CN115619999A

CN115619999A - Real-time monitoring method and device for power equipment, electronic equipment and readable medium

Info

Publication number: CN115619999A
Application number: CN202110798702.9A
Authority: CN
Inventors: 尚芳剑; 李信; 张希; 王博; 彭柏; 陈重韬; 王艺霏; 李欣怡; 娄竞; 张少军; 闫忠平; 王东升; 李贤�; 来骥; 赵婷; 陈江琦; 夏卫尚; 刘思言; 吴春鹏; 刘慧丽
Original assignee: State Grid Corp of China SGCC; Global Energy Interconnection Research Institute; Information and Telecommunication Branch of State Grid Jibei Electric Power Co Ltd
Current assignee: State Grid Corp of China SGCC; Global Energy Interconnection Research Institute; Information and Telecommunication Branch of State Grid Jibei Electric Power Co Ltd
Priority date: 2021-07-15
Filing date: 2021-07-15
Publication date: 2023-01-17

Abstract

The disclosure relates to a real-time monitoring method and device for power equipment, electronic equipment and a computer readable medium. The method comprises the following steps: acquiring a real-time image of the power equipment through the Internet of things; preprocessing the real-time image to generate image data; inputting the image data into a target detection model to generate a target detection result, wherein the target detection model is a deep neural network model which is based on Newton method optimization loss function for compression; and monitoring the state of the power equipment in real time based on the target detection result. According to the real-time monitoring method and device for the power equipment, the electronic equipment and the computer readable medium, a high-precision and low-complexity calculation model can be arranged in the edge calculation equipment, so that the power equipment can be monitored in real time, the safe operation of a power grid is ensured, and the data pressure of the Internet of things is reduced.

Description

Real-time monitoring method and device for power equipment, electronic equipment and readable medium

Technical Field

The disclosure relates to the field of operation and detection of power equipment, in particular to a real-time monitoring method and device for power equipment, electronic equipment and a computer readable medium.

Background

The power transformation equipment of the power grid in China has the characteristics of various types, wide distribution and different structural parameters, and in the long-term operation process, the power equipment has faults which are almost unavoidable, and the reasons for causing the faults comprise the equipment defects left in the manufacturing process, the problems existing in the installation, overhaul and maintenance, the factors of insulation aging, structural degradation and the like caused after long-term operation. At present, the intelligent tour equipment of tour business in the transformer substation is used more to go on, such as unmanned aerial vehicle, the robot, intelligent helmet etc., the work load of fortune dimension staff has been lightened greatly, reduce the daily fortune dimension's of power equipment cost, through shooing in real time to the equipment status, can produce a large amount of equipment image data, and whether need the manual work to judge the equipment status according to the image normal, this process has increased fortune dimension staff's work load, occupy a large amount of manpowers, consequently, utilize intelligent recognition technique to carry out analysis processes to data, acquire key information, and then it is very necessary to realize that data processing is automatic.

Edge computing allows more of the monitoring application's computational tasks to be performed at distributed nodes at the edge of the network, so time delays can be reduced with these edge devices and real-time online decisions can be made. In the field of operation and maintenance and detection of power equipment, real-time state monitoring needs to be performed on power equipment in a power internet of things environment, and since a power internet of things deployed by a power enterprise is composed of a large number of power terminal devices, an electric power monitoring application system based on edge computing is generally adopted to analyze and process acquired real-time images of the power equipment so as to monitor whether the equipment is good or not. However, the complex network model has resource limitation problem when terminal or edge device is deployed.

The above information disclosed in this background section is only for enhancement of understanding of the background of the disclosure and therefore it may contain information that does not constitute prior art that is already known to a person of ordinary skill in the art.

Disclosure of Invention

In view of this, the present disclosure provides a method and an apparatus for monitoring an electrical device in real time, an electronic device, and a computer readable medium, which can arrange a high-precision and low-complexity calculation model in an edge calculation device, so as to monitor the electrical device in real time, and reduce data pressure of the internet of things while ensuring safe operation of a power grid.

Additional features and advantages of the disclosure will be set forth in the detailed description which follows, or in part will be obvious from the description, or may be learned by practice of the disclosure.

According to an aspect of the present disclosure, a method for monitoring an electric power device in real time is provided, the method including: acquiring a real-time image of the power equipment through the Internet of things; preprocessing the real-time image to generate image data; inputting the image data into a target detection model to generate a target detection result, wherein the target detection model is a deep neural network model which is based on a Newton method to optimize a loss function for compression; and monitoring the state of the power equipment in real time based on the target detection result.

In an exemplary embodiment of the present disclosure, further comprising: acquiring a plurality of historical images of the power equipment; preprocessing the plurality of historical images to generate a plurality of historical image data; training a deep neural network model based on the plurality of historical image data to generate an initial target detection model; and optimizing a loss function based on a Newton method to compress the initial target detection model to generate the target detection model.

In an exemplary embodiment of the present disclosure, preprocessing the plurality of history images to generate a plurality of history image data includes: performing data cleansing on the plurality of historical images to generate a plurality of historical image data; and/or performing data conversion on the plurality of historical images to generate a plurality of historical image data; and/or performing data normalization processing on the plurality of historical images to generate a plurality of historical image data.

In an exemplary embodiment of the present disclosure, optimizing a loss function based on newton's method to compress the initial target detection model to generate the target detection model includes: generating an exponential loss function; generating a compression function; performing optimization calculation on the exponential loss function based on a Newton method to obtain a minimized exponential loss function and corresponding model parameters thereof; decoupling the compression function to obtain a weight of the compressed target detection model; generating the target detection model based on the model parameters and the weights.

In an exemplary embodiment of the present disclosure, generating an exponential-loss function includes: acquiring initialization parameters, wherein the initialization parameters comprise a compression target, algorithm iteration times and sample iteration times; generating the exponential-loss function based on the initialization parameter.

In an exemplary embodiment of the present disclosure, generating a compression function includes: acquiring an original weight tensor of the initial target detection model; generating the compression function based on the original weight tensor.

In an exemplary embodiment of the present disclosure, performing an optimization calculation on the exponential loss function based on newton's method to obtain a minimized exponential loss function and a corresponding model parameter thereof includes: determining an iteration initial value; determining a searching direction; performing iterative calculations based on the search direction to solve a minimized exponential function; determining the model parameters based on the minimization exponential function.

In an exemplary embodiment of the present disclosure, determining a search direction includes: and performing linear transformation on the gradient through a Hessian matrix based on the iteration initial value to obtain a search direction.

In an exemplary embodiment of the present disclosure, decoupling the compression function to obtain the weights of the compressed target detection model includes: determining a dual variable; decoupling the compression function based on the dual variable and an alternating direction multiplier method; calculating the decoupled compression function based on iterative calculation to obtain the weight of the target detection model.

According to an aspect of the present disclosure, a real-time monitoring device for power equipment is provided, the device including: the image module is used for acquiring a real-time image of the power equipment through the Internet of things; the processing module is used for preprocessing the real-time image to generate image data; the calculation module is used for inputting the image data into a target detection model to generate a target detection result, wherein the target detection model is a deep neural network model which is based on Newton method optimization loss function for compression; and the monitoring module is used for monitoring the state of the power equipment in real time based on the target detection result.

In an exemplary embodiment of the present disclosure, further comprising: the model training module is used for acquiring a plurality of historical images of the power equipment; preprocessing the plurality of historical images to generate a plurality of historical image data; training a deep neural network model based on the plurality of historical image data to generate an initial target detection model; and optimizing a loss function based on a Newton method to compress the initial target detection model to generate the target detection model.

According to an aspect of the present disclosure, an electronic device is provided, the electronic device including: one or more processors; storage means for storing one or more programs; when executed by one or more processors, cause the one or more processors to implement a method as above.

According to an aspect of the disclosure, a computer-readable medium is proposed, on which a computer program is stored, which program, when being executed by a processor, carries out the method as above.

According to the real-time monitoring method and device for the power equipment, the electronic equipment and the computer readable medium, the real-time image of the power equipment is acquired through the Internet of things; preprocessing the real-time image to generate image data; inputting the image data into a target detection model to generate a target detection result, wherein the target detection model is a deep neural network model which is based on a Newton method to optimize a loss function for compression; the mode of monitoring the state of the power equipment in real time based on the target detection result can arrange a high-precision and low-complexity calculation model in the edge calculation equipment so as to monitor the power equipment in real time, ensure the safe operation of a power grid and reduce the data pressure of the Internet of things.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The above and other objects, features and advantages of the present disclosure will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings. The drawings described below are merely some embodiments of the present disclosure, and other drawings may be derived from those drawings by those of ordinary skill in the art without inventive effort.

Fig. 1 is a system block diagram illustrating a method and apparatus for real-time monitoring of power equipment according to an exemplary embodiment.

FIG. 2 is a flow chart illustrating a method for real-time monitoring of electrical devices, according to an exemplary embodiment.

Fig. 3 is a flow chart illustrating a method for real-time monitoring of an electrical device, according to another exemplary embodiment.

Fig. 4 is a flow chart illustrating a method for real-time monitoring of an electrical device, according to another exemplary embodiment.

Fig. 5 is a schematic diagram illustrating a method for real-time monitoring of an electrical device, according to another exemplary embodiment.

Fig. 6 is a schematic diagram illustrating a method for real-time monitoring of an electrical device, according to another exemplary embodiment.

Fig. 7 is a block diagram illustrating a real-time monitoring apparatus for electrical equipment according to an exemplary embodiment.

FIG. 8 is a block diagram of an electronic device shown in accordance with an example embodiment.

FIG. 9 is a block diagram illustrating a computer-readable medium in accordance with an example embodiment.

Detailed Description

Example embodiments will now be described more fully with reference to the accompanying drawings. Example embodiments may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and a repetitive description thereof will be omitted.

Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the disclosure. One skilled in the relevant art will recognize, however, that the embodiments of the disclosure can be practiced without one or more of the specific details, or with other methods, components, devices, steps, and the like. In other instances, well-known methods, devices, implementations, or operations have not been shown or described in detail to avoid obscuring aspects of the disclosure.

The block diagrams shown in the figures are functional entities only and do not necessarily correspond to physically separate entities. I.e. these functional entities may be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different networks and/or processor means and/or microcontroller means.

The flowcharts shown in the figures are illustrative only and do not necessarily include all of the contents and operations/steps, nor do they necessarily have to be performed in the order described. For example, some operations/steps may be decomposed, and some operations/steps may be combined or partially combined, so that the actual execution sequence may be changed according to the actual situation.

It will be understood that, although the terms first, second, third, etc. may be used herein to describe various components, these components should not be limited by these terms. These terms are used to distinguish one element from another. Thus, a first component discussed below may be termed a second component without departing from the teachings of the disclosed concept. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.

It is to be understood by those skilled in the art that the drawings are merely schematic representations of exemplary embodiments, and that the blocks or processes shown in the drawings are not necessarily required to practice the present disclosure and, therefore, are not intended to limit the scope of the present disclosure.

Network equipment in the environment of the internet of things accumulates a large amount of data, and cloud computing related technologies have the potential to be successful in processing the large amount of data, but for most typical monitoring applications, processing and analyzing the large amount of monitoring data still presents a challenge. The complex network model has the problem of resource limitation when the terminal or the edge device is deployed, the neural network model compression technology is researched, and the transplantation and the calculation optimization of the neural network model at the terminal or the edge device become the key for reducing the edge calculation complexity based on the monitoring application on the premise of ensuring the model precision.

The inventors of the present application have found that over the past few years, various techniques for compressing DNN (deep neural network) models have been proposed. Pruning and quantification are the two most widely used methods in practice. The pruning method not only uses weight pruning to compress the model, but also provides channel (filter/neuron) pruning to remove the whole filter of CNN weight, thereby realizing reasoning acceleration. Quantization is considered as another direction of DNN compression, in addition to parameter reduction by pruning. The quantization intervals may be uniform or non-uniform, and in general, non-uniform quantization may achieve a higher compression rate, while uniform quantization may provide acceleration. The quantization bit width can be further reduced by Hoffman coding. In addition to scalar quantization, vector quantization may also be applied to DNN model compression.

In order to maximize compression performance, there are some methods that train pruning and quantization together, but in this case there is a problem that the sparsity of the hierarchy and the quantization bit width affect each other, and these methods rely on setting a hyper-parameter to compress the hierarchy, increasing the difficulty of manually selecting compression ratio or hyper-parameter adjustment. In view of the technical bottleneck existing in the prior art, the application provides a real-time monitoring method for electrical equipment.

In the real-time monitoring method for the power equipment disclosed by the present disclosure, a specific technical description is given by taking a monitoring application of operation and maintenance of the power equipment as an example, and it is to be understood that the method disclosed by the present disclosure may also be applied to other fields, and the present application is not limited thereto.

More specifically, in a power distribution operation and maintenance target identification task, a real-time device state image shot by intelligent inspection equipment adopted by inspection business in power distribution operation and maintenance is collected, and a proper research sample is selected from the real-time device state image. And then preprocessing the data by adopting an image enhancement technology. And secondly, manually labeling the image to obtain a data set for target detection of the power equipment. The image dataset may then be used for an object detection task based on edge calculation. In the automatic DNN compression framework provided by the invention, firstly, an image data set needs to be preprocessed, the image data is subjected to data cleaning to remove dirty data, then the image data is converted into tensor, and then the mean value and the variance are set to normalize the image channel by channel; and then, carrying out model training based on the processed image data set, compressing the trained model by adopting an optimized automatic DNN compression method, accelerating the training convergence process of the loss function by utilizing the optimization method of the loss function, and finally obtaining the compressed neural network model.

Compared with a DNN model in the prior art, the target detection model provided by the real-time monitoring method for the power equipment has higher compression ratio and accuracy, the size of the DNN model is obviously reduced, and the complexity of the model can be effectively reduced. The method of the present disclosure is described below with the aid of specific examples.

As shown in fig. 1, the system architecture 10 may include

terminal devices

101, 102, 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the

terminal devices

101, 102, 103 and the server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.

The

terminal devices

101, 102, 103 may interact with a server 105 over a network 104 to receive or transmit data or the like. Various communication client applications, such as a video monitoring application, a web browser application, an instant data transmission application, a mailbox client, etc., may be installed on the

terminal devices

101, 102, 103.

The

terminal devices

101, 102, 103 may be various electronic devices having a monitoring function and supporting data transmission or calculation, including but not limited to power electronics, smart cameras, smart monitoring meters, and the like.

The

terminal devices

101, 102, 103 may obtain real-time images of the power devices, for example, through the internet of things; the

terminal devices

101, 102, 103 may, for example, pre-process the real-time image to generate image data; the

terminal devices

101, 102, 103 may, for example, input the image data into a target detection model to generate a target detection result, where the target detection model is a deep neural network model that optimizes a loss function based on a newton method for compression; the

terminal devices

101, 102, 103 may monitor the status of the electrical devices in real time, for example, based on the target detection results.

The server 105 may be a server providing various services, such as a backend server providing support for tasks processed by the

terminal devices

101, 102, 103. The background server can analyze and process the received request and feed back the processing result (the compressed monitoring model) to the terminal equipment.

The server 105 may, for example, obtain a plurality of historical images of the electrical devices; the server 105 may, for example, pre-process the plurality of historical images to generate a plurality of historical image data; server 105 may train a deep neural network model to generate an initial target detection model, e.g., based on the plurality of historical image data; server 105 may generate the object detection model based on newton's method optimization loss function to compress the initial object detection model, for example.

The server 105 may be a physical server, and may also be composed of a plurality of servers, for example, it should be noted that the method for monitoring the power equipment in real time provided by the embodiment of the present disclosure may be executed by the server 105 and/or the

terminal devices

101, 102, and 103, and accordingly, the device for monitoring the power equipment in real time may be disposed in the server 105 and/or the

terminal devices

101, 102, and 103.

Fig. 2 is a flow chart illustrating a method for real-time monitoring of an electrical device, according to an example embodiment. The real-time monitoring method 20 for the power equipment can be applied to the edge equipment of the internet of things, and at least includes steps S202 to S208.

As shown in fig. 2, in S202, a real-time image of the power device is acquired through the internet of things. The internet of things (IOT) is used for collecting any object or process needing monitoring, connection and interaction in real time through various devices and technologies such as various information sensors, radio frequency identification technologies, global positioning systems, infrared sensors and laser scanners, collecting various required information such as sound, light, heat, electricity, mechanics, chemistry, biology and positions of the object or process, realizing ubiquitous connection of the object and people through various possible network accesses, and realizing intelligent sensing, identification and management of the object and the process. An edge device in the internet of things (lot) may obtain and process real-time images, where an edge device (edge device) is a device that provides an entry point to an enterprise or service provider core network.

In S204, the real-time image is preprocessed to generate image data. In order to purify a data set, the obtained original power equipment image data needs to be screened in a certain scale, fuzzy images, repeated images and damaged image files are removed, unnecessary influence on the quality of a training model is avoided, and image data with clear targets are reserved.

The data conversion can store the acquired image data sets with different types of training data in different folders according to labels respectively, and then load the data, and the image gray scale range is converted from [0,255] to [0,1 ]. The given image is randomly cropped to different sizes and aspect ratios, the cropped image is then scaled to the specified size, and the given PIL image is flipped at a given probability random level. Converting the image into a storage format in a memory, inputting bytes in a stream form, converting the bytes into a one-dimensional tensor, reorganizing and transposing the tensor, and dividing each element of the current tensor by 255 to output the tensor.

In order to increase the convergence rate of the model, the image is normalized channel by channel, and the calculation formula is shown as formula (1), wherein mean represents the mean value of each channel, and std represents the standard deviation of each channel. The data can be normalized using this formula, with a mean value of 0 and a standard deviation of 1.

output＝(input-mean)/std(1)

In S206, the image data is input into a target detection model, and a target detection result is generated, where the target detection model is a deep neural network model that optimizes a loss function based on a newton method for compression.

In S208, the state of the power equipment is monitored in real time based on the target detection result. When the detection result contains the preset target, warning information can be generated. The preset target may be an image that marks damage to the power equipment.

According to the real-time monitoring method for the power equipment, a real-time image of the power equipment is obtained through the Internet of things; preprocessing the real-time image to generate image data; inputting the image data into a target detection model to generate a target detection result, wherein the target detection model is a deep neural network model which is based on a Newton method to optimize a loss function for compression; the mode of monitoring the state of the power equipment in real time based on the target detection result can arrange a high-precision and low-complexity calculation model in the edge calculation equipment so as to monitor the power equipment in real time, ensure the safe operation of a power grid and simultaneously reduce the data pressure of the Internet of things.

It should be clearly understood that this disclosure describes how to make and use particular examples, but the principles of this disclosure are not limited to any details of these examples. Rather, these principles can be applied to many other embodiments based on the teachings of the present disclosure.

FIG. 3 is a flow chart illustrating a method for real-time monitoring of electrical devices, according to another exemplary embodiment. The real-time power equipment monitoring method 30 can be applied to a server supporting an edge computing device, and the flow 30 shown in fig. 3 is a supplementary description of the flow shown in fig. 2.

As shown in fig. 3, in S302, a plurality of history images of the electric power device are acquired.

In S304, the plurality of history images are preprocessed to generate a plurality of history image data. The method comprises the following steps: performing data cleansing on the plurality of historical images to generate a plurality of historical image data; and/or data-converting the plurality of historical images to generate a plurality of historical image data; and/or performing data normalization processing on the plurality of historical images to generate a plurality of historical image data.

In S306, a deep neural network model is trained based on the plurality of historical image data to generate an initial target detection model. And inputting the image data subjected to data preprocessing into a DNN (digital noise network) model for training to generate an initial target detection model.

In S308, a loss function is optimized based on newton' S method to compress the initial target detection model to generate the target detection model. An exponential loss function may be generated, for example; generating a compression function; performing optimization calculation on the exponential loss function based on a Newton method to obtain a minimized exponential loss function and a corresponding model parameter; decoupling the compression function to obtain a weight of the compressed target detection model; generating the target detection model based on the model parameters and the weights.

The method is based on the combination of pruning and quantification of the model, under the premise of carrying out data preprocessing on image data of the power equipment, original weight and initialization parameters of the model are calculated, and decoupling is carried out on the parameters by utilizing an alternative iteration multiplier Algorithm (ADMM).

In the real-time monitoring method of the power equipment, a model compression optimization technology based on a Newton method loss function is provided. Different from classical loss functions such as traditional information entropy and mean square error, the method adopts a Newton method to accelerate the training convergence process of the loss functions and improve the precision of the classification result of the model.

Experimental results show that compared with the existing popular method, the method provided by the patent has higher compression ratio and accuracy, reduces the size of the DNN model, and can effectively reduce the complexity of the model.

Fig. 4 is a flow chart illustrating a method for real-time monitoring of an electrical device, according to another exemplary embodiment. The process 40 shown in fig. 4 is a detailed description of S308 "optimizing a loss function based on newton' S method to compress the initial object detection model to generate the object detection model" in the process shown in fig. 3.

As shown in fig. 4, in S402, an exponential-loss function is generated. The method comprises the following steps: acquiring initialization parameters, wherein the initialization parameters comprise a compression target, algorithm iteration times and sample iteration times; generating the exponential-loss function based on the initialization parameter.

In S404, a compression function is generated. The method comprises the following steps: acquiring an original weight tensor of the initial target detection model; generating the compression function based on the raw weight tensor.

In S406, the exponential loss function is optimized and calculated based on newton' S method to obtain a minimized exponential loss function and its corresponding model parameters. The method comprises the following steps: determining an iteration initial value; determining a search direction; performing iterative calculations based on the search direction to solve a minimized exponential function; determining the model parameters based on the minimization exponential function.

Wherein, determining the search direction comprises: and performing linear transformation on the gradient through a Hessian matrix based on the iteration initial value to obtain a search direction.

In S408, the compression function is decoupled to obtain the weights of the compressed target detection model. The method comprises the following steps: determining a dual variable; decoupling the compression function based on the dual variable and an alternating direction multiplier method; calculating the decoupled compression function based on iterative calculation to obtain the weight of the target detection model.

In S410, the object detection model is generated based on the model parameters and the weights.

In one embodiment, the original weight tensor W of the model may be computed, and the initialization parameters include the target size for model compression, the total number of SGD iterations of the algorithm, and the number of iterations required to train all samples in the training set once. Then, a Loss function l is defined, and the present disclosure performs the calculation of the Loss using an Exponential Loss function (Exponential Loss), as shown in equation (2). Where n is the number of samples, y is the true value of the samples, f (x) _i ) Is the weight of the ith iteration model.

The general function of DNN compression is shown in equation (3), which is compressedThe constraint of reducing the total size of the DNN weights aims at minimizing the loss function. Wherein

Is a set of weight vectors with L-level DNN, b (W) ⁽ⁱ⁾ ) Is the minimum bit width of all non-zero elements in the code W, L0 norm | | | W | | survival ₀ Is the number of non-zero elements in W, Z is the model target size, and l is the loss function.

Because b (|) and | | | | | | | | | ₀ It is an irreducible function, and cannot be solved by a normal training algorithm, and in the embodiment of the present application, an ADMM method may be used to decouple the L0 norm and the bit width portion thereof, which may be specifically shown in formula (4). By introducing dual variables

And absorbing the equality constraint into the augmented Lagrangian function, wherein lambda > 0, is a hyper-parameter.

Is a copy of the DNN weight W.

The model compression method based on joint pruning and quantization can be normally trained by using the ADMM method, and b (.) and | can be solved by iteratively updating 3 variables W, V and Y in formula (4) ₀ An insubstantial problem. Finally, the weight of the compressed DNN model is output.

In order to optimize the loss function l, in the present disclosure, a search direction may be obtained by performing linear transformation on a gradient through a Hessian matrix using the idea of newton method, thereby accelerating the convergence process of the loss function. Firstly, an iteration initial value x0 epsilon omega is selected, epsilon is selected to be larger than 0, and the following operations are repeated. If it is

The cycle is stopped. Then, the gradient G is calculated, as shown in equation (5), where the total number of training samples is n, t =0, 1.., n, f (x) is a loss function, and x is an optimized parameter object:

the calculation of the Hessian matrix H is shown in equation (6);

the search direction d is calculated as shown in equation (7):

the calculation of the last update iteration point is shown in equation (8):

x _t+1 ＝x _t -d _t (8)

as shown above, the minimum value of the function is solved through continuous iteration, the minimized loss function and the model parameter value are finally obtained, and the parameter optimization of the model is realized.

The following describes the usage effect of the model compression method according to a specific experimental result:

experimental setup and data set

The experiments were performed using Win10, GPU with GTX1070 and 8G memory and CUDA/CUDNN. The training data set is 5944 collected power equipment state images, wherein the 5944 collected power equipment state images comprise six types of equipment state images of a breather, a meter, an insulator, oil leakage, foreign matters, metal corrosion and the like. Standard training/testing data partitioning is carried out on the acquired data set, 70% of the whole data set is used for training the model to be used as a training set, and 30% of the whole data set is used for testing the model to be used as a testing set. The performance of the disclosed compression method was tested using AlexNet as a DNN model for training and testing power image data.

The batch size may be set to 128, the exponential loss function l (W) optimized using momentum SGD, the initial learning rate to 0.005, the decay learning rate using a cosine annealing strategy, and the hyperparameter λ =0.05 set. In order to make the effect of the comparison between the methods more evident, the compression budget Z may also be set to a value close to or smaller than the comparison method. The experiment was performed 120 iterations on a data set containing 4160 training examples.

Index of experiment

In order to evaluate the performance of model compression, the compression factor and accuracy were used as evaluation indexes, as shown in formula (9) and formula (10).

Wherein R is _C Denotes the compression factor, N _original Representing the number of original model parameters, N _compressed Representing the number of model parameters after compression.

Wherein A represents the accuracy, N _correct Number of samples representing correct classification, N _test Representing the total number of samples in the test set. In the accuracy index, the fact that the sample classification is correct means that the predicted label takes the largest one of the last probability vectors as the prediction result, and if the part with the largest probability in the prediction results is used as the part with the largest probabilityAnd if the class is correct, predicting to be correct, and then calculating the frequency of correct classification to obtain the accuracy.

Comparison method

In order to verify the effectiveness of the automatic neural network compression method using a combined pruning and quantization strategy, the present disclosure compares the automatic DNN compression method with the pruning method in the depth compression of the currently popular model compression method. Unlike the end-to-end framework employed by the present disclosure, this approach requires setting the clipping rate as a hyper-parameter.

The pruning method is described below. The method first learns the connection through normal network training. Next, the low weight connections are pruned, i.e. all connections with weights below a threshold are removed from the network. Finally, the network is retrained to learn the final weights of the remaining sparse connections. The pruning method reduces the number of parameters of the model.

Results and analysis of the experiments

In the experiment, the method adopted by the disclosure and the depth compression method for comparison are respectively used for training and testing on the AlexNet model, the training and testing data all adopt power equipment operation and detection image data sets, and equipment in different states is detected, as shown in fig. 5 and 6, the experimental results of the model compression multiple and the classification accuracy of the compressed model of the two methods are respectively displayed.

Fig. 5 shows that the compression factor for the AlexNet model using the automatic DNN compression method is significantly higher than that produced by the depth compression method.

And after the AlexNet model is compressed by adopting an automatic neural network compression method and a deep compression method respectively, training the compressed model by using the training set again, and testing the trained compressed model by using the testing set. As shown in fig. 6, the models compressed using the method of the present disclosure have higher classification accuracy than using the depth compression method.

By combining the analysis of fig. 5 and fig. 6, the compression method introduced by the present disclosure is more effective than the neural network compression method in the prior art, has higher compression multiple and accuracy, and is very suitable for being used in a model which needs to process and analyze a large amount of image data in an edge computing environment. Therefore, the method adopted by the disclosure is effective in the application based on the image target detection of the power equipment, and particularly, the size of the DNN model is reduced, the complexity of the model is reduced, and the load of the edge calculation is further reduced while the higher model prediction accuracy is achieved.

The method comprises the steps of carrying out training convergence on a loss function, and carrying out training convergence on the loss function by using a Newton method. The effectiveness and the practicability of the method for the neural network compression are verified through a large number of experiments. The method can be used for target detection edge service for processing and analyzing real-time power equipment defect image data, reduces the complexity of the model, reduces the load of edge calculation, and particularly reduces the size of the model while achieving higher model prediction accuracy.

Those skilled in the art will appreciate that all or part of the steps to implement the above embodiments are implemented as a computer program executed by a CPU. When executed by the CPU, performs the functions defined by the above-described methods provided by the present disclosure. The program may be stored in a computer readable storage medium, which may be a read-only memory, a magnetic or optical disk, or the like.

Furthermore, it should be noted that the above-mentioned figures are only schematic illustrations of the processes involved in the methods according to exemplary embodiments of the present disclosure, and are not intended to be limiting. It will be readily understood that the processes shown in the above figures are not intended to indicate or limit the chronological order of the processes. In addition, it is also readily understood that these processes may be performed synchronously or asynchronously, e.g., in multiple modules.

The following are embodiments of the disclosed apparatus that may be used to perform embodiments of the disclosed methods. For details not disclosed in the embodiments of the apparatus of the present disclosure, refer to the embodiments of the method of the present disclosure.

Fig. 7 is a block diagram illustrating a real-time monitoring apparatus for electrical equipment according to an exemplary embodiment. As shown in fig. 7, the power equipment real-time monitoring device 70 includes: an image module 702, a processing module 704, a calculation module 706, a monitoring module 708, and a model training module 710.

The image module 702 is configured to obtain a real-time image of the power device through the internet of things;

the processing module 704 is configured to pre-process the real-time image to generate image data;

the calculation module 706 is configured to input the image data into a target detection model, and generate a target detection result, where the target detection model is a deep neural network model that is based on a newton method to optimize a loss function for compression;

the monitoring module 708 is configured to monitor the state of the power device in real time based on the target detection result.

The model training module 710 is used for acquiring a plurality of historical images of the power equipment; preprocessing the plurality of historical images to generate a plurality of historical image data; training a deep neural network model based on the plurality of historical image data to generate an initial target detection model; and optimizing a loss function based on a Newton method to compress the initial target detection model to generate the target detection model.

According to the real-time monitoring device for the power equipment, a real-time image of the power equipment is obtained through the Internet of things; preprocessing the real-time image to generate image data; inputting the image data into a target detection model to generate a target detection result, wherein the target detection model is a deep neural network model which is based on a Newton method to optimize a loss function for compression; the mode of monitoring the state of the power equipment in real time based on the target detection result can arrange a high-precision and low-complexity calculation model in the edge calculation equipment so as to monitor the power equipment in real time, ensure the safe operation of a power grid and simultaneously reduce the data pressure of the Internet of things.

An electronic device 800 according to this embodiment of the disclosure is described below with reference to fig. 8. The electronic device 800 shown in fig. 8 is only an example and should not bring any limitations to the functionality and scope of use of the embodiments of the present disclosure.

As shown in fig. 8, the electronic device 800 is in the form of a general purpose computing device. The components of the electronic device 800 may include, but are not limited to: at least one processing unit 810, at least one memory unit 820, a bus 830 connecting the various system components (including the memory unit 820 and the processing unit 810), a display unit 840, and the like.

Wherein the storage unit stores program code that can be executed by the processing unit 810, such that the processing unit 810 performs the steps according to various exemplary embodiments of the present disclosure described in this specification. For example, the processing unit 810 may perform the steps as shown in fig. 2, 3, 4.

The memory unit 820 may include readable media in the form of volatile memory units such as a random access memory unit (RAM) 8201 and/or a cache memory unit 8202, and may further include a read only memory unit (ROM) 8203.

The memory unit 820 may also include a program/utility module 8204 having a set (at least one) of program modules 8205, such program modules 8205 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination thereof may comprise an implementation of a network environment.

Bus 830 may be any one or more of several types of bus structures including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

The electronic device 800 can also communicate with one or more external devices 800' (e.g., keyboard, pointing device, bluetooth device, etc.) such that a user can communicate with the devices with which the electronic device 800 interacts, and/or any device (e.g., router, modem, etc.) with which the electronic device 800 can communicate with one or more other computing devices. Such communication may occur over input/output (I/O) interfaces 850. Also, the electronic device 800 may communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network such as the internet) via the network adapter 860. The network adapter 860 may communicate with other modules of the electronic device 800 via the bus 830. It should be understood that although not shown in the figures, other hardware and/or software modules may be used in conjunction with the electronic device 800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, among others.

Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein may be implemented by software, or by software in combination with necessary hardware. Therefore, as shown in fig. 9, the technical solution according to the embodiment of the present disclosure may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (which may be a CD-ROM, a usb disk, a removable hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which may be a personal computer, a server, or a network device, etc.) to execute the above method according to the embodiment of the present disclosure.

The software product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The computer readable storage medium may include a propagated data signal with readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A readable storage medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out operations of the present disclosure may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., through the internet using an internet service provider).

The computer readable medium carries one or more programs which, when executed by a device, cause the computer readable medium to perform the functions of: acquiring a real-time image of the power equipment through the Internet of things; preprocessing the real-time image to generate image data; inputting the image data into a target detection model to generate a target detection result, wherein the target detection model is a deep neural network model which is based on Newton method optimization loss function for compression; and monitoring the state of the power equipment in real time based on the target detection result.

Those skilled in the art will appreciate that the modules described above may be distributed in the apparatus according to the description of the embodiments, or may be modified accordingly in one or more apparatuses unique from the embodiments. The modules of the above embodiments may be combined into one module, or further split into multiple sub-modules.

Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein may be implemented by software, or by software in combination with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (which may be a CD-ROM, a usb disk, a removable hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which may be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

Exemplary embodiments of the present disclosure are specifically illustrated and described above. It is to be understood that the present disclosure is not limited to the precise arrangements, instrumentalities, or instrumentalities described herein; on the contrary, the disclosure is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A real-time monitoring method for power equipment is characterized by comprising the following steps:

acquiring a real-time image of the power equipment through the Internet of things;

preprocessing the real-time image to generate image data;

inputting the image data into a target detection model to generate a target detection result, wherein the target detection model is a deep neural network model which is based on a Newton method to optimize a loss function for compression;

and monitoring the state of the power equipment in real time based on the target detection result.

2. The real-time monitoring method for the power equipment as claimed in claim 1, further comprising:

acquiring a plurality of historical images of the power equipment;

preprocessing the plurality of history images to generate a plurality of history image data;

training a deep neural network model based on the plurality of historical image data to generate an initial target detection model;

and optimizing a loss function based on a Newton method to compress the initial target detection model to generate the target detection model.

3. The power equipment real-time monitoring method of claim 2, wherein preprocessing the plurality of historical images to generate a plurality of historical image data comprises:

performing data cleansing on the plurality of historical images to generate a plurality of historical image data; and/or

Performing data conversion on the plurality of history images to generate a plurality of history image data; and/or

And performing data normalization processing on the plurality of historical images to generate a plurality of historical image data.

4. The power device real-time monitoring method of claim 2, wherein optimizing a loss function based on newton's method to compress the initial target detection model to generate the target detection model comprises:

generating an exponential loss function;

generating a compression function;

performing optimization calculation on the exponential loss function based on a Newton method to obtain a minimized exponential loss function and corresponding model parameters thereof;

decoupling the compression function to obtain weights of the compressed target detection model;

generating the target detection model based on the model parameters and the weights.

5. The method of real-time monitoring of power equipment of claim 4, wherein generating an exponential loss function comprises:

acquiring initialization parameters, wherein the initialization parameters comprise a compression target, algorithm iteration times and sample iteration times;

generating the exponential-loss function based on the initialization parameter.

6. The power device real-time monitoring method of claim 4, wherein generating a compression function comprises:

acquiring an original weight tensor of the initial target detection model;

generating the compression function based on the original weight tensor.

7. The method for monitoring the power equipment in real time as claimed in claim 4, wherein the optimization calculation of the exponential loss function based on Newton's method to obtain the minimized exponential loss function and the corresponding model parameters thereof comprises:

determining an iteration initial value;

determining a searching direction;

performing iterative calculations based on the search direction to solve a minimized exponential function;

determining the model parameters based on the minimization exponential function.

8. The real-time monitoring method for the power equipment as claimed in claim 7, wherein the determining of the search direction comprises:

and performing linear transformation on the gradient through a Hessian matrix based on the iteration initial value to obtain a search direction.

9. The power equipment real-time monitoring method of claim 4, wherein decoupling the compression function to obtain the weights of the compressed target detection model comprises:

determining a dual variable;

decoupling the compression function based on the dual variable and an alternating direction multiplier method;

calculating the decoupled compression function based on iterative calculation to obtain the weight of the target detection model.

10. The utility model provides an electrical equipment real-time supervision device which characterized in that includes:

the image module is used for acquiring a real-time image of the power equipment through the Internet of things;

the processing module is used for preprocessing the real-time image to generate image data;

the calculation module is used for inputting the image data into a target detection model to generate a target detection result, wherein the target detection model is a deep neural network model which is compressed based on a Newton method optimization loss function;

and the monitoring module is used for monitoring the state of the power equipment in real time based on the target detection result.

11. The real-time monitoring device for electric power equipment according to claim 10, further comprising:

the model training module is used for acquiring a plurality of historical images of the power equipment; preprocessing the plurality of historical images to generate a plurality of historical image data; training a deep neural network model based on the plurality of historical image data to generate an initial target detection model; and optimizing a loss function based on a Newton method to compress the initial target detection model to generate the target detection model.

12. An electronic device, comprising:

one or more processors;

storage means for storing one or more programs;

when executed by the one or more processors, cause the one or more processors to implement the method of any one of claims 1-9.

13. A computer-readable medium, on which a computer program is stored, which, when being executed by a processor, carries out the method according to any one of claims 1-9.