label values.

[0201] The neural network training module is used to train the neural network. When the maximum number of training rounds is reached, or the value of the cross-entropy loss function reaches its minimum, the training stops, thus obtaining a trained neural network.

[0202] The analysis result output module is used to input the real-time collected GIS switch status data into the trained neural network to obtain the GIS equipment status analysis results.

[0203] Specifically, the training set construction module is also used for:

[0204] Construct datasets of grounding current, surge arrester leakage current, and enclosure vibration for GIS switch maintenance, cold standby, and hot standby states. ,in, This represents a dataset of ground currents. This represents the surge arrester leakage current dataset. This represents a dataset of vibration data for GIS equipment casings.

[0205] Specifically, the neural network building module is also used for:

[0206] Step 2.1: Construct a time-frequency memory recurrent neural network based on IMF energy entropy. This neural network includes a time-frequency memory recurrent layer and a fully connected layer. The time-frequency memory recurrent layer includes a time-frequency memory unit, a time-frequency memory forgetting part, and a cell renewal unit.

[0207] Step 2.2: Input the GIS switch status data into the time-frequency memory recursive layer to obtain the first... The GIS switch status data in the time-frequency memory unit GIS switch status selection information at time steps The time-frequency decomposition and filtering are performed to obtain each IMF component, and the energy entropy of each IMF component is calculated; the energy entropy is input into the time-frequency memory-forgetting part to obtain the first IMF component. The first GIS switch status data Output set at each time step ;

[0208] Step 2.3, the The first GIS switch status data Data at each time step and the Hidden status values ​​of GIS switch status data at each time step The input to the cell renewal unit yields the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status information to be updated The time-frequency decomposition and filtering are performed to obtain each IMF component, and the energy entropy of each IMF component is calculated. ;

[0209] Step 2.4, the cell renewal unit will... The first GIS switch status data Output set at each time step With energy entropy Adding them together gives the result of the first addition. The first GIS switch status data Memory state at each time step After processing, the first output is... GIS switch status information for all time steps, including hidden status. For hidden state Take the mean value as the GIS switch status feature vector ;

[0210] Step 2.5, GIS switch status feature vector In the fully connected input layer, the cross-entropy loss function is designed to consider both the classification result and the true state label value.

[0211] More specifically, step 2.2 includes:

[0212] Step 2.2.1, the first The first GIS switch status data Data at each time step In the input time-frequency memory recursive layer, obtain the first... The GIS switch status data in the time-frequency memory unit GIS switch status selection information at time steps :

[0213] (1)

[0214] (2)

[0215] (3)

[0216] in Indicates the first The first GIS switch status data Hidden status values ​​of GIS switch status information over a time step; This represents an activation function that maps variables to values ​​between 0 and 1. , and , All are weight matrices.

[0217] More specifically, step 2.2 further includes:

[0218] Step 2.2.2: Select information for GIS switch status. Time-frequency decomposition and filtering are performed to obtain each IMF component:

[0219] (4)

[0220] (5)

[0221] (6)

[0222] in For the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status selection information. This refers to the frequency domain information obtained after time-frequency decomposition. For decomposing sequence number and , Angular velocity, This is the output value of this filter. and These are the filter coefficients. and These are two sampled values, The previous sample value, for New information obtained from time-frequency conversion.

[0223] More specifically, step 2.2 further includes:

[0224] Step 2.2.3: Calculate the energy entropy of each component using equations (8)-(9).

[0225] (7)

[0226] (8)

[0227] (9)

[0228] (10)

[0229] in, This represents the energy value of each IMF component. This represents the total energy value of the IMF components. This indicates the proportion of energy in each component to the total energy. This represents the energy entropy value of each IMF component;

[0230] The forgetting part of the time-frequency memory is obtained using equation (11). The first GIS switch status data Output set at each time step :

[0231] (11)

[0232] in, This represents the outer product operation. It is the first The GIS switch status data in the time-frequency memory unit The memory information of the time step.

[0233] More specifically, step 2.3 includes:

[0234] Step 2.3.1: The cell renewal unit uses equations (12)-(14) to obtain the first... The first GIS switch status data GIS switch status input information at each time step and GIS switch status modulation information :

[0235] (12)

[0236] (13)

[0237] (14)

[0238] in, For the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status information awaiting update; and Both are input weight matrices. Represents the input bias vector; and All are modulation weight matrices. Represents the modulation deviation vector; This represents element-wise multiplication. For activation functions;

[0239] Step 2.3.2, regarding the first... The GIS switch status data in the time-frequency memory unit Time step GIS switch status information to be updated Perform time-frequency decomposition and filtering to obtain the first... The first GIS switch status data Each IMF component of the GIS switch status information output at each time step ;

[0240] Step 2.3.3: Calculate each IMF component. Energy entropy .

[0241] More specifically, step 2.4 includes:

[0242] Step 2.4.1: The output gate of the cell renewal unit is used to obtain the first... The first GIS switch status data Memory state at each time step :

[0243] (twenty two)

[0244] The output gate obtains the first... using equation (23). The first GIS switch status data A composite signal with a time step :

[0245] (twenty three)

[0246] in, and Both are weight matrices. This represents the bias vector of the output gate;

[0247] Based on equations (22) and (23), we can obtain the first equation using equations (24) and (25). GIS switch status information for all time steps; hidden status. :

[0248] (twenty four)

[0249] (25)

[0250] Step 2.4.2, take the first... GIS switch status information for all time steps; hidden status. The mean value is used as the feature vector of GIS switch status. .

[0251] More specifically, step 2.5 includes:

[0252] Step 2.5.1, the GIS switch state feature vector Input into the fully connected layer, and use equation (26) to obtain the first... GIS switch status data with tags The probability value corresponding to all GIS switch state types N :

[0253] (26)

[0254] in, and These represent the weighted average GIS switch state feature vectors, respectively. The weight matrix and bias vector;

[0255] Step 2.5.2: Construct the cross-entropy loss function using equation (27). :

[0256] (27)

[0257] in, Indicates the first The first GIS switch status data The probability of predicting a GIS switch status label at each time step. Indicates the first The first GIS switch status data The true label value for each time step.

[0258] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing the condition of GIS equipment based on IMF energy entropy, characterized in that, The method includes: Step 1: Obtain GIS switch status data and their corresponding status label values, and construct a training set; Step 2: Construct a time-frequency memory recurrent neural network based on IMF energy entropy. Input GIS switch state data into the neural network to obtain the hidden state of GIS switch state information for all time steps, and use the mean of these hidden states as the GIS switch state feature vector. Use the GIS switch state feature vector to classify GIS switch states, and design a cross-entropy loss function considering the classification results and the true state label values. Step 2.1: Construct a time-frequency memory recurrent neural network based on IMF energy entropy. This neural network includes a time-frequency memory recurrent layer and a fully connected layer. The time-frequency memory recurrent layer includes a time-frequency memory unit, a time-frequency memory forgetting part, and a cell renewal unit. Step 2.2: Input the GIS switch status data into the time-frequency memory recursive layer to obtain the first... The GIS switch status data in the time-frequency memory unit GIS switch status selection information at time steps The time-frequency decomposition and filtering are performed to obtain each IMF component, and the energy entropy of each IMF component is calculated; the energy entropy is input into the time-frequency memory-forgetting part to obtain the first IMF component. The first GIS switch status data Output set at each time step ; Step 2.3, the The first GIS switch status data Data at each time step and the Hidden status values ​​of GIS switch status data at each time step The input to the cell renewal unit yields the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status information to be updated The time-frequency decomposition and filtering are performed to obtain each IMF component, and the energy entropy of each IMF component is calculated. ; Step 2.4, the cell renewal unit will... The first GIS switch status data Output set at each time step With energy entropy Adding them together gives the result of the first addition. The first GIS switch status data Memory state at each time step After processing, the first output is... GIS switch status information for all time steps, including hidden status. For hidden state Take the mean value as the GIS switch status feature vector ; Step 2.5, GIS switch status feature vector In the fully connected input layer, the cross-entropy loss function is designed considering both the classification result and the true state label value. Step 3: Train the neural network. Stop training when the maximum number of training rounds is reached or the value of the cross-entropy loss function reaches its minimum, thus obtaining a trained neural network. Step 4: Input the real-time collected GIS switch status data into the trained neural network to obtain the GIS equipment status assessment results.

2. The GIS equipment status assessment method based on IMF energy entropy according to claim 1, characterized in that, Step one includes: Construct datasets of grounding current, surge arrester leakage current, and enclosure vibration for GIS switch maintenance, cold standby, and hot standby states. ,in, This represents a dataset of ground currents. This represents the surge arrester leakage current dataset. This represents a dataset of vibration data for GIS equipment casings.

3. The GIS equipment status assessment method based on IMF energy entropy according to claim 1, characterized in that, Step 2.2 includes: Step 2.2.1, the first The first GIS switch status data Data at each time step In the input time-frequency memory recursive layer, obtain the first... The GIS switch status data in the time-frequency memory unit GIS switch status selection information at time steps : （1） （2） （3） in Indicates the first The first GIS switch status data Hidden status values ​​of GIS switch status information over a time step; This represents an activation function that maps variables to values ​​between 0 and 1. , and , All are weight matrices.

4. The GIS equipment status assessment method based on IMF energy entropy according to claim 3, characterized in that, Step 2.2 further includes: Step 2.2.2: Select information for GIS switch status. Time-frequency decomposition and filtering are performed to obtain each IMF component: （4） （5） （6） in For the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status selection information. This refers to the frequency domain information obtained after time-frequency decomposition. For decomposing sequence number and , Angular velocity, This is the output value of this filter. and These are the filter coefficients. and These are two sampled values, The previous sample value, for New information obtained from time-frequency conversion.

5. The GIS equipment status assessment method based on IMF energy entropy according to claim 4, characterized in that, Step 2.2 further includes: Step 2.2.3: Calculate the energy entropy of each component using equations (8)-(9). （7） （8） （9） （10） in, This represents the energy value of each IMF component. This represents the total energy value of the IMF components. This indicates the proportion of energy in each component to the total energy. This represents the energy entropy value of each IMF component; The forgetting part of the time-frequency memory is obtained using equation (11). The first GIS switch status data Output set at each time step : （11） in, This represents the outer product operation. It is the first The GIS switch status data in the time-frequency memory unit The memory information of the time step.

6. The GIS equipment status assessment method based on IMF energy entropy according to claim 5, characterized in that, Step 2.3 includes: Step 2.3.1: The cell renewal unit uses equations (12)-(14) to obtain the first... The first GIS switch status data GIS switch status input information at each time step and GIS switch status modulation information : （12） （13） （14） in, For the first The GIS switch status data in the time-frequency memory unit Time step GIS switch status information awaiting update; and Both are input weight matrices. Represents the input bias vector; and All are modulation weight matrices. Represents the modulation deviation vector; This represents element-wise multiplication. For activation functions; Step 2.3.2, regarding the first... The GIS switch status data in the time-frequency memory unit Time step GIS switch status information to be updated Perform time-frequency decomposition and filtering to obtain the first... The first GIS switch status data Each IMF component of the GIS switch status information output at each time step ; Step 2.3.3: Calculate each IMF component. Energy entropy .

7. The GIS equipment status assessment method based on IMF energy entropy according to claim 6, characterized in that, Step 2.4 includes: Step 2.4.1: The output gate of the cell renewal unit is used to obtain the first... The first GIS switch status data Memory state at each time step : (22) The output gate obtains the first... using equation (23). The first GIS switch status data A composite signal with a time step : （23） in, and Both are weight matrices. This represents the bias vector of the output gate; Based on equations (22) and (23), we can obtain the first equation using equations (24) and (25). GIS switch status information for all time steps; hidden status. : （24） （25） Step 2.4.2, take the first... GIS switch status information for all time steps; hidden status. The mean value is used as the feature vector of GIS switch status. .

8. The GIS equipment status assessment method based on IMF energy entropy according to claim 7, characterized in that, Step 2.5 includes: Step 2.5.1, the GIS switch state feature vector Input into the fully connected layer, and use equation (26) to obtain the first... GIS switch status data with tags The probability value corresponding to all GIS switch state types N : （26） in, and These represent the weighted average GIS switch state feature vectors, respectively. The weight matrix and bias vector; Step 2.5.2: Construct the cross-entropy loss function using equation (27). : （27） in, Indicates the first The first GIS switch status data The probability of predicting a GIS switch status label at each time step. Indicates the first The first GIS switch status data The true label value for each time step.

9. An apparatus for implementing the GIS equipment status assessment method based on IMF energy entropy as described in any one of claims 1-8, characterized in that, The device includes: The training set construction module is used to obtain GIS switch status data and their corresponding status label values ​​to construct a training set. The neural network construction module is used to construct a time-frequency memory recurrent neural network based on IMF energy entropy. GIS switch state data is input into the neural network to obtain the hidden state of GIS switch state information at all time steps, and the mean of these hidden states is used as the GIS switch state feature vector. The GIS switch state feature vector is used to classify GIS switch states. The cross-entropy loss function is designed considering the classification results and the true state label values. The neural network training module is used to train the neural network. When the maximum number of training rounds is reached, or the value of the cross-entropy loss function reaches its minimum, the training stops, thus obtaining a trained neural network. The analysis result output module is used to input the real-time collected GIS switch status data into the trained neural network to obtain the GIS equipment status analysis results.

Images