Substation cable trench small animal invasion identification method

Abstract

The invention provides a transformer substation cable trench small animal invasion identification method, which comprises the following steps of: installing a multi-gas sensing module in a cable trench to detect odor and generate gas data, uploading the gas data to a data processing terminal by using a data transmission module, receiving the gas data by the data processing terminal to perform preprocessing and feature extraction, and transmitting the gas data to a server; and then whether small animals exist in the cable trench is judged through the intelligent identification system according to the processed gas data, and a corresponding identification result is output. According to the intelligent identification system, a deep learning algorithm is adopted, network training is carried out through a large amount of sample data, and the identification accuracy and stability are improved. Meanwhile, operation and maintenance personnel can be reminded to take corresponding protection measures according to the identification result, intelligent safety protection of the cable trench is achieved, and the safety of the cable trench of the transformer substation can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of safety protection, and in particular to a method for identifying small animals intruding into a cable trench of a substation. Background Art

[0002] Small animals in substations have always been one of the key safety hazards in the station. Small animals can cause cable damage, switch tripping, and even enter the switch cabinet, causing serious consequences such as switch cabinet explosion and fire. In the field of cable trench safety protection, traditional methods of preventing small animals mainly rely on physical barriers, such as installing protective nets and using sealing materials to block them. However, these methods have many limitations. First, physical barriers may not be able to completely prevent small animals from entering the cable trench through various gaps or holes. Secondly, after long-term use, the protective net or sealing material may lose its protective effect due to aging and damage, and requires regular maintenance and replacement, which increases maintenance costs and workload. In addition, the physical barrier method cannot detect and deal with small animals that have entered the cable trench in a timely manner, and there is a long inspection dead zone, which poses certain safety hazards.

[0003] With the development of technology, intelligent safety protection has gradually become a trend. At present, although there are some devices on the market that use sensor technology for safety monitoring, intelligent identification devices for cable trench protection against small animals are still relatively rare. Small animals such as mice will produce gas substances (such as hydrogen sulfide, ammonia, methane, carbon dioxide, etc.) after decomposition in the cable trench, which has its own unique gas "atlas" information.

[0004] Therefore, the present invention studies a method for identifying small animals intruding into a substation cable trench, which has important practical application value and market demand. By using a multi-gas module to detect the smell of small animals, and combining network training for intelligent identification, it is intended to solve the limitations of traditional methods for preventing small animals. At the same time, during the training process, one is to consider the smell of small animals rotting in the cable trench, and the other is to consider the gases (hydrogen sulfide, mercaptan, sulfide, ammonia, methane, carbon dioxide, etc.) produced by the smell of corruption caused by small animals bringing food to the cable trench for comprehensive consideration. By real-time monitoring of gas data in the cable trench, the present invention can accurately determine whether there are small animals in the cable trench, and automatically trigger an alarm or take corresponding protective measures to improve the safety protection level of the cable trench. Summary of the invention

[0005] The purpose of the present invention is to propose a method for identifying the intrusion of small animals in the cable trench of a substation. By using a gas sensor, different odor detection can be achieved and it can be determined whether there are small animals in the cable trench. An alarm can be automatically triggered or corresponding protective measures can be taken. This can improve the safety of the substation and has strong practicality.

[0006] To achieve the above-mentioned purpose, the technical solution of the present invention is: a method for identifying the intrusion of small animals in a cable trench of a substation, wherein a multi-gas sensor module is installed in the cable trench to detect odors and generate gas data, and the gas data is uploaded to a data processing terminal using a data transmission module. The data processing terminal receives the gas data for preprocessing and feature extraction, and then uses an intelligent identification system to determine whether there are small animals in the cable trench based on the processed gas data, and outputs the corresponding identification results.

[0007] Preferably, the multi-gas sensing module includes 16 groups of gas sensors, and the 16 groups of gas sensors are made of different semiconductor materials, including: tin oxide, zinc oxide, titanium dioxide, tungsten trioxide, indium oxide, iron oxide, chromium trioxide, magnesium oxide, iridium dioxide, cobalt oxide, nickel oxide, palladium oxide, molybdenum oxide, vanadium oxide, niobium oxide and antimony oxide. After reaching the operating temperature, the semiconductor material can adsorb gas molecules or react chemically with them, resulting in changes in conductivity, volt-ampere characteristics or surface potential, so as to distinguish between gases such as nitric oxide, nitrogen, hydrogen sulfide, thiol, sulfide, ammonia, methane, carbon dioxide, etc.; for example, SnO2 is a typical n-type semiconductor material. When it adsorbs oxygen molecules and accepts electrons, it will form O2- or O-, and then in the presence of reducing gases (such as H2, CO, CH4), the surface resistance of the semiconductor will decrease and the conductivity will increase.

[0008] Preferably, the preprocessing is to denoise the gas data using a Kalman filter algorithm.

[0009] Preferably, the feature extraction is to reduce the dimension of the denoised gas data using PCA principal component analysis.

[0010] Preferably, the intelligent identification system determines whether there are small animals in the cable trench through a fully connected convolutional neural network model FCNN, and the fully connected convolutional neural network model FCNN is trained using an established sample set.

[0011] Preferably, the sample set is established by: using a multi-gas sensor module to collect multiple groups of gas data on the decomposition of small animal carcasses in closed containers in different time periods, using a Kalman filter algorithm to denoise the gas data to obtain clean samples, using PCA principal component analysis to reduce the dimension of the denoised gas data, and using the reduced-dimensional sample data as the sample set.

[0012] Preferably, the denoising of gas data by using the Kalman filter algorithm is specifically as follows:

[0013] Establish a state transition model:

[0014] s t =As t-1 +w t-1

[0015] Among them, s t and t-1 They represent the state vectors of the system at time t and t-1 respectively, A is the state transfer matrix, and w t-1 is the process noise, whose covariance is Q;

[0016] Build the observation model:

[0017] x t =Hs t +n t

[0018] Where H is the observation matrix, n t is the observation noise, whose covariance is R;

[0019] Through the prediction equation:

[0020]

[0021] P t|t-1 =AP t-1|t-1 A T +Q

[0022] in, It represents the state vector predicted at time t by time t-1, P t|t-1 is the prediction error covariance matrix at time t, P t-1|t-1 is the prediction error covariance matrix calculated at time t-1;

[0023] And update it with the following formula:

[0024] Calculate the Kalman gain: K t =P t|t-1 H T (HP t|t-1 H T +R) -1 , where H is the observation matrix, mapping the system state to the observation space, and R is the observation noise covariance matrix;

[0025] Using the observed value x t To update the state estimate: The difference between the observed and predicted values ​​is weighted by the Kalman gain to adjust the state estimate;

[0026] Update the error covariance matrix P t|t =(IK t H)P t|t-1 , where K t is the Kalman gain, R is the observation covariance, I is the identity matrix, Indicates that the state obtained at this moment is obtained by estimating the potential clean signal from the noisy observation data X:

[0027]

[0028] Preferably, the specific steps of using PCA principal component analysis to reduce the dimension of the denoised gas data are:

[0029] Step S1: Data centering; Calculate the sample mean vector for a dataset S with m samples and n features

[0030]

[0031] Where S = [s_1, s_2, ..., s_m], each sample s_i is an n-dimensional vector, n represents the number of gas sensor channels; then, the data set is centered, and the corresponding mean is subtracted from each feature to obtain the centralized data set S centered :

[0032]

[0033] Step S2: Calculate the covariance matrix: Based on the centralized data set, calculate the covariance matrix:

[0034]

[0035] The covariance matrix is ​​a symmetric matrix, the elements on the diagonal are the variance of each feature, and the elements on the off-diagonal are the covariances between different features; the covariance matrix Σ is subjected to eigenvalue decomposition to obtain eigenvalues ​​λ_1,λ_2,...,λ_n and corresponding eigenvectors v_1,v_2,...,v_n; the size of the eigenvalue indicates the degree of data variation in the direction of the corresponding eigenvector, that is, the size of the variance;

[0036] Step S3: Select the principal components; select the eigenvectors corresponding to the first k eigenvalues ​​as the principal components according to the size of the eigenvalues, and these eigenvectors constitute a projection matrix W, in which each column is an eigenvector; the number of principal components k is determined based on the cumulative contribution rate, that is, the proportion of the sum of the variances of the first k principal components to the total variance;

[0037] Step S4: Data projection: The centralized data set S centered Project it into the subspace formed by the first k eigenvectors to obtain the reduced-dimensional data set S reduced , whose dimension is m×k, that is, the coordinate of each sample in the k-dimensional space.

[0038] Preferably, the fully connected convolutional neural network model FCNN includes an input layer, a hidden layer and an output layer, wherein the hidden layer output = σ(W1·x+b1), the output layer output = W2·σ(W1·x+b1)+b2, x represents the input data, W1 and W2 are both weight matrices, connecting the input layer to the hidden layer and the hidden layer to the output layer respectively, b1 and b2 are both bias terms, and σ is an activation function; the output layer outputs the final recognition result, and the recognition result is divided into no small animals, presence of small animals and presence of food decomposition.

[0039] Preferably, the method further comprises: automatically triggering an alarm according to the identification result to remind operation and maintenance personnel to take corresponding protective measures.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The present invention uses multiple groups of semiconductor materials of different materials as gas sensors, takes into account the influence of mutual interference between different gases, performs full feature acquisition, homogeneously, and denoises the gas data through the Kalman filter algorithm to effectively improve the accuracy of the data; PCA principal component analysis is used to reduce the dimension of the denoised gas data, thereby reducing the complexity of the calculation; the intelligent identification system uses a deep learning algorithm to perform network training through a large amount of sample data to improve the accuracy and stability of identification. At the same time, the present invention can also automatically trigger an alarm or take corresponding protective measures according to the identification results to achieve intelligent safety protection of the cable trench. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0043] The following is combined with Figure 1 , the technical solution of the present invention is specifically described.

[0044] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0045] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations.

[0046] The present invention proposes a method for identifying the intrusion of small animals in a cable trench of a substation. A multi-gas sensor module is installed in the cable trench to detect odors and generate gas data. The gas data is uploaded to a data processing terminal using a data transmission module. The data processing terminal receives the gas data for preprocessing and feature extraction. Then, an intelligent identification system is used to determine whether there are small animals in the cable trench based on the processed gas data, and output corresponding identification results.

[0047] In this embodiment, the multi-gas sensing module includes 16 groups of gas sensors, and the 16 groups of gas sensors are made of different semiconductor materials, including: tin oxide, zinc oxide, titanium dioxide, tungsten trioxide, indium oxide, iron oxide, chromium trioxide, magnesium oxide, iridium dioxide, cobalt oxide, nickel oxide, palladium oxide, molybdenum oxide, vanadium oxide, niobium oxide and antimony oxide. After reaching the operating temperature, the semiconductor material can adsorb gas molecules or react chemically with them, resulting in changes in conductivity, volt-ampere characteristics or surface potential to distinguish gases such as nitric oxide, nitrogen, hydrogen sulfide, mercaptan, sulfide, ammonia, methane, carbon dioxide, etc.; for example, SnO2 is a typical n-type semiconductor material. When it adsorbs oxygen molecules and accepts electrons, it will form O2- or O-, and then in the presence of reducing gases (such as H2, CO, CH4), the semiconductor surface resistance decreases and the conductivity increases.

[0048] The analysis method introduced in the present invention adopts 16 groups of gas sensors as the dimensions of the odor map, and utilizes the characteristic that a single gas sensor will produce differentiated response signals to different gases. According to the gas substances (such as nitric oxide, nitrogen, hydrogen sulfide, thiol, sulfide, ammonia, methane, carbon dioxide, etc.) produced after the decomposition of small animal corpses, each gas sensor becomes a "dimension" of odor detection.

[0049] In this embodiment, the preprocessing is to use a Kalman filter algorithm to denoise the gas data.

[0050] In this embodiment, the feature extraction is to use PCA principal component analysis to reduce the dimension of the denoised gas data.

[0051] In this embodiment, the intelligent identification system determines whether there are small animals in the cable trench through a fully connected convolutional neural network model FCNN, and the fully connected convolutional neural network model FCNN is trained using an established sample set.

[0052] In this embodiment, the sample set is established specifically as follows: using a multi-gas sensor module to collect multiple groups of gas data on the decomposition of small animal carcasses in a closed container in different time periods, using a Kalman filter algorithm to denoise the gas data to obtain clean samples, using PCA principal component analysis to reduce the dimension of the denoised gas data, and using the reduced-dimensional sample data as the sample set.

[0053] The present invention uses multiple groups of small animal carcasses decomposed in sealed containers in different time periods as training samples, and generates different responses in different gas sensors to obtain a multi-dimensional space of odor data, thereby realizing the "identity" detection of different odors. Through the sensor's response curve to the odor, considering the sensor's noise disturbance, a set of multi-dimensional time series sample sets are formed for the response of the gas sensor. In order to improve the accuracy of the sample, Kalman filtering is performed to denoise, and the clean sample is subjected to PCA principal component analysis to obtain dimensionality-reduced sample data, and combined with neural network training to obtain judgment.

[0054] In this embodiment, the Kalman filter algorithm is used to denoise the gas data as follows:

[0055] Establish a state transition model:

[0056] s t =As t-1 +w t-1

[0057] Among them, s t and t-1 They represent the state vectors of the system at time t and t-1 respectively, A is the state transfer matrix, and w t-1 is the process noise, whose covariance is Q;

[0058] Build the observation model:

[0059] x t =Hs t +n t

[0060] Where H is the observation matrix, n t is the observation noise, whose covariance is R;

[0061] Through the prediction equation:

[0062]

[0063] P t|t-1 =AP t-1|t-1 A T +Q

[0064] in, It represents the state vector predicted at time t by time t-1, P t|t-1 is the prediction error covariance matrix at time t, P t-1|t-1 is the prediction error covariance matrix calculated at time t-1;

[0065] And update it with the following formula:

[0066] Calculate the Kalman gain: K t=P t|t-1 H T (HP t|t-1 H T +R) -1 , where H is the observation matrix, mapping the system state to the observation space, and R is the observation noise covariance matrix;

[0067] Using the observed value x t To update the state estimate: The difference between the observed and predicted values ​​is weighted by the Kalman gain to adjust the state estimate;

[0068] Update the error covariance matrix P t|t =(IK t H)P t|t-1 , where K t is the Kalman gain, R is the observation covariance, I is the identity matrix, Indicates that the state obtained at this moment is obtained by estimating the potential clean signal from the noisy observation data X:

[0069]

[0070] In this embodiment, the specific steps of using PCA principal component analysis to reduce the dimension of the denoised gas data are as follows:

[0071] Step S1: Data centering: For a data set S with m samples and n features (the number of gas sensor channels), calculate the mean vector of the samples

[0072]

[0073] Among them, S = [s_1, s_2, ..., s_m], each sample s_i is an n-dimensional vector; then, the data set is centered, and the corresponding mean is subtracted from each feature to obtain the centralized data set S centered :

[0074]

[0075] Step S2: Calculate the covariance matrix: Based on the centralized data set, calculate the covariance matrix:

[0076]

[0077] The covariance matrix is ​​a symmetric matrix, the elements on the diagonal are the variance of each feature, and the elements on the off-diagonal are the covariances between different features;

[0078] Perform eigenvalue decomposition on the covariance matrix Σ to obtain eigenvalues ​​λ_1,λ_2,...,λ_n and corresponding eigenvectors v_1,v_2,...,v_n; the size of the eigenvalue indicates the degree of data variation in the direction of the corresponding eigenvector, that is, the size of the variance;

[0079] Step S3: selecting principal components; selecting the eigenvectors corresponding to the first k eigenvalues ​​as principal components according to the size of the eigenvalues, and these eigenvectors constitute a projection matrix W, in which each column is an eigenvector; the number of principal components k is determined based on the cumulative contribution rate (i.e., the ratio of the sum of the variances of the first k principal components to the total variance);

[0080] Step S4: Data projection: The centralized data set S centered Project it into the subspace formed by the first k eigenvectors to obtain the reduced-dimensional data set S reduced , S reduced is the reduced-dimensional dataset, whose dimension is m×k, i.e., the coordinates of each sample in the k-dimensional space;

[0081] The specific explanations are as follows:

[0082] S centered :This is a centered matrix whose columns are subtracted from the mean of the column so that the mean of each column is 0. Assume S centered is an m×n matrix, where m is the number of observations and n is the number of variables.

[0083] W: This is a transformation matrix whose dimensions depend on the dimensions to which we want to reduce the data. In PCA, W usually consists of the eigenvectors of the data covariance matrix, which correspond to the largest eigenvalues. In more general linear transformations, W can be any matrix that meets our needs. Assume W is an n×k matrix, where k≤n is the dimension to which we want to reduce the data.

[0084] S reduced : This is the data matrix after W transformation, with dimension m × k. Each row represents the coordinates of an observation in the transformed space.

[0085] Therefore, the principal component analysis data set S reduced .

[0086] In this embodiment, the fully connected convolutional neural network model FCNN includes an input layer, a hidden layer and an output layer. reduced A fully connected convolutional neural network (FCNN) is used for training by passing the data as input to the input layer for forward propagation.

[0087] Hidden layer output = σ(W1 x+b1),

[0088] Output layer output = W2·σ(W1·x+b1)+b2,

[0089] Among them, x represents the input data, W1 and W2 are weight matrices, connecting the input layer to the hidden layer and the hidden layer to the output layer respectively, b1 and b2 are bias terms, and σ is the activation function.

[0090] The weighted summation is performed through each hidden layer with the corresponding input. Then, a bias term is added to the result of the weighted summation, and the output of the neuron is obtained through the activation function. This process is repeated in each layer until the output layer is reached.

[0091] The final prediction result is generated: In the output layer, the final classification is calculated based on the weights, which belongs to the smell of rotting small animals or the smell of rotting food, and the final prediction result is generated.

[0092] In this embodiment, the method further includes: automatically triggering an alarm according to the identification result to remind the operation and maintenance personnel to take corresponding protective measures

[0093] The above are preferred embodiments of the present invention. Any changes made according to the technical solution of the present invention, as long as the resulting functions do not exceed the scope of the technical solution of the present invention, belong to the protection scope of the present invention.

Claims

1. A method for identifying small animals intruding into a substation cable trench, characterized in that: A multi-gas sensor module is installed in the cable trench to detect odors and generate gas data. The gas data is uploaded to the data processing terminal using the data transmission module. The data processing terminal receives the gas data for preprocessing and feature extraction. The intelligent identification system then determines whether there are small animals in the cable trench based on the processed gas data and outputs the corresponding identification results.

2. The method for identifying small animals intruding into a substation cable trench according to claim 1 is characterized in that: The multi-gas sensing module includes 16 groups of gas sensors, which are made of different semiconductor materials, including: tin oxide, zinc oxide, titanium dioxide, tungsten trioxide, indium oxide, iron oxide, chromium trioxide, magnesium oxide, iridium dioxide, cobalt oxide, nickel oxide, palladium oxide, molybdenum oxide, vanadium oxide, niobium oxide and antimony oxide. After reaching the operating temperature, the semiconductor material can adsorb gas molecules or react chemically with them, resulting in changes in conductivity, volt-ampere characteristics or surface potential.

3. The method for identifying small animals intruding into a substation cable trench according to claim 1 is characterized in that: The preprocessing is to use Kalman filter algorithm to remove noise from gas data.

4. The method for identifying small animals intruding into a substation cable trench according to claim 3 is characterized in that: The feature extraction is to use PCA principal component analysis to reduce the dimension of the denoised gas data.

5. The method for identifying small animals intruding into a substation cable trench according to claim 1, characterized in that: The intelligent identification system determines whether there are small animals in the cable trench through a fully connected convolutional neural network model FCNN, and the fully connected convolutional neural network model FCNN is trained using an established sample set.

6. The method for identifying small animals intruding into a substation cable trench according to claim 4 is characterized in that: The sample set is established specifically as follows: using a multi-gas sensor module to collect multiple sets of gas data on the decomposition of small animal carcasses in closed containers in different time periods, using a Kalman filter algorithm to denoise the gas data to obtain clean samples, using PCA principal component analysis to reduce the dimension of the denoised gas data, and using the reduced-dimensional sample data as the sample set.

7. The method for identifying small animals intruding into a substation cable trench according to claim 4 or 6, characterized in that: The Kalman filter algorithm is used to denoise the gas data as follows: Establish a state transition model: s t =As t-1 +w t-1 Among them, s t and t-1 They represent the state vectors of the system at time t and t-1 respectively, A is the state transfer matrix, and w t-1 is the process noise, whose covariance is Q; Build the observation model: x t =Hs t +n t Where H is the observation matrix, n t is the observation noise, whose covariance is R; Through the prediction equation: P t|t-1 =AP t-1|t-1 From T +Q in, It represents the state vector predicted at time t by time t-1, P t|t-1 is the prediction error covariance matrix at time t, P t-1|t-1 is the prediction error covariance matrix calculated at time t-1; And update it with the following formula: Calculate the Kalman gain: K t =P t|t-1 H T (HP t|t-1 H T +R) -1 , where H is the observation matrix, mapping the system state to the observation space, and R is the observation noise covariance matrix; Using the observed value x t To update the state estimate: The difference between the observed and predicted values ​​is weighted by the Kalman gain to adjust the state estimate; Update the error covariance matrix P t|t =(IK t H)P t|t-1 , where K t is the Kalman gain, R is the observation covariance, I is the identity matrix, Indicates that the state obtained at this moment is obtained by estimating the potential clean signal from the noisy observation data X:

8. The method for identifying small animals intruding into a substation cable trench according to claim 7, characterized in that: The specific steps of using PCA principal component analysis to reduce the dimension of the denoised gas data are as follows: Step S1: Data centering, calculate the mean vector of the sample for the dataset S of m samples and n features after the Kalman filter Where S = [s_1, s_2, ..., s_m], each sample s_i is an n-dimensional vector, n represents the number of gas sensor channels; then, the data set is centered, and the corresponding mean is subtracted from each feature to obtain the centralized data set S centered : Step S2: Calculate the covariance matrix: Based on the centralized data set, calculate the covariance matrix: The covariance matrix is ​​a symmetric matrix, the elements on the diagonal are the variance of each feature, and the elements on the off-diagonal are the covariances between different features; the covariance matrix Σ is subjected to eigenvalue decomposition to obtain eigenvalues ​​λ_1,λ_2,...,λ_n and corresponding eigenvectors v_1,v_2,...,v_n; the size of the eigenvalue indicates the degree of data variation in the direction of the corresponding eigenvector, that is, the size of the variance; Step S3: Select the principal components; select the eigenvectors corresponding to the first k eigenvalues ​​as the principal components according to the size of the eigenvalues, and these eigenvectors constitute a projection matrix W, in which each column is an eigenvector; the number of principal components k is determined based on the cumulative contribution rate, that is, the proportion of the sum of the variances of the first k principal components to the total variance; Step S4: Data projection: The centralized data set S centered Project it into the subspace formed by the first k eigenvectors to obtain the reduced-dimensional data set S reduced , whose dimension is m×k, that is, the coordinate of each sample in the k-dimensional space.

9. The method for identifying small animals intruding into a substation cable trench according to claim 7, characterized in that: The fully connected convolutional neural network model FCNN includes an input layer, a hidden layer and an output layer, wherein the hidden layer output = σ(W1·x+b1), the output layer output = W2·σ(W1·x+b1)+b2, x represents the input data, W1 and W2 are weight matrices, connecting the input layer to the hidden layer and the hidden layer to the output layer respectively, b1 and b2 are bias terms, and σ is an activation function; the output layer outputs the final recognition result, and the recognition result is divided into no small animals, small animals and food decomposition.

10. The method for identifying small animals intruding into a substation cable trench according to claim 1, characterized in that: The method also includes: automatically triggering an alarm according to the identification result to remind operation and maintenance personnel to take corresponding protective measures.

Images