Non-intrusive digital power grid asset identification method and system based on deep learning

Abstract

The invention discloses a non-intrusive digital power grid asset identification method and system based on deep learning, and relates to the technical field of asset security management and control, and the method comprises the steps: collecting network flow data, and carrying out the data preprocessing of original data; building a first model based on the network traffic data; and performing asset decision on the equipment through a first method. According to the non-intrusive digital power grid asset identification method based on deep learning provided by the invention, by passively monitoring the network traffic instead of actively sending the packet, the extra burden and security risk caused to the operation of the power grid are avoided, the recurrent neural network RNN is selected as a core model, the time dependence and periodic change in the network traffic are effectively captured, and the identification accuracy of the assets of the power grid is improved. Compared with the prior art, the method has the advantages that strong technical support is provided for asset identification, the One-Hot coding feature engineering technology is adopted, the generalization ability of the model is improved, the Moore voting mechanism is introduced to carry out final asset decision making, convenience and flexibility are improved, and better effects are achieved in the aspects of safety, generalization ability and flexibility.

Description

Technical Field

[0001] The present invention relates to the technical field of asset security management and control, and specifically to a non-intrusive digital power grid asset identification method based on deep learning. Background Art

[0002] In recent years, with the acceleration of informatization and digital transformation of the power industry, digital grids have gradually become an important part of the energy Internet. Traditional power grids have improved their data collection, transmission and processing capabilities by introducing virtualization technology, private cloud deployment and big data applications. However, with the rapid increase in power grid equipment and network assets, the complexity of power grid operation and management has increased significantly. Especially in the field of power grid asset management, how to accurately identify and manage network equipment has become an important issue to ensure the safe and stable operation of the power grid. Traditional asset identification technologies are mostly based on active detection methods, such as obtaining asset information by actively sending packets to interact with devices. However, this method not only increases the network load, but may also cause potential safety hazards. In order to address these problems, deep learning technology has been introduced in recent years. By building an intelligent asset description model, the intelligent development of digital power grid asset identification has been realized, providing a new technical path for efficient management of the power grid.

[0003] Although current technology has made certain progress in the field of power grid asset identification, it still has obvious shortcomings in many aspects and cannot meet the needs of efficient identification in complex scenarios of digital power grids. Traditional methods mostly use intrusive detection methods, which pose a potential threat to the normal operation of the network, especially in scenarios with high load or strict security requirements. It is easy to cause network performance degradation and security risks. Existing asset identification technology is relatively rough in data processing and modeling. Most of them are based on simple rules or static models and cannot flexibly respond to complex scenarios with dynamic changes in assets. The recognition accuracy and generalization ability are limited. In terms of intelligent decision support, traditional methods usually lack targeted analysis and optimization, and it is difficult to provide customized asset identification solutions according to the needs of different scenarios. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problems solved by the present invention are: the existing asset identification methods have the problems that the identification models lack flexibility and dynamic adaptability, the intrusive detection causes burden and security risks to the network, and how to optimize the non-intrusive, intelligent and precise identification of digital power grid assets.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a non-intrusive digital power grid asset identification method based on deep learning, including collecting network traffic data and performing data preprocessing on the original data; building a first model based on the network traffic data; and making asset decisions for the equipment through the first method.

[0007] As a preferred solution of the deep learning-based non-intrusive digital power grid asset identification method described in the present invention, the collection of network traffic data includes capturing the network traffic data of the power grid equipment by monitoring the device hub or switch.

[0008] As a preferred solution of the deep learning-based non-intrusive digital power grid asset identification method described in the present invention, the data preprocessing of the original data includes preliminary cleaning of the acquired original network traffic data.

[0009] As a preferred solution of the deep learning-based non-intrusive digital power grid asset identification method described in the present invention, the data preprocessing of the original data also includes extracting and encoding features in the network traffic data.

[0010] As a preferred solution of the deep learning-based non-intrusive digital power grid asset identification method described in the present invention, wherein: the building of the first model based on network traffic data includes building a first model based on the traffic data of the power grid assets.

[0011] As a preferred solution of the deep learning-based non-intrusive digital power grid asset identification method described in the present invention, wherein: the building of the first model based on network traffic data also includes training the first model by using power grid asset traffic.

[0012] As a preferred solution of the deep learning-based non-intrusive digital power grid asset identification method described in the present invention, wherein: the asset decision-making for the device through the first method includes obtaining a network traffic determination result.

[0013] Another object of the present invention is to provide a non-intrusive digital power grid asset identification system based on deep learning, which can determine whether the flow data comes from the target asset through a model building module, solving the current problem of being unable to dynamically adapt to changes in the network environment.

[0014] As a preferred solution of the non-intrusive digital power grid asset identification system based on deep learning described in the present invention, it includes: an acquisition and processing module, a model building module, and an asset decision module; the acquisition and processing module is used to collect network traffic data and perform data preprocessing on the original data; the model building module is used to build a first model based on the network traffic data; the asset decision module is used to make asset decisions on the equipment through a first method.

[0015] A computer device includes a memory and a processor, wherein the memory stores a computer program, and wherein the processor executes the computer program to implement a step of a non-intrusive digital power grid asset identification method based on deep learning.

[0016] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of a non-intrusive digital power grid asset identification method based on deep learning are implemented.

[0017] Beneficial effects of the present invention: The non-intrusive digital power grid asset identification method based on deep learning provided by the present invention avoids additional burden and security risks on power grid operation by passively listening to network traffic rather than actively sending packets. The recurrent neural network RNN ​​is selected as the core model to effectively capture the time dependency and periodic changes in network traffic, providing strong technical support for asset identification. The One-Hot coding feature engineering technology is adopted to improve the generalization ability of the model, and the Moore voting mechanism is introduced to make the final asset decision, which improves convenience and flexibility. The present invention achieves better results in terms of security, generalization ability and flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0019] Figure 1 An overall flow chart of the deep learning-based non-intrusive digital power grid asset identification method provided for the first and second embodiments of the present invention.

[0020] Figure 2 RNN structure diagram of the deep learning-based non-intrusive digital power grid asset identification method provided in the second embodiment of the present invention.

[0021] Figure 3 An RNN recurrent connection diagram of a deep learning-based non-intrusive digital power grid asset identification method provided in the second embodiment of the present invention.

[0022] Figure 4 An asset feature description model diagram of a non-intrusive digital power grid asset identification method based on deep learning provided in the second embodiment of the present invention.

[0023] Figure 5 A deep learning model operation diagram of a deep learning-based non-intrusive digital power grid asset identification method provided in the second embodiment of the present invention.

[0024] Figure 6 An overall module diagram of a deep learning-based non-intrusive digital power grid asset identification system provided for the fourth embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0026] Example 1

[0027] Reference Figure 1 , which is an embodiment of the present invention, provides a non-intrusive digital power grid asset identification method based on deep learning, comprising:

[0028] S1: Collect network traffic data and perform data preprocessing on the original data.

[0029] Furthermore, collecting network traffic data includes capturing network traffic data of power grid devices by monitoring device hubs or switches.

[0030] It should be noted that by intercepting the network bypass, capturing and analyzing the network traffic data of the power grid equipment, without making any physical or logical changes to the existing network, it will not affect the network of the digital power grid, nor the normal operation of the equipment. Under the premise of not affecting the normal operation of the network, real-time network traffic information can be obtained to provide data support for subsequent deep learning analysis. By configuring monitoring equipment, copying network traffic, and processing the collected traffic data, the status of the power grid equipment can be monitored and identified.

[0031] Deploy monitoring device hubs or switches at key nodes in the network. The monitoring device needs to be configured in bypass mode. By setting up a shared hub or mirror switch, after the monitoring device is configured, when the power grid devices communicate with each other, all network traffic passing through the hub or mirror switch will be copied and sent to the monitoring host through the mirror port. This copying process is completely transparent and will not have any impact on the actual network traffic, ensuring that the entire process is non-intrusive and does not cause additional burden on the network. After the monitoring host receives the copied network traffic data, it inputs it into the processing unit configured later.

[0032] Furthermore, data preprocessing of the raw data includes preliminary cleaning of the acquired raw network traffic data.

[0033] It should be noted that the acquired raw network traffic data is preliminarily cleaned to filter out noise data, duplicate data, and irrelevant data packets to ensure the quality and consistency of the input data. Users can customize data rules or screening criteria, such as using the IP address of an existing asset as the destination IP address for traffic data screening.

[0034] Furthermore, data preprocessing of the raw data also includes extracting and encoding features in the network traffic data.

[0035] It should be noted that for non-continuous features in network traffic data, such as IP addresses, MAC addresses, network protocol types, etc., One-Hot encoding is used to convert different IP addresses or protocol types into binary vectors. As a commonly used encoding method in machine learning and data processing, it is mainly used to convert discrete data into numerical form so that it can be input into the model for training. In One-Hot encoding, it is assumed that a categorical variable X has k different categories, and each category i corresponds to a category value X i , where i∈{1,2,…,k}, for each sample X, convert it into a k-dimensional one-hot vector v. For a sample X=x i , its One-Hot encoding vector v is expressed as:

[0036] v=[v1,v2,…,v k ]

[0037]

[0038] Among them, v j is the value of the jth element in the vector v. The continuous features in the network traffic data, such as packet size, traffic length, time interval, etc., are properly retained and normalized, and the value range of these features is scaled to a suitable range through standardization or normalization methods to reduce the scale difference between the features, so that the deep learning model can better learn the relationship between these features, which is expressed as:

[0039]

[0040] The encoded non-continuous features are combined with the retained continuous features to form a complete feature vector as the input of the deep learning model. The preprocessed data is formatted and converted into CSV file format to ensure that it meets the input requirements of the deep learning model.

[0041] Example 2

[0042] Reference Figure 1-Figure 5, which is an embodiment of the present invention, provides a non-intrusive digital power grid asset identification method based on deep learning, comprising:

[0043] S2: Build the first model based on network traffic data.

[0044] Furthermore, building the first model based on the network traffic data includes building the first model based on the traffic data of the power grid assets.

[0045] In the embodiment of the present application, the first model is an asset feature description model, and a deep learning model is constructed by a recurrent neural network RNN. Before the emergence of RNN, traditional neural networks could only deal with each input in isolation when processing data, and the correlation between the previous and next inputs could not be reflected in the traditional model. In reality, many tasks require an in-depth understanding of the context of sequence information, which requires the network to capture and understand the order and dependencies between inputs. RNN uses a unique loop structure to enable information to be transmitted and accumulated within the network, effectively processing data with contextual characteristics. In power grid flow data, the communication between devices is usually time-series. RNN can remember and utilize past input information through its loop structure, so as to better understand the dynamic changes of flow and the interaction mode between devices. Therefore, RNN is suitable for flow data with contextual relationships in digital power grids;

[0046] A simple RNN structure consists of an input layer, a hidden layer, and an output layer. The input layer has N input units. The input of this layer is a vector sequence {X1, X2, ..., X t}, where X t =(x1,x2,...,x N ), the input units in a fully connected RNN are connected to the hidden units in the hidden layer, where the connections are represented by a weight matrix W IH Definition: The hidden layer has M hidden units H t =(h1,h2,...,h M ), the hidden units are connected to each other on a time scale through recurrent connections, and the hidden layer defines the state space of the system as:

[0047] H t =f H (o t )

[0048] o t =W IH X t +W HH H t +b t

[0049] Among them, f His the activation function of the hidden layer, b t is the bias vector of the hidden unit, and the hidden unit is weighted by the weight matrix W HO Connected to the output layer, the output layer contains P units, expressed as:

[0050] Y T =f o (W HO H t +b o )

[0051] Y T =(y1,y2,...,y P )

[0052] Among them, f o and b o are the activation function and bias vector of the output layer respectively. Since the input and output are continuous in time, the above steps are repeated at time t. The formula proves that an RNN consists of certain nonlinear state equations, which are iterated over time. At each time step, the hidden state provides a prediction of the output layer based on the input vector. The hidden state of the RNN is a set of values. In addition to being affected by any external factors, it also summarizes all the only necessary information about the past state of the network over multiple time steps. This integrated information can define the future behavior of the network and make accurate predictions at the output layer.

[0053] In an optional embodiment, the first model may also be implemented in other ways, such as by using an asset feature description model that combines clustering and classification, performing preliminary clustering of traffic data by an unsupervised learning method, and then further verifying and refining the clustering results by using a supervised classification model;

[0054] After preprocessing, the collected network traffic data is input into the K-Means or DBSCAN (density clustering) based algorithm. The clustering algorithm groups the traffic according to traffic characteristics, such as protocol type, packet size, time interval, etc., and initially forms a feature classification of assets. Data can be analyzed without prior labels, and potential hidden patterns can be discovered, providing strong support for subsequent classification.

[0055] In order to improve the computational efficiency and generalization ability of the model, feature selection and dimensionality reduction are performed on the clustering results. Principal component analysis (PCA) or feature embedding technology is used to extract the key features of traffic data, reduce the data dimension, and retain the most discriminative information.

[0056] On the basis of clustering, supervised classification models are introduced, such as support vector machine SVM or gradient boosting tree GBDT, to further classify and verify the data. The training data of the classification model comes from the labeled asset samples. Combined with the clustering results of the previous step, the model can more accurately identify the characteristics of each type of asset. Through cross-validation and hyperparameter optimization, the robustness and efficiency of the model are ensured. The hybrid model outputs the asset category and the corresponding feature description results, and identifies each category of equipment assets.

[0057] Furthermore, building the first model based on network traffic data also includes training the first model by using power grid asset traffic.

[0058] It should be noted that based on the RNN model, by using known power grid asset traffic to train the RNN, the model automatically learns multi-dimensional features such as IP address, port information, traffic size and time interval, thereby improving the ability to describe asset features, designing and building an asset feature description model dedicated to the power grid, and because the model used is based on deep learning technology, it can often mine more complete information for modeling, and by adjusting the activation function of the output layer, the model can output clear binary classification results, namely positive examples (source IP device is an asset) and negative examples (source IP device is not an asset);

[0059] The asset feature description model is mainly based on the preprocessed power grid flow data and is trained using the RNN model. It aims to identify and determine whether the source IP device of the network flow data packet is a power grid asset. By designing the structure of the model and the activation function of the output layer, the model can effectively distinguish between positive and negative examples and achieve accurate identification of power grid assets.

[0060] S3: Make asset decisions for the equipment using the first method.

[0061] Furthermore, making an asset decision for the device through the first method includes obtaining a network traffic determination result.

[0062] In the embodiment of the present application, the first method is Moore's voting method, which is a simple and effective decision-making method. The flow data is predicted by the asset feature description model to obtain the judgment results of multiple network flows sent by each device. These results include two categories: belonging to assets or not belonging to assets;

[0063] The final asset decision is determined by voting on the judgment results of all flow data of a certain device. For all flow judgment results of the device, the number of judgments that belong to assets and the number of judgments that do not belong to assets are counted, and then the final asset category of the device is determined by majority voting. If the number of votes belonging to assets exceeds half, the device is judged as an asset, otherwise, it is judged as a non-asset.

[0064] Based on Moore's voting, decisions are adjusted according to the user's adjustments. Since the deep learning model may have a small number of misjudgments, a decision factor is set. When the user wants to discover more grid assets with higher certainty, the rules of Moore's voting can be changed, and the decision factor for determining the assets can be increased accordingly. If the number of votes belonging to an asset exceeds 80%, it is assumed that the device is an asset of the digital grid. At this time, the decision factor is 0.8, so that potential grid equipment can be discovered with greater certainty.

[0065] In an optional embodiment, the first method may also be implemented in other ways, such as an asset decision method based on a weighted scoring mechanism, and a final judgment of the equipment assets is made by a method based on a weighted scoring mechanism;

[0066] Assign corresponding weights to each network traffic data according to its characteristics and sources. Traffic involving key protocols, such as TCP or UDP, or traffic with high packet frequency is assigned higher weights, while data originating from lower priority ports is assigned lower weights. The weight values ​​can be determined by historical data analysis or confidence prediction values ​​of deep learning models.

[0067] Calculate the comprehensive score for each device’s traffic data;

[0068] The system presets an asset determination threshold. If the comprehensive score of a device exceeds the threshold, it is determined to be an asset. Otherwise, it is determined to be a non-asset. The threshold can be adjusted dynamically and can be customized according to the network security level, device importance or identification requirements. In high-security scenarios, the threshold can be increased to ensure the reliability of the determination results.

[0069] The scoring mechanism supports dynamic adjustment. When it is found that the scores of some devices are close to the threshold, the system can increase the adjustment range of the weight according to the real-time traffic characteristics or introduce additional features as auxiliary judgment basis to improve the accuracy of the results.

[0070] Example 3

[0071] One embodiment of the present invention provides a non-intrusive digital power grid asset identification method based on deep learning. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0072] The experiment simulated different power grid operation scenarios, including real-time flow monitoring and historical data analysis, and compared the performance of the existing technology (intrusive detection and rule-based methods) with the method of the present invention (non-intrusive detection and deep learning model). Through the non-intrusive flow monitoring unit, real-time and historical flow data were collected from the key nodes of the test network. The experiment used a shared hub to bypass the flow to ensure that the collection process did not interfere with the network operation. The flow data covered multi-dimensional features such as protocol type (TCP, UDP), packet size, IP address and time interval. The collected data was feature extracted and encoded. Non-continuous variables (such as protocol type and IP address) were The preprocessed data is stored in a standardized format for model training. The model is trained using a recurrent neural network (RNN) using the preprocessed data. Two sets of data are used in the training process, one for training the model and the other for verifying the generalization ability of the model. The model training parameters include training rounds (200 rounds) and batch size (32). The model is deployed in the experimental network, and real-time traffic and historical data are collected for asset identification. The performance of the model in terms of identification accuracy, false alarm rate, and processing delay is tested and compared.

[0073] As shown in Table 1, the accuracy of the traditional intrusive and rule-based methods in real-time monitoring and historical data analysis is 78.5% and 85.1% respectively, while the non-intrusive method and deep learning model of the present invention reach 92.3% and 94.5% respectively, indicating that the deep learning model can capture asset features more efficiently and significantly improve recognition accuracy. The false alarm rate of the traditional method is high, while the method of the present invention reduces the false alarm rate to 4.7% and 3.1% through the deep learning model. This advantage stems from the model's deep mining and intelligent judgment capabilities for traffic features. In terms of data processing time, The method of the present invention reduces the delay in traditional methods, and the non-invasive method only takes 8.6ms, which proves that it has significant advantages in real-time performance. Due to the characteristics of active detection, traditional methods impose a certain burden on network operation, and their impact scores are 3.5 and 2.9 respectively. The passive collection method of the method of the present invention reduces the impact to 0.8 and 0.6, ensuring the stability and security of network operation. The method of the present invention significantly improves the credibility of decision-making by introducing an intelligent decision-making mechanism, from 82.4% and 88.9% of traditional methods to 96.7% and 98.3%.

[0074] Table 1 Experimental data table

[0075] Traditional methods Method of the present invention Traditional methods Method of the present invention Traffic Type Real-time monitoring Real-time monitoring Historical data analysis Historical data analysis Detection accuracy 78.5％ 92.3％ 85.1％ 94.5％ False Positive Rate 15.6％ 4.7％ 9.3％ 3.1％ Data processing time 12.4ms 8.6ms 10.2ms 7.5ms Model training time 45min 38min 50min 35min Network Impact Score 3.5 0.8 2.9 0.6 Decision Credibility 82.4％ 96.7％ 88.9％ 98.3％

[0076] Example 4

[0077] Reference Figure 6, is an embodiment of the present invention, and provides a non-intrusive digital power grid asset identification system based on deep learning, including: an acquisition and processing module, a model building module, and an asset decision module.

[0078] The collection and processing module is used to collect network traffic data and perform data preprocessing on the original data; the model building module is used to build a first model based on the network traffic data; and the asset decision module is used to make asset decisions on the equipment through a first method.

[0079] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0080] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0081] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0082] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0083] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A non-intrusive digital power grid asset identification method based on deep learning, characterized in that: include: Collect network traffic data and pre-process the raw data; Build the first model based on network traffic data; The first method is used to make asset decisions for equipment.

2. The non-intrusive digital power grid asset identification method based on deep learning according to claim 1, characterized in that: The collecting of network traffic data includes capturing the network traffic data of the power grid equipment by monitoring a device hub or a switch.

3. The non-intrusive digital power grid asset identification method based on deep learning according to claim 2, characterized in that: The data preprocessing of the original data includes preliminary cleaning of the acquired original network traffic data.

4. The non-intrusive digital power grid asset identification method based on deep learning as claimed in claim 3, characterized in that: The data preprocessing of the original data also includes extracting and encoding features in the network traffic data.

5. The non-intrusive digital power grid asset identification method based on deep learning according to claim 4, characterized in that: The building of the first model based on the network traffic data includes building the first model based on the traffic data of the power grid assets.

6. The non-intrusive digital power grid asset identification method based on deep learning according to claim 1, characterized in that: The step of building the first model based on network traffic data also includes training the first model by using grid asset traffic.

7. The non-intrusive digital power grid asset identification method based on deep learning according to claim 6, characterized in that: The making of asset decision on the device by the first method includes obtaining a network traffic determination result.

8. A system using the non-intrusive digital power grid asset identification method based on deep learning as claimed in any one of claims 1 to 7, characterized in that: It includes collection and processing module, model building module and asset decision module; The acquisition and processing module is used to collect network traffic data and perform data preprocessing on the original data; The model building module is used to build a first model based on network traffic data; The asset decision module is used to make asset decisions on equipment through a first method.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the deep learning-based non-intrusive digital grid asset identification method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the deep learning-based non-intrusive digital grid asset identification method described in any one of claims 1 to 7 are implemented.

Images