Convolutional neural network-based power supply quality evaluation method and system

Abstract

The invention discloses a power supply quality evaluation method and system based on a convolutional neural network, and the method comprises the steps: carrying out the convolution processing and nonlinear feature mapping of electric energy data collected from a power supply link in real time according to a preset convolution module and an activation function, and extracting a first feature map of the electric energy data; performing pooling operation and vector conversion on the first feature map to obtain a first feature value; according to a support vector machine optimized by a locust optimization algorithm, classifying and identifying the first characteristic value to obtain a power supply quality evaluation result; and overhauling the power supply link according to the power supply quality evaluation result. According to the method, the convolutional neural network is used for extracting the characteristic values of the electric energy data in the whole process, the locust optimization algorithm is used for improving the efficiency of classifying and identifying the characteristic values, and the accuracy of evaluating the overall power supply quality is improved.

Description

Technical Field

[0001] The present invention belongs to the field of power quality assessment, and specifically relates to a power quality assessment method and system based on convolutional neural network. Background Art

[0002] With the development of modern power systems, in order to reduce carbon emissions and promote the efficient operation of power systems, a variety of renewable energy sources have been increasingly used, and as a result, more power electronic equipment has been put into operation in the power grid. In this context, the risk of power quality disturbances (PQDs) is increasing, and power quality problems usually appear in the form of sudden changes in current, voltage or frequency. In actual situations, many reasons may lead to the generation of PQDs, such as the application of power electronic equipment, lightning, nonlinear loads, etc. Therefore, it is necessary to evaluate the power quality and detect power quality disturbances in a timely manner.

[0003] Existing power quality assessment methods all rely on inspection and partial measurement point data. Signal processing technology is first used to extract the characteristic quantity of the signal, and then a classification method is used to determine the category of the disturbance event. The assessment results obtained in this way only represent the power supply quality of the sampled samples, and cannot represent the overall power supply quality of the power supply area. Therefore, the obtained assessment results are not accurate and cannot represent the overall power supply quality of the power supply area. In addition, traditional signal processing technology generally uses algorithms such as S transform, Hilbert-Huang transform, discrete wavelet transform, short-time Fourier transform, fast Fourier transform, and discrete Fourier transform to extract characteristic values ​​from the signal waveform that can accurately express the signal difference. Recognition and classification is to accurately distinguish the disturbance type based on the extracted characteristic values. However, the above algorithms are prone to partial feature loss, which will affect the accuracy of classification. Summary of the invention

[0004] The present invention proposes a power supply quality assessment method and system based on convolutional neural network, which uses convolutional neural network to extract characteristic values ​​of electric energy data of the whole process, and uses locust optimization algorithm to improve the efficiency of classification and identification of characteristic values, and improves the accuracy of assessing the overall power supply quality.

[0005] A first aspect of the present invention provides a power supply quality assessment method based on a convolutional neural network, the method comprising:

[0006] According to a preset convolution module and activation function, convolution processing and nonlinear feature mapping are performed on the electric energy data collected in real time from the power supply link to extract a first feature map of the electric energy data;

[0007] Performing a pooling operation and a vector conversion on the first feature map to obtain a first eigenvalue;

[0008] Classifying and identifying the first eigenvalues ​​according to a support vector machine optimized by a locust optimization algorithm to obtain a power supply quality assessment result;

[0009] According to the power supply quality assessment results, the power supply link is inspected and repaired.

[0010] The above scheme first convolves the electric energy data collected in real time from the power supply link to extract the feature map of the electric energy data, and then performs nonlinear mapping on the feature map to obtain the first feature map, thereby improving the data expression capability of the feature map; then performs a pooling operation on the feature map to reduce the dimension of the first feature map, shorten the data processing time, and then converts the first feature map into a vector to facilitate subsequent classification calculations; then the first feature map is input into a support vector machine optimized by the locust optimization algorithm, and the first eigenvalue is classified and identified by using the obtained optimal feature spacing to obtain an accurate power supply quality assessment result, while reducing the amount of calculation and improving the efficiency of classifying and identifying the eigenvalues.

[0011] In a possible implementation method of the first aspect, according to a preset convolution module and activation function, convolution processing and nonlinear feature mapping are performed on the electric energy data collected in real time from the power supply link to extract a first feature map of the electric energy data, specifically:

[0012] Inputting the electric energy data into a convolution module to extract local area information of the electric energy data;

[0013] Extracting local features of the electric energy data according to the local area information and a plurality of convolution kernels; wherein each convolution kernel includes a corresponding weight coefficient and a deviation;

[0014] The local features of the electric energy data are input into a preset activation function for nonlinear feature mapping to obtain the first feature map.

[0015] The above scheme first extracts the local area information of the electric energy data, which is used to extract the local area features of the electric energy data in combination with the convolution kernel to obtain the overall features of the electric energy data; then the local features of the electric energy data are subjected to nonlinear feature mapping, so that the obtained first feature map can participate in subsequent nonlinear operation calculations, providing data support for subsequent calculations.

[0016] In a possible implementation method of the first aspect, before performing convolution processing and nonlinear feature mapping on the electric energy data collected in real time from the power supply link, the electric energy data is first preprocessed, and then the preprocessed electric energy data is converted into a data format required by the convolutional neural network; wherein the preprocessing includes de-averaging, normalization and dimensionality reduction.

[0017] In a possible implementation method of the first aspect, performing a pooling operation and a vector conversion on the first feature map to obtain a first eigenvalue is specifically:

[0018] If the dimension of the first feature map is not a multiple of the first threshold, the first feature map is processed using an edge zero padding method so that the first feature map satisfies a pooling condition;

[0019] Using a preset pooling window and step size, sliding on the first feature map that meets the pooling condition to perform a pooling operation to obtain a second feature map;

[0020] Perform vector transformation on the second eigenmap to obtain the first eigenvalue.

[0021] The above scheme reduces the dimension and data volume of the first feature map through pooling operation, thereby effectively reducing the overall amount of data calculation, and uses the edge padding method to process the first feature map to facilitate the pooling operation of the data; then the data is converted into vector output to provide data support for subsequent calculations.

[0022] In a possible implementation method of the first aspect, if the dimension of the first feature map is not a multiple of the first threshold, the first feature map is processed using an edge zero padding method, specifically:

[0023] Calculate the difference between the dimension of the first feature map and the multiple of the next first threshold value to obtain the number of zero padding;

[0024] Add zero padding number zeros to the edge of the first feature map to obtain the first feature map that meets the pooling condition.

[0025] The above scheme processes the first feature map by using the edge zero padding method to fill the edge size in the first feature map that does not meet the pooling condition, so that the first feature map can perform subsequent pooling operations.

[0026] In a possible implementation method of the first aspect, the first eigenvalue is classified and identified according to a support vector machine optimized by a locust optimization algorithm to obtain a power supply quality assessment result, specifically:

[0027] Taking the maximum characteristic distance of the first eigenvalue as the target, using the locust optimization algorithm to calculate the first eigenvalue to obtain the optimal characteristic distance;

[0028] According to the optimal feature spacing, the penalty factor and kernel function of the support vector machine are optimized;

[0029] The first eigenvalue is input into the optimized support vector machine, and the first eigenvalue is classified and identified according to the optimal feature distance to obtain a power supply quality assessment result.

[0030] The above scheme uses the locust optimization algorithm to calculate the optimal feature spacing that allows the data to be correctly classified, and uses the optimal feature spacing to optimize the penalty factor and kernel function of the support vector machine, thereby improving the efficiency of the support vector machine in classifying data, avoiding overfitting of the calculation results, and improving the accuracy of data classification.

[0031] A second aspect of the present invention provides a power supply quality assessment system based on a convolutional neural network, the system comprising: a feature extraction module, a data conversion module, a data classification module and a power supply link maintenance module;

[0032] The feature extraction module is used to perform convolution processing and nonlinear feature mapping on the electric energy data collected in real time from the power supply link according to a preset convolution module and activation function, and extract a first feature map of the electric energy data;

[0033] The data conversion module is used to perform pooling operation and vector conversion on the first feature map to obtain a first eigenvalue;

[0034] The data classification module is used to classify and identify the first eigenvalue according to the support vector machine optimized by the locust optimization algorithm to obtain a power supply quality assessment result;

[0035] The power supply link maintenance module is used to inspect and repair the power supply link according to the power supply quality evaluation result.

[0036] In a possible implementation of the second aspect, the feature extraction module includes: a data convolution and mapping unit;

[0037] The data convolution and mapping unit is used to input the electric energy data into a convolution module to extract local area information of the electric energy data; extract local features of the electric energy data based on the local area information and a plurality of convolution kernels; wherein each convolution kernel includes a corresponding weight coefficient and a deviation; and input the local features of the electric energy data into a preset activation function for nonlinear feature mapping to obtain the first feature map.

[0038] In a possible implementation of the second aspect, the feature extraction module includes: a data preprocessing unit;

[0039] The data preprocessing unit is used to preprocess the electric energy data collected in real time from the power supply link before performing convolution processing and nonlinear feature mapping on the electric energy data, and then convert the preprocessed electric energy data into the data format required by the convolutional neural network; wherein the preprocessing includes de-averaging, normalization and dimensionality reduction.

[0040] In a possible implementation of the second aspect, the data conversion module includes: a data pooling unit;

[0041] The data pooling unit is used to process the first feature map using an edge zero padding method if the dimension of the first feature map is not a multiple of a first threshold value so that the first feature map meets the pooling condition; use a preset pooling window and step size to slide on the first feature map that meets the pooling condition to perform a pooling operation to obtain a second feature map; and perform vector conversion on the second feature map to obtain a first eigenvalue.

[0042] In a possible implementation of the second aspect, the data conversion module includes: a data edge zero padding unit;

[0043] The data edge zero-padding unit is used to calculate the difference between the dimension of the first feature map and the multiple of the next first threshold to obtain the number of zero padding; add the number of zero padding zeros to the edge of the first feature map to obtain the first feature map that meets the pooling condition.

[0044] In a possible implementation of the second aspect, the data classification module includes: a data classification unit;

[0045] The data classification unit is used to calculate the first eigenvalue using the locust optimization algorithm with the maximum feature distance of the first eigenvalue as the target, so as to obtain the optimal feature distance; optimize the penalty factor and kernel function of the support vector machine according to the optimal feature distance; input the first eigenvalue into the optimized support vector machine, classify and identify the first eigenvalue according to the optimal feature distance, and obtain the power supply quality assessment result. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the implementation mode will be briefly introduced below. Obviously, the drawings described below are only some implementation modes of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 A specific flow chart of a power supply quality assessment method based on a convolutional neural network is provided in a certain embodiment of the present invention;

[0048] Figure 2 A structural diagram of a power supply quality assessment system based on a convolutional neural network is provided in accordance with a certain embodiment of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] It should be understood that the step numbers used in this article are only for the convenience of description and are not intended to limit the order in which the steps are executed.

[0051] With the rapid development of deep learning theory and practice, the technological innovation of computer vision, natural language processing, and speech recognition in the field of artificial intelligence is in the ascendant. Compared with traditional machine learning methods, deep learning has higher accuracy in image classification and speech recognition, and has better generalization performance. It has good application potential in power quality disturbance identification.

[0052] The convolutional neural network (CNN) is a deep learning method based on convolution calculation, which can learn the features of input data and predict category labels end-to-end. The convolutional neural network adopts three strategies: local receptive field, weight sharing and downsampling. In the case of translation, rotation, scale scaling, etc., the complexity of the network model is reduced. It is widely used in image classification, target recognition, speech recognition, etc. Therefore, CNN (convolutional neural network) can be used for power feature extraction.

[0053] like Figure 1 As shown, Figure 1 A specific flow chart of a power supply quality assessment method based on a convolutional neural network is provided for a certain embodiment of the present invention. The power supply quality assessment method based on a convolutional neural network in this embodiment includes steps S1 to S4, which are described in detail as follows:

[0054] Step S1, performing convolution processing and nonlinear feature mapping on the electric energy data collected in real time from the power supply link according to a preset convolution module and activation function, and extracting a first feature map of the electric energy data;

[0055] In this step, the electric energy data collected from the power supply side and the power consumption side of the power supply link are first obtained, and the electric energy data is processed such as de-averaging, normalization and dimension reduction, and then the electric energy data is converted into the data format required by the convolutional neural network.

[0056] The electric energy data is input into a preset convolution module, and the local areas of the electric energy data are sequentially passed through convolution kernels of specific sizes to calculate the corresponding local area information; wherein, the sizes of all convolution kernels in the same convolution layer are usually the same, so as to perform efficient matrix operations. However, before different convolution layers, the size of the convolution kernel can be different to meet different feature extraction requirements, and each convolution layer can have a different number of convolution kernels, which depends on the design requirements of the convolutional neural network.

[0057] Then, according to the local area information and a plurality of convolution kernels, local features of the electric energy data are extracted; wherein each convolution kernel includes a corresponding weight coefficient and a deviation;

[0058] Finally, the local features of the electric energy data are input into a preset activation function for nonlinear feature mapping, and the linear electric energy data is converted into nonlinear data to obtain the first feature map of the electric energy data, so as to provide a suitable data format for subsequent calculations. Common activation functions include tanh, sigmoid, and relu, etc. Among them, the activation function used in this embodiment is the relu function, which has a fast convergence speed and solves the gradient saturation problem in sigmoid, and is therefore widely used. Relu function has the advantages of fast convergence and simple gradient calculation compared to sigmoid and nonlinear functions of traditional neural networks. In some embodiments, an unsaturated nonlinear activation function is generally used.

[0059] Step S2, performing a pooling operation and a vector conversion on the first feature map to obtain a first eigenvalue;

[0060] In this step, in order to solve the problem of increased data size and dimensionality due to convolution, it is necessary to continue the pooling operation on the first feature map to further extract the features and identification features of the interference data, thereby effectively reducing the complexity of feature extraction and shortening the calculation time.

[0061] First, the first feature map is input into the pooling layer and the dimension of the first feature map is detected. Although the pooling layer retains the depth dimension, in this embodiment, when the unit of the input data of the pooling layer is not an integer multiple of 2, the edge padding method is usually adopted to pad it to a multiple of 2. The main purpose of edge padding is to ensure that the pooling operation can be evenly applied to the entire input data, avoiding the appearance of areas that cannot be covered by pooling at the edge, which reduces the accuracy of the data.

[0062] If the dimension of the first feature map is not a multiple of 2, the first feature map is processed using an edge zero padding method so that the first feature map satisfies a pooling condition;

[0063] The specific process of the edge zero padding method is as follows:

[0064] 1. Determine the number of zero padding. First, you need to calculate the difference between the dimension of the first feature map and the next multiple of 2 as the number of zero padding; for example, if the width of the first feature map is 5, then it is 1 unit away from the next multiple of 2, 6, so you need to pad 0.5 units of zero on both sides of the dimension, but since you can't pad half a unit, you usually choose to pad 1 unit of zero, so that the total width becomes 6;

[0065] 2. Perform zero padding. According to the calculated zero padding amount, add a corresponding number of zeros to the edges of the corresponding dimensions of the first feature map. In some embodiments, zeros are added to the front and back (for one-dimensional data) or four edges (for two-dimensional data) of the first feature map;

[0066] 3. Perform pooling. After zero padding, the size of the first feature map becomes an integer multiple of 2, and the pooling operation can be applied at this time. The pooling operation will slide on the first feature map after zero padding according to the preset pooling window and step size, perform pooling (such as maximum pooling or average pooling) operation, and obtain the second feature map.

[0067] It is worth noting that the amount and method of zero padding may vary depending on the specific deep learning framework and settings. Some frameworks may allow different amounts of zero padding to be set for different dimensions of the input data, while others may only support a unified zero padding method. Therefore, in actual applications, the method of zero padding needs to be determined based on the framework used and specific needs.

[0068] Finally, the second feature map is converted into a vector through the fully connected layer, and multiplication is performed to reduce its dimension. Then, the second feature map in vector form is classified to obtain the first eigenvalue. Among them, the fully connected layer is the last layer of the convolutional neural network and has the same connection method as the traditional neural network.

[0069] Among them, the first characteristic value is the characteristic value of electric energy data, which mainly includes the following indicators:

[0070] 1. Voltage stability: refers to the degree to which the grid voltage remains stable within a certain time and space range. Common measurement indicators include voltage deviation, flicker and short-term fluctuation;

[0071] 2. Current stability: Similar to voltage stability, it focuses on the stability of current within a certain time and space range. The main indicators include harmonic distortion, interharmonic distortion and instantaneous overload, etc.

[0072] 3. Power supply reliability: refers to the ability of the power system to continuously and stably supply power to users. It is usually measured by power supply reliability index (such as average fault time interval, average fault duration, etc.);

[0073] 4. Frequency stability: Pay attention to the range and rate of change of the power grid frequency. The main indicators are frequency deviation and frequency change rate, etc.

[0074] 5. Unbalance: reflects the degree of asymmetry between phases in a three-phase power system. Mainly includes voltage imbalance, current imbalance and negative sequence component, etc.

[0075] 6. Instantaneous power quality: focus on the instantaneous characteristics of the power system, such as instantaneous power, instantaneous reactive power, etc.;

[0076] 7. Harmonics and interharmonics: refers to the non-fundamental frequency components in the power system. The main parameters include harmonic content, total harmonic distortion and interharmonic content.

[0077] Step S3, classifying and identifying the first eigenvalues ​​according to the support vector machine optimized by the locust optimization algorithm to obtain a power supply quality assessment result;

[0078] In this step, the locust optimization algorithm is first used to optimize the support vector machine. Support vector machine SVM is a machine learning tool that can effectively solve the classification and prediction of problems. The key to the support vector machine is to find the largest feature space spacing in order to correctly classify the data.

[0079] The support vector machine requires the function interval to be greater than 1 when performing classification. However, when there are noise points in the data set, the support vector machine will cause the interval of the hyperplane to become smaller in order to correctly divide the noise points, which greatly reduces the generalization performance. In addition, when the noise points are mixed into another class, it becomes a linear inseparable problem and must be converted through the kernel function, which will complicate the model and may cause overfitting. Therefore, the locust optimization algorithm is needed for optimization processing, which can not only avoid the complexity of the model, but also improve the generalization performance to optimize the output results.

[0080] The locust optimization algorithm (GOA) is a new metaheuristic algorithm proposed by S. Saremi et al. in 2017. This method is inspired by the large-scale movement of adults and adult groups and the aggregation behavior during foraging, and has a small amount of computation and a simple calculation formula. Through optimization experiments on standard test functions, it is proved that GOA converges better than the particle swarm optimization algorithm (PSO).

[0081] Firstly, the first eigenvalue is calculated using the locust optimization algorithm with the maximum eigenvalue spacing of the first eigenvalue as a target.

[0082] In the locust optimization algorithm, we first assume that there are N locusts in the population, and the specific position of each locust is X =

[0083] (x1,x2,...,xn), the quality of the position is determined by setting the objective function. Therefore, the optimal position update method of the locust swarm is shown in the following formula:

[0084]

[0085] In the formula, d is the dimension of the first eigenvalue, i and j are locust individuals, is the optimal position of the locust individual, S i,j is the distance between the i-th locust and the j-th locust, are the upper and lower limits of the d-th dimension space, respectively; c is a control parameter used to control the dynamic and uncertain search capabilities of the algorithm; and s is the degree of interaction influence.

[0086] The expressions for c and s are as follows:

[0087]

[0088]

[0089] In the formula, c max and c min are the maximum and minimum values ​​of the decreasing interval, respectively, with values ​​of 1 and 0.0001, t is the current number of calculations, T max is the maximum number of calculations, f is the attraction intensity value, which is 0.5, l is the attraction scale value, which is 1.5, and s(r) is the social influence coefficient.

[0090] After the locust optimization algorithm, the optimal feature spacing is obtained.

[0091] Then, according to the optimal feature spacing, the penalty factor and kernel function of the vector machine are optimized.

[0092] The first eigenvalue is input into the optimized vector machine, and the first eigenvalue is classified and identified according to the optimal feature distance to obtain a power supply quality assessment result.

[0093] Step S4: According to the power supply quality assessment result, the power supply link is inspected and repaired.

[0094] In this step, based on the high-precision power supply quality assessment results obtained, the nodes with risks in the power supply link are repaired to improve the operational stability of the power supply link.

[0095] Furthermore, in order to execute the power supply quality assessment system based on convolutional neural network corresponding to the above method embodiment to achieve the corresponding functions and technical effects, Figure 2A structural diagram of a power supply quality assessment system based on a convolutional neural network is provided. For ease of description, only the parts related to this embodiment are shown. The power supply quality assessment system based on a convolutional neural network provided by the embodiment of the present invention includes:

[0096] The feature extraction module 201 is used to perform convolution processing and nonlinear feature mapping on the electric energy data collected in real time from the power supply link according to a preset convolution module and activation function, and extract a first feature map of the electric energy data;

[0097] A data conversion module 202, configured to perform a pooling operation and a vector conversion on the first feature map to obtain a first eigenvalue;

[0098] A data classification module 203 is used to classify and identify the first eigenvalue according to a support vector machine optimized by a locust optimization algorithm to obtain a power supply quality assessment result;

[0099] The power supply link maintenance module 204 is used to inspect and repair the power supply link according to the power supply quality evaluation result.

[0100] In some embodiments, the feature extraction module 201 further includes:

[0101] A data convolution and mapping unit, used for inputting the electric energy data into a convolution module to extract local area information of the electric energy data; extracting local features of the electric energy data according to the local area information and a plurality of convolution kernels; wherein each convolution kernel includes a corresponding weight coefficient and a deviation; inputting the local features of the electric energy data into a preset activation function for nonlinear feature mapping to obtain the first feature map.

[0102] A data preprocessing unit is used to preprocess the electric energy data collected in real time from the power supply link before performing convolution processing and nonlinear feature mapping on the electric energy data, and then convert the preprocessed electric energy data into the data format required by the convolutional neural network; wherein the preprocessing includes de-averaging, normalization and dimensionality reduction.

[0103] In some embodiments, the data conversion module 202 further includes:

[0104] A data pooling unit is used to process the first feature map using an edge-padding method if the dimension of the first feature map is not a multiple of a first threshold value so that the first feature map meets the pooling condition; use a preset pooling window and step size to slide on the first feature map that meets the pooling condition to perform a pooling operation to obtain a second feature map; and perform vector conversion on the second feature map to obtain a first eigenvalue.

[0105] A data edge zero padding unit is used to calculate the difference between the dimension of the first feature map and the multiple of the next first threshold to obtain the number of zero padding; add the number of zero padding zeros to the edge of the first feature map to obtain the first feature map that meets the pooling condition.

[0106] In some embodiments, the data classification module 203 further includes:

[0107] A data classification unit is used to calculate the first eigenvalue using the locust optimization algorithm with the maximum feature distance of the first eigenvalue as the target to obtain the optimal feature distance; according to the optimal feature distance, the penalty factor and kernel function of the support vector machine are optimized; the first eigenvalue is input into the optimized support vector machine, and the first eigenvalue is classified and identified according to the optimal feature distance to obtain a power supply quality assessment result.

[0108] This embodiment proposes a power supply quality assessment and system based on a convolutional neural network: according to a preset convolution module and activation function, convolution processing and nonlinear feature mapping are performed on the electric energy data collected in real time from the power supply link to extract the first feature map of the electric energy data; pooling operation and vector conversion are performed on the first feature map to obtain the first eigenvalue; according to the support vector machine optimized by the locust optimization algorithm, the first eigenvalue is classified and identified to obtain the power supply quality assessment result; according to the power supply quality assessment result, the power supply link is repaired. Its beneficial effects are: using a convolutional neural network to extract the eigenvalues ​​of the electric energy data of the whole process, and using the locust optimization algorithm to improve the efficiency of classifying and identifying the eigenvalues, and improve the accuracy of evaluating the overall power supply quality.

[0109] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A power supply quality assessment method based on convolutional neural network, characterized in that: include: According to a preset convolution module and activation function, convolution processing and nonlinear feature mapping are performed on the electric energy data collected in real time from the power supply link to extract a first feature map of the electric energy data; Performing a pooling operation and a vector conversion on the first feature map to obtain a first eigenvalue; Classifying and identifying the first eigenvalues ​​according to a support vector machine optimized by a locust optimization algorithm to obtain a power supply quality assessment result; According to the power supply quality assessment results, the power supply link is inspected and repaired.

2. The power supply quality assessment method based on convolutional neural network according to claim 1, characterized in that: According to the preset convolution module and activation function, the electric energy data collected in real time from the power supply link is convoluted and nonlinearly mapped to extract the first feature map of the electric energy data, which is specifically: Inputting the electric energy data into a convolution module to extract local area information of the electric energy data; Extracting local features of the electric energy data according to the local area information and a plurality of convolution kernels; wherein each convolution kernel includes a corresponding weight coefficient and a deviation; The local features of the electric energy data are input into a preset activation function for nonlinear feature mapping to obtain the first feature map.

3. The power supply quality assessment method based on convolutional neural network according to claim 1, characterized in that: Before performing convolution processing and nonlinear feature mapping on the electric energy data collected in real time from the power supply link, the electric energy data is first preprocessed, and then the preprocessed electric energy data is converted into the data format required by the convolutional neural network; wherein the preprocessing includes de-averaging, normalization and dimensionality reduction.

4. The power supply quality assessment method based on convolutional neural network according to claim 1, characterized in that: The first feature map is subjected to a pooling operation and a vector conversion to obtain a first feature value, specifically: If the dimension of the first feature map is not a multiple of the first threshold, the first feature map is processed using an edge zero padding method so that the first feature map satisfies a pooling condition; Using a preset pooling window and step size, sliding on the first feature map that meets the pooling condition to perform a pooling operation to obtain a second feature map; Perform vector transformation on the second eigenmap to obtain the first eigenvalue.

5. The power supply quality assessment method based on convolutional neural network according to claim 4 is characterized in that: If the dimension of the first feature map is not a multiple of the first threshold, the first feature map is processed using an edge zero padding method, specifically: Calculate the difference between the dimension of the first feature map and the multiple of the next first threshold value to obtain the number of zero padding; Add zero padding number zeros to the edge of the first feature map to obtain the first feature map that meets the pooling condition.

6. The power supply quality assessment method based on convolutional neural network according to claim 1, characterized in that: The first eigenvalue is classified and identified according to the support vector machine optimized by the locust optimization algorithm to obtain a power supply quality assessment result, which is specifically: Taking the maximum characteristic distance of the first eigenvalue as the target, using the locust optimization algorithm to calculate the first eigenvalue to obtain the optimal characteristic distance; According to the optimal feature spacing, the penalty factor and kernel function of the support vector machine are optimized; The first eigenvalue is input into the optimized support vector machine, and the first eigenvalue is classified and identified according to the optimal feature distance to obtain a power supply quality assessment result.

7. A power supply quality assessment system based on convolutional neural network, characterized in that: include: Feature extraction module, data conversion module, data classification module and power supply link maintenance module; The feature extraction module is used to perform convolution processing and nonlinear feature mapping on the electric energy data collected in real time from the power supply link according to a preset convolution module and activation function, and extract a first feature map of the electric energy data; The data conversion module is used to perform pooling operation and vector conversion on the first feature map to obtain a first eigenvalue; The data classification module is used to classify and identify the first eigenvalue according to the support vector machine optimized by the locust optimization algorithm to obtain a power supply quality assessment result; The power supply link maintenance module is used to inspect and repair the power supply link according to the power supply quality evaluation result.

8. The power supply quality assessment system based on convolutional neural network according to claim 7, characterized in that: The feature extraction module includes: a data convolution and mapping unit; The data convolution and mapping unit is used to input the electric energy data into a convolution module to extract local area information of the electric energy data; extract local features of the electric energy data based on the local area information and a plurality of convolution kernels; wherein each convolution kernel includes a corresponding weight coefficient and a deviation; and input the local features of the electric energy data into a preset activation function for nonlinear feature mapping to obtain the first feature map.

9. The power supply quality assessment system based on convolutional neural network according to claim 7, characterized in that: The feature extraction module includes: a data preprocessing unit; The data preprocessing unit is used to preprocess the electric energy data collected in real time from the power supply link before performing convolution processing and nonlinear feature mapping on the electric energy data, and then convert the preprocessed electric energy data into the data format required by the convolutional neural network; wherein the preprocessing includes de-averaging, normalization and dimensionality reduction.

10. The power supply quality assessment system based on convolutional neural network according to claim 7, characterized in that: The data conversion module includes: a data pooling unit; The data pooling unit is used to process the first feature map using an edge zero padding method if the dimension of the first feature map is not a multiple of a first threshold value so that the first feature map meets the pooling condition; use a preset pooling window and step size to slide on the first feature map that meets the pooling condition to perform a pooling operation to obtain a second feature map; and perform vector conversion on the second feature map to obtain a first eigenvalue.

11. The power supply quality assessment system based on convolutional neural network according to claim 7, characterized in that: The data conversion module comprises: a data edge zero padding unit; The data edge zero-padding unit is used to calculate the difference between the dimension of the first feature map and the multiple of the next first threshold to obtain the number of zero padding; add the number of zero padding zeros to the edge of the first feature map to obtain the first feature map that meets the pooling condition.

12. The power supply quality assessment system based on convolutional neural network according to claim 7, characterized in that: The data classification module includes: a data classification unit; The data classification unit is used to calculate the first eigenvalue using the locust optimization algorithm with the maximum feature distance of the first eigenvalue as the target, so as to obtain the optimal feature distance; optimize the penalty factor and kernel function of the support vector machine according to the optimal feature distance; input the first eigenvalue into the optimized support vector machine, classify and identify the first eigenvalue according to the optimal feature distance, and obtain the power supply quality assessment result.

Images