A construction method of a lightweight artificial intelligence algorithm library for distribution transformer areas

Abstract

The present invention proposes a method for constructing a lightweight artificial intelligence algorithm library for a distribution transformer area, including the following steps: Step S1, reducing the historical operation scenarios of the transformer area based on dimensionality reduction and clustering algorithms, and extracting representative typical operation scenarios of the transformer area; Step S2, by recording the real-time operation data of edge devices when artificial intelligence algorithms are applied in the low-voltage distribution transformer area, constructing the matching relationship between the operation state of the transformer area and the application of each artificial intelligence algorithm, and further realizing the matching between the artificial intelligence algorithm and the operation scenario of the transformer area; Step S3, considering the data transmission bandwidth, internal storage space size, and calculation speed differences of each fusion terminal, and on the premise of ensuring sufficient accuracy, using artificial intelligence algorithm model simplification technology to construct lightweight artificial intelligence algorithms, forming a lightweight artificial intelligence algorithm library; The present invention can simplify artificial intelligence algorithms to solve the limitations and compatibility problems of deploying artificial intelligence algorithms in transformer area fusion terminals.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent distribution networks, and in particular to a method for constructing a lightweight artificial intelligence algorithm library for distribution substations. Background Art

[0002] Currently, the distribution network is transforming from an uncontrollable traditional distribution network to a partially controllable modern distribution network and a fully controllable future distribution network. The development goal of the distribution network is also gradually evolving from the geographical full coverage of user loads to reliable, efficient, and high-quality power supply services, to a green, low-carbon, intelligent, controllable, and supply-demand interactive service platform. The distribution network has also gradually evolved from the original power transmission and customer service to the stage of a distribution Internet of Things that comprehensively perceives, fuses data, and applies intelligence. Under this background, the State Grid Equipment Department has proposed intelligent fusion terminal technology for low-voltage distribution substations, and accelerated the construction of a low-voltage distribution Internet of Things system with fusion terminals as the core. By expanding the coverage of distribution transformer intelligent fusion terminals in substations, the scale effect is exerted to meet the current needs of grass-roots power supply stations and operation and maintenance teams for effective management and immediate perception of medium- and low-voltage distribution networks, and to improve the effective response ability of low-voltage distribution substations to the massive access of distributed photovoltaic, energy storage and other devices, as well as the sudden increase in electric vehicle charging capacity, facilitating the "operation, inspection, emergency repair, and storage" business of grass-roots service personnel, improving the operation efficiency of low-voltage distribution substations, and enhancing the edge coordination ability of substation resources.

[0003] With the advancement of the practical application of fusion terminals, based on fusion terminals and combined with existing edge devices in each substation, the ability to perceive and analyze loads can be greatly improved, providing the possibility for constructing a new form of regional autonomy for the distribution network. For a long time, the construction of the power grid has mainly focused on equipment management, lacking analysis and management of distribution network operation data. Moreover, the internal data structure of the power grid is complex, with a large amount of data that is difficult to integrate, which also leads to great difficulties in data analysis and collaborative control in low-voltage distribution substations to a certain extent. However, through the advancement of the practical application of fusion terminals, the data of a large number of edge devices in substations can be collected and aggregated into the fusion terminals. It has become possible to analyze, identify, and manage the diversified loads in low-voltage distribution substations through the fusion terminals as an edge computing platform, laying a technical foundation for realizing regional autonomy of the distribution network.

[0004] However, the application of artificial intelligence algorithms, scenario analysis, management, and resource collaborative control mainly based on the integrated terminal of the transformer substation area still face many problems. First, with the continuous access of edge devices such as the photovoltaic energy storage charging system and IoT switches in the transformer substation area, it is difficult to comprehensively analyze the massive low-voltage data. It is urgent to extract representative and typical data from the massive low-voltage data for modeling, so as to further carry out more advanced function applications. Second, traditional artificial intelligence algorithms often face the server side. With sufficient computing power resources and storage space, the accuracy and efficiency of the algorithms are guaranteed, which is in a certain conflict with the computing resources of the integrated terminal of the transformer substation area. Taking the State Grid integrated terminal as an example, the main frequency of its core CPU is not less than 700MHz, lower than the common 2GHz of the server; its memory is not less than 512MB, and the FLASH is not less than 4GB, which is far from the cloud server with dozens of GB of memory and thousands of GB of FLASH. It can be seen that the traditional artificial intelligence algorithm model is too large to be deployed and needs to be simplified to a certain extent before it can be deployed to the integrated terminal.

[0005] Therefore, aiming at the problem of certain limitations in deploying artificial intelligence algorithms for the integrated terminal of the transformer substation area, a method for constructing a lightweight artificial intelligence algorithm library for distribution transformer substations is proposed. Summary of the Invention

[0006] The present invention proposes a method for constructing a lightweight artificial intelligence algorithm library for distribution transformer substations, which can simplify artificial intelligence algorithms and is used to solve the limitations and compatibility problems of deploying artificial intelligence algorithms for the integrated terminal of the transformer substation area.

[0007] The present invention adopts the following technical solutions.

[0008] A method for constructing a lightweight artificial intelligence algorithm library for distribution transformer substations includes the following steps;

[0009] Step S1: Considering the uncertainty of the load of edge devices in the distribution transformer substation area, reduce the historical operation scenarios of the transformer substation area based on dimensionality reduction and clustering algorithms, and extract representative typical operation scenarios of the transformer substation area;

[0010] Step S2: By recording the real-time operation data of edge devices when artificial intelligence algorithms are applied in the low-voltage distribution transformer substation area, construct the matching relationship between the operation state of the transformer substation area and the application of each artificial intelligence algorithm, and then realize the matching between the artificial intelligence algorithm and the operation scenario of the transformer substation area;

[0011] Step S3: Considering the data transmission bandwidth, internal storage space size, and calculation speed differences of each integrated terminal, on the premise of ensuring sufficient accuracy, use artificial intelligence algorithm model simplification technology to construct lightweight artificial intelligence algorithms and form a lightweight artificial intelligence algorithm library.

[0012] Specifically, step S1 is as follows: fully considering the uncertain characteristics of distributed photovoltaic output, the charging characteristics of electric vehicle charging piles, and the charge and discharge characteristics of energy storage in the low-voltage distribution substation area, using the daily load data of the distribution transformer and the photovoltaic-storage-charging as input and output, using the autoencoder AE to extract the dimensionality reduction features of the load data, and using the dimensionality reduction features of the load data to reduce the historical operation scenarios of the substation area based on the K-medoids clustering algorithm, extracting representative typical operation scenarios of the substation area. At the same time, combining the external environment data in the operation process of power enterprises, using the above-mentioned typical scenario construction method of the substation area, establishing typical scenarios of the distribution substation area under different influencing factors, and establishing a typical operation scenario library for the low-voltage distribution substation area.

[0013] The external environment data includes ambient temperature and weather characteristics;

[0014] In step S1, for the complex multi-source heterogeneous massive data generated in the operation process of power enterprises, technologies such as source data verification, source data error correction, and data reduction are used to fuse and integrate the multi-source heterogeneous data, and the massive measurement data of multiple power grid systems are connected. At the same time, based on the integrated multi-source data, combined with the influencing factors of ambient temperature and season, clustering analysis, classification analysis, association analysis, and regression analysis methods are used to analyze the changes in the typical operation scenarios of the substation area under different influencing factors, and the construction of typical operation scenarios of the distribution network under different influencing factors is realized.

[0015] Step S2 includes: combining the measurement data generated during the operation of the low-voltage distribution substation area, recording the real-time operation data of the substation area edge devices when executing the artificial intelligence algorithm, and constructing the matching relationship between the substation area operation state and the application of each artificial intelligence algorithm;

[0016] When the artificial intelligence algorithm matches multiple different operation states of the substation area, by performing AE feature extraction and K-medoids clustering of the dimensionality reduction features on the multiple operation state data, reducing the number of substation area operation states matching the artificial intelligence algorithm, and finding the typical scenario of the substation area that is most similar to each operation state of the artificial intelligence algorithm in the substation area, the matching between the artificial intelligence algorithm and the typical scenario of the substation area is completed, and its objective function is:

[0017]

[0018] Among them, S j is the measurement data of the jth substation area operation state after reduction corresponding to the artificial intelligence algorithm, k is the total number of reduced substation area operation states, S ci is the operation data of the ith established typical substation area operation scenario, and m is the total number of established typical substation area operation scenarios.

[0019] The substation area edge devices include, for example, photovoltaic, energy storage, charging piles, and Internet of Things switch devices;

[0020] In step S2, by using the multi-source massive data stored during the operation of the low-voltage distribution substation area, and recording the real-time operation data of the substation area edge devices when applying algorithms such as heavy overload prediction, photovoltaic power output prediction, energy storage coordination, and source following load movement, the matching relationship between the substation area operation scenarios and the applications of various artificial intelligence algorithms is constructed.

[0021] When a specific artificial intelligence algorithm matches multiple different scenarios in the substation area, AE is used to extract features from all the matching scenarios, and the big data analysis method of K-medoids clustering is used to reduce the number of substation area operation scenarios matching the artificial intelligence algorithm. The substation area operation scenarios corresponding to the artificial intelligence algorithm and the established typical substation area scenarios use Euclidean distance and Pearson correlation coefficient techniques to achieve the matching between the artificial intelligence algorithm and the typical substation area scenarios.

[0022] The specific content of step S3 is as follows: After the training of the traditional artificial intelligence algorithm is completed, considering the data transmission bandwidth, internal storage space size, calculation speed difference, and system environment of each fusion terminal, on the premise of ensuring sufficient accuracy, the simplification technology of the artificial intelligence algorithm model is used to construct a lightweight artificial intelligence algorithm based on the trained traditional artificial intelligence algorithm, forming a lightweight artificial intelligence algorithm library.

[0023] In step S3, when constructing the lightweight artificial intelligence algorithm library, the specific method is: On the premise of completing the construction of the typical substation area operation scenarios, the analysis of the application requirements of the artificial intelligence algorithm, and the collation of the information of the substation area fusion terminal devices, select an artificial intelligence algorithm application for development;

[0024] Use the artificial intelligence algorithm model simplification technology to develop a lightweight artificial intelligence algorithm on the host computer, and use the historical load data of the edge device as offline data to debug the algorithm until the result obtained by the algorithm is close to the actual operation result;

[0025] When there are differences between the host computer development environment and the target embedded machine, which may cause the transplanted algorithm to be unable to run or have low running efficiency, then the algorithm is modified and tested specifically. According to the corresponding development environment, a deployable lightweight artificial intelligence algorithm program is formed through the programming language, and the program is packaged using the docker containerization technology to achieve rapid deployment and partition isolation from other computing programs, so that the model can quickly sink to the fusion terminal;

[0026] According to the above steps, the deployment of the lightweight artificial intelligence algorithm application in the substation area is completed in an orderly manner. Under the condition that the operation result of the algorithm meets the expectation, it is incorporated into the lightweight artificial intelligence algorithm library; more other artificial intelligence algorithms are still developed according to the above steps and incorporated into the algorithm library to complete the construction of the lightweight artificial intelligence algorithm library.

[0027] The present invention can simplify artificial intelligence algorithms and is used to solve the limitations and compatibility problems of deploying artificial intelligence algorithms in substation integration terminals. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The present invention will be further described in detail below with reference to the drawings and specific embodiments:

[0029] FIG Figure 1 is a schematic flowchart of a method for constructing a lightweight artificial intelligence algorithm library for a distribution substation area;

[0030] FIG Figure 2 is a schematic comparison diagram of the line loss rates of the substation areas with and without the algorithm executed by the integration terminal. SPECIFIC EMBODIMENTS

[0031] As shown in the figure, a method for constructing a lightweight artificial intelligence algorithm library for a distribution substation area includes the following steps;

[0032] Step S1: Considering the uncertainty of the load of edge devices in the distribution substation area, the historical operation scenarios of the substation area are reduced based on dimensionality reduction and clustering algorithms, and representative typical operation scenarios of the substation area are extracted;

[0033] Step S2: By recording the real-time operation data of edge devices when artificial intelligence algorithms are applied in the low-voltage distribution substation area, the matching relationship between the operation state of the substation area and the application of each artificial intelligence algorithm is constructed, and then the matching between the artificial intelligence algorithm and the operation scenario of the substation area is realized;

[0034] Step S3: Considering the data transmission bandwidth, internal storage space size, and calculation speed differences of each integration terminal, on the premise of ensuring sufficient accuracy, the artificial intelligence algorithm model simplification technology is used to construct a lightweight artificial intelligence algorithm to form a lightweight artificial intelligence algorithm library.

[0035] The specific content of step S1 is as follows: fully considering the uncertain characteristics of distributed photovoltaic power output, electric vehicle charging pile charging characteristics, and energy storage charging and discharging characteristics in the low-voltage distribution substation area, using the daily load data of distribution transformers, photovoltaic energy storage charging as input and output, using an autoencoder AE to extract the dimensionality reduction features of the load data, and using the dimensionality reduction features of the load data, the historical operation scenarios of the substation area are reduced based on the K-medoids clustering algorithm, and representative typical operation scenarios of the substation area are extracted. At the same time, combined with the external environment data in the operation process of power enterprises, using the above-mentioned typical scenario construction method of the substation area, typical scenarios of the distribution substation area under different influencing factors are established, and a typical operation scenario library of the low-voltage distribution substation area is established.

[0036] The external environment data includes environmental temperature and weather characteristics;

[0037] In step S1, for the complex multi-source heterogeneous massive data generated during the operation of power enterprises, technologies such as source data verification, source data error correction, and data reduction are used to integrate and integrate multi-source heterogeneous data, and the massive measurement data of multiple power grid systems are connected; at the same time, based on the integrated multi-source data, combined with the influencing factors of environmental temperature and season, clustering analysis, classification analysis, association analysis, and regression analysis methods are used to analyze the changes in typical operation scenarios of the substation area under different influencing factors, and the construction of typical operation scenarios of the distribution network under different influencing factors is realized.

[0038] The said step S2 includes: combining the measurement data generated during the operation of the low-voltage distribution substation area, recording the real-time operation data of the substation area edge equipment when executing the artificial intelligence algorithm, and constructing the matching relationship between the operation state of the substation area and the application of each artificial intelligence algorithm;

[0039] When the artificial intelligence algorithm matches multiple different operation states of the substation area, by performing AE feature extraction and K-medoids clustering of the reduced-dimensional features on the multiple operation state data, the number of substation area operation states matching the artificial intelligence algorithm is reduced, and the typical substation area scenario most similar to each substation area operation state of the artificial intelligence algorithm is found to complete the matching between the artificial intelligence algorithm and the typical substation area scenario, and its objective function is:

[0040]

[0041] Among them, S j is the measurement data of the j-th substation area operation state after reduction corresponding to the artificial intelligence algorithm, k is the total number of reduced substation area operation states, S ci is the operation data of the i-th established typical substation area operation scenario, and m is the total number of established typical substation area operation scenarios.

[0042] The said substation area edge equipment includes, such as, photovoltaic, energy storage, charging piles, and Internet of Things switch equipment;

[0043] In step S2, using the multi-source massive data stored during the operation of the low-voltage distribution substation area, by recording the real-time operation data of the substation area edge equipment when applying algorithms such as overload prediction, photovoltaic output prediction, energy storage coordination, and source following load movement, the matching relationship between the operation scenario of the substation area and the application of each artificial intelligence algorithm is constructed;

[0044] When a certain specific artificial intelligence algorithm matches multiple different scenarios of the substation area, by extracting features using AE for all the matching scenarios and using the big data analysis method of K-medoids clustering, the number of substation area operation scenarios matching the artificial intelligence algorithm is reduced, and the substation area operation scenario corresponding to the artificial intelligence algorithm and the established typical substation area scenario are matched using Euclidean distance and Pearson correlation coefficient technologies to realize the matching between the artificial intelligence algorithm and the typical substation area scenario.

[0045] Step S3 is specifically as follows: After completing the training of traditional artificial intelligence algorithms, considering the data transmission bandwidth, internal storage space size, computing speed differences, and system environment of each fusion terminal, and on the premise of ensuring sufficient accuracy, using the simplification technology of the artificial intelligence algorithm model, a lightweight artificial intelligence algorithm is constructed based on the trained traditional artificial intelligence algorithm to form a lightweight artificial intelligence algorithm library.

[0046] In step S3, when constructing the lightweight artificial intelligence algorithm library, the specific method is as follows: On the premise of completing the construction of typical operation scenarios in the substation area, the analysis of artificial intelligence algorithm application requirements, and the collation of information on substation area fusion terminal devices, select an artificial intelligence algorithm application for development;

[0047] Develop a lightweight artificial intelligence algorithm using the artificial intelligence algorithm model simplification technology on the host, and use the historical load data of the edge device as offline data to debug the algorithm until the result obtained by the algorithm is close to the actual operation result;

[0048] When there are differences between the host development environment and the target embedded machine, which may cause the transplanted algorithm to fail to run or have low running efficiency, the algorithm is modified and tested accordingly. A deployable lightweight artificial intelligence algorithm program is formed through the programming language according to the corresponding development environment, and the program is packaged using the docker containerization technology to achieve fast deployment and partition isolation from other computing programs, so that the model can quickly sink to the fusion terminal;

[0049] According to the above steps, the lightweight artificial intelligence algorithm applications in the deployed substation area are completed in an orderly manner. Under the condition of ensuring that the algorithm operation results meet the expectations, they are incorporated into the lightweight artificial intelligence algorithm library; more artificial intelligence algorithms are still developed according to the above steps and incorporated into the algorithm library to complete the construction of the lightweight artificial intelligence algorithm library.

[0050] In this example, the construction of the algorithm library is simulated and verified. Specifically, the measurement operation data of a certain distribution substation area on a certain day is selected for verification and analysis. The line loss rate of the substation area is used as the operation benefit index of the substation area, and the line loss rates of executing the algorithm and not executing the algorithm are compared, as Figure 2 shown.

[0051] The substation area enters the typical operation state at 8 o'clock and 18 o'clock respectively, and the corresponding lightweight artificial intelligence algorithm is executed. The line loss rate of the substation area decreases relative to not executing the algorithm after this time.

[0052] The comparison of lightweight artificial intelligence algorithms and traditional artificial intelligence algorithms in terms of algorithm execution time, CPU occupancy rate, and memory occupancy rate is shown in Table 1 below. The execution time of the lightweight artificial intelligence algorithm is reduced, and the resource occupancy is decreased. Therefore, it has great advantages when applied to devices with limited resources such as fusion terminals.

[0053] Table 1 Comparison between Traditional Artificial Intelligence Algorithm and Lightweight Artificial Intelligence Algorithm

[0054]

Claims

1. A method for constructing a lightweight artificial intelligence algorithm library for distribution transformer areas, characterized in that: It includes the following steps; Step S1: Considering the uncertainty of the load of edge devices in the distribution transformer area, based on dimensionality reduction and clustering algorithms, reduce the historical operation scenarios of the transformer area, and extract representative typical operation scenarios of the transformer area; Step S2: By recording the real-time operation data of edge devices when artificial intelligence algorithms are applied in the low-voltage distribution transformer area, construct the matching relationship between the operation state of the transformer area and the application of each artificial intelligence algorithm, and then realize the matching between the artificial intelligence algorithm and the operation scenario of the transformer area; Step S3: Considering the data transmission bandwidth, internal storage space size, and calculation speed differences of each fusion terminal, on the premise of ensuring sufficient accuracy, use artificial intelligence algorithm model simplification technology to construct lightweight artificial intelligence algorithms, and form a lightweight artificial intelligence algorithm library; The said Step S2 includes: Combining the measurement data generated during the operation of the low-voltage distribution transformer area, record the real-time operation data of the edge devices in the transformer area when the artificial intelligence algorithm is executed, and construct the matching relationship between the operation state of the transformer area and the application of each artificial intelligence algorithm; When the artificial intelligence algorithm matches multiple different operation states of the transformer area, through AE feature extraction and K-medoids clustering of the dimensionality reduction features of the multiple operation state data, reduce the number of operation states of the transformer area matching the artificial intelligence algorithm, and find the typical scenarios of the transformer area that are most similar to each operation state of the artificial intelligence algorithm in the transformer area, and complete the matching between the artificial intelligence algorithm and the typical scenarios of the transformer area. Its objective function is: Among them, S j is the measurement data of the operation state of the j-th substation area after the reduction of the operation state of the substation area corresponding to the artificial intelligence algorithm, k is the total number of reduced substation area operation states, and S ci is the operation data of the i-th established typical substation area operation scenario, and m is the total number of established typical substation area operation scenarios; In the said Step S3, when constructing the lightweight artificial intelligence algorithm library, the specific method is: On the premise of completing the construction of the typical operation scenarios of the transformer area, the analysis of the application requirements of the artificial intelligence algorithm, and the collation of the information of the fusion terminal devices in the transformer area, select an artificial intelligence algorithm application for development; Develop a lightweight artificial intelligence algorithm using artificial intelligence algorithm model simplification technology on the host computer, and use the historical load data of the edge device as offline data to debug the algorithm until the result obtained by the algorithm is close to the actual operation result; When there are differences between the host computer development environment and the target embedded machine, which will cause the transplanted algorithm to be unable to run or have low operation efficiency, then make targeted algorithm modifications and tests; Form a deployable lightweight artificial intelligence algorithm program through the program development language according to the corresponding development environment, and use the docker containerization technology to package the program to achieve fast deployment and partition isolation from other computing programs, so that the model can quickly sink to the fusion terminal; According to the above steps, orderly complete the deployment of the lightweight artificial intelligence algorithm application in the transformer area. When ensuring that the operation result of the algorithm meets the expectations, incorporate it into the lightweight artificial intelligence algorithm library; Other more artificial intelligence algorithms are still developed according to the above steps and incorporated into the algorithm library to complete the construction of the lightweight artificial intelligence algorithm library.

2. A method for constructing a lightweight artificial intelligence algorithm library for distribution transformer areas according to claim 1, characterized in that: The specific steps of step S1 are as follows: fully consider the uncertain characteristics of distributed photovoltaic power output, the charging characteristics of electric vehicle charging piles, and the charge and discharge characteristics of energy storage in the low-voltage distribution area. Taking the daily load data of distribution transformers and the integrated energy storage and charging as input and output, use the autoencoder AE to extract the dimensionality reduction features of the load data. Using the dimensionality reduction features of the load data, reduce the historical operation scenarios of the distribution area based on the K-medoids clustering algorithm, and extract representative typical operation scenarios of the distribution area. At the same time, combined with the external environment data in the operation process of power enterprises, use the above-mentioned method for constructing typical operation scenarios of the distribution area to establish typical scenarios of the distribution area under different influencing factors, and establish a typical operation scenario library for the low-voltage distribution area.

3. A method for constructing a lightweight artificial intelligence algorithm library for a distribution area according to claim 2, characterized in that: the external environment data includes environmental temperature and weather characteristics; In step S1, for the complex multi-source heterogeneous massive data generated during the operation of power enterprises, source data verification, source data error correction, and data reduction technologies are used to integrate and integrate the multi-source heterogeneous data, and the massive measurement data of multiple power grid systems are connected. At the same time, based on the integrated multi-source data, combined with the influencing factors of environmental temperature and season, use clustering analysis, classification analysis, association analysis, and regression analysis methods to analyze the changes in the typical operation scenarios of the distribution area under different influencing factors, and realize the construction of typical operation scenarios of the distribution network under different influencing factors.

4. A method for constructing a lightweight artificial intelligence algorithm library for a distribution area according to claim 1, characterized in that: the edge devices of the distribution area include, for example, photovoltaic, energy storage, charging piles, and Internet of Things switch devices; in step S2, using the multi-source massive data stored during the operation of the low-voltage distribution area, by recording the real-time operation data of the edge devices of the distribution area when applying overload prediction, photovoltaic power output prediction, energy storage coordination, and source-following-load algorithms, construct the matching relationship between the operation scenarios of the distribution area and the applications of various artificial intelligence algorithms; When a specific artificial intelligence algorithm matches multiple different scenarios of the distribution area, extract features from all the matching scenarios using AE, and use the big data analysis method of K-medoids clustering to reduce the number of operation scenarios of the distribution area matching the artificial intelligence algorithm, and use the Euclidean distance and Pearson correlation coefficient technologies to match the operation scenarios of the distribution area corresponding to the artificial intelligence algorithm with the established typical scenarios of the distribution area to achieve the matching between the artificial intelligence algorithm and the typical scenarios of the distribution area.

5. A method for constructing a lightweight artificial intelligence algorithm library for a distribution area according to claim 1, characterized in that: the specific steps of step S3 are as follows: after completing the training of traditional artificial intelligence algorithms, considering the data transmission bandwidth, internal storage space size, calculation speed difference, and system environment of each fusion terminal, on the premise of ensuring sufficient accuracy, use the simplification technology of artificial intelligence algorithm models to construct lightweight artificial intelligence algorithms based on the trained traditional artificial intelligence algorithms, and form a lightweight artificial intelligence algorithm library.

Images