AI digital human autonomous learning method and system based on deep learning

Abstract

The invention relates to the technical field of AI digital humans, and particularly discloses an AI digital human autonomous learning method and system based on deep learning, and the method comprises the steps: collecting user interaction behavior data and operation maintenance data of application equipment in an application scene in real time; performing optimization training on the built-in function model of the AI digital human by using the latest collected user interaction behavior data and the operation maintenance data of the application equipment to obtain an AI digital human autonomous learning result; the method and the device are used for improving the quality and the effect of interaction between AI digital people and users and providing services which are more suitable for user expectations. And the functions of the AI digital human can be continuously improved, so that the AI digital human has higher adaptability and competitiveness.

Description

Technical Field

[0001] The present invention relates to the field of AI digital human technology, and in particular to an AI digital human autonomous learning method and system based on deep learning. Background Art

[0002] At present, with the rapid development of artificial intelligence technology, AI digital humans have been widely used in many fields. In the current technological environment, the development of AI digital humans aims to interact with users more intelligently and efficiently, and better adapt to various complex application scenarios. In order to achieve this goal, researchers are committed to exploring various methods to improve the learning ability and intelligence level of AI digital humans. In terms of data collection, existing technologies can already obtain a certain amount of user interaction behavior data and operation and maintenance data of application equipment.

[0003] However, existing data collection methods cannot cover all possible user interaction behaviors and the operating conditions of application devices, resulting in the inability to ensure the integrity and timeliness of data, which affects the learning effect of digital humans. Training based on fixed models and limited data sets limits the learning ability and adaptability of AI digital humans to new situations, making it difficult to truly achieve autonomous learning and intelligence improvement. Due to the limitations of data and training methods, the autonomous learning results of AI digital humans often cannot meet the needs of actual application scenarios well, and cannot provide accurate and efficient services. It is impossible to optimize the built-in functional model of digital humans in a timely manner based on the latest data, resulting in the performance of digital humans gradually lagging behind changes in actual needs.

[0004] Therefore, the present invention proposes an AI digital human autonomous learning method and system based on deep learning. Summary of the invention

[0005] The present invention provides an AI digital human autonomous learning method and system based on deep learning, which is used to collect user interaction behavior data and equipment operation and maintenance data in real time, and can obtain the latest and most authentic information in a timely manner, providing rich and timely materials for the learning of AI digital humans. Optimizing and training the built-in functional model using the latest data can help AI digital humans quickly adapt to changing application scenarios and user needs. The obtained autonomous learning results can improve the quality and effect of the interaction between AI digital humans and users, and provide services that are more in line with user expectations. It can enhance the AI ​​digital human's understanding and response capabilities to the operating conditions of application equipment, and improve the stability and reliability of the system. It helps to continuously improve the functions of AI digital humans, making them more adaptable and competitive.

[0006] The present invention provides an AI digital human autonomous learning method based on deep learning, comprising:

[0007] S1: Real-time collection of user interaction behavior data and application equipment operation and maintenance data in application scenarios;

[0008] S2: Utilize the latest collected user interaction behavior data and application equipment operation and maintenance data to optimize the built-in functional model of AI digital human and obtain the autonomous learning results of AI digital human.

[0009] Preferably, the AI ​​digital human autonomous learning method based on deep learning, S1: real-time collection of user interaction behavior data in application scenarios and operation and maintenance data of application equipment, including:

[0010] Collect user voice interaction behavior data, user face interaction behavior data, user touch interaction behavior data, and user online and offline interaction behavior data in real time during each complete interaction process in the application scenario to obtain user interaction behavior data in the application scenario;

[0011] Collect historical operation data, operation performance data and manual operation and maintenance data of application equipment in real time as operation and maintenance data of application equipment.

[0012] Preferably, the AI ​​digital human autonomous learning method based on deep learning collects user voice interaction behavior data, user face interaction behavior data, user touch interaction behavior data, and user online and offline interaction behavior data in real time during each complete interaction process in the application scenario to obtain user interaction behavior data in the application scenario, including:

[0013] Collect user voice interaction behavior data, user face interaction behavior data, user touch interaction behavior data, and user online and offline interaction behavior data in real time during each complete interaction process in the application scenario;

[0014] Generate the identity information of each complete interaction process based on the user's facial interaction behavior data and the user's online and offline interaction behavior data during each complete interaction process;

[0015] The identity information of each complete interaction process is bound and stored with the user voice interaction behavior data, user face interaction behavior data, user touch interaction behavior data and user online and offline interaction behavior data of the corresponding complete interaction process to obtain the user interaction behavior data of each complete interaction process in the application scenario.

[0016] Preferably, the AI ​​digital human autonomous learning method based on deep learning generates the identity information of each complete interaction process based on the user's facial interaction behavior data and the user's online and offline interaction behavior data in each complete interaction process, including:

[0017] Extract the valid facial data of each complete interaction process from the user facial interaction behavior data of each complete interaction process, and extract the system login information or payment information of each complete interaction process from the user online and offline interaction behavior data of each complete interaction process;

[0018] Based on the valid facial data and system login information or payment information of each complete interaction process, the identity information of each complete interaction process is generated.

[0019] Preferably, the AI ​​digital human autonomous learning method based on deep learning, S2: using the latest collected user interaction behavior data and application equipment operation and maintenance data to optimize the built-in function model of the AI ​​digital human and obtain the AI ​​digital human autonomous learning results, including:

[0020] Use the latest collected user interaction behavior data to optimize and train the product recommendation model and product demand prediction model currently built into the AI ​​digital human to obtain optimized product recommendation model and product demand prediction model;

[0021] Use the latest collected operation and maintenance data of application equipment to optimize the equipment prediction operation and maintenance model of AI digital human to obtain an optimized equipment prediction operation and maintenance model;

[0022] Among them, the built-in functional models of AI digital humans include: product recommendation model, product demand prediction model, and equipment prediction and operation and maintenance model.

[0023] Preferably, the AI ​​digital human autonomous learning method based on deep learning uses the latest collected user interaction behavior data to optimize and train the product recommendation model and product demand prediction model currently built into the AI ​​digital human, respectively, to obtain the optimized product recommendation model and product demand prediction model, including:

[0024] Based on the latest collected user interaction behavior data of each complete interaction process, all the final purchased products of the user and the purchase intention index of each product are analyzed;

[0025] Based on the latest collected identity information of each complete interaction process, all final purchased products, and the purchase intention index of each product, the product recommendation training sample set corresponding to the identity information is updated;

[0026] Input the product recommendation training sample set of the updated identity information into the built-in product recommendation model of the AI ​​digital human for model optimization training to obtain an optimized product recommendation model;

[0027] Based on the latest collected identity information of all complete interaction processes, all final purchased products, and the purchase intention index of each product, the product demand prediction training sample set is updated to obtain a new product demand training sample set;

[0028] The latest product demand training sample set is input into the product demand forecasting model built into the AI ​​digital human for model optimization training to obtain an optimized product demand forecasting model.

[0029] Preferably, the AI ​​digital human autonomous learning method based on deep learning analyzes all the user's final purchased products and the purchase intention index of each product based on the latest collected user interaction behavior data of each complete interaction process, including:

[0030] Determine all final purchase products of the user based on the latest collected user interaction behavior data of each complete interaction process, and set the purchase intention index of each final purchase product to 1;

[0031] Treat all available products, except for all final purchased products in each complete interaction process, as unpurchased products;

[0032] Based on the most recently collected user interaction behavior data of each complete interaction process, determine the browsing time, viewing times, consultation frequency, comparison times with other products, total number of comparison information items, and consultation initiation initiative index of each unpurchased product by the user in each complete interaction process;

[0033] Based on the browsing time, viewing times, consultation frequency, number of comparisons with other products, total number of comparison information items and consultation initiation initiative index of each unpurchased product during each complete interaction process, the purchase intention index of each unpurchased product is determined.

[0034] Preferably, the AI ​​digital human autonomous learning method based on deep learning determines the purchase intention index of each unpurchased product based on the browsing time, viewing times, consultation frequency, comparison times with other products, total number of comparison information items, and consultation initiation initiative index of each unpurchased product by the user in each complete interaction process, including:

[0035] Determine the longest browsing time, maximum number of views, maximum consultation frequency, maximum number of comparisons with other products, and maximum total number of comparison items for all historically purchased products of the user;

[0036] Based on the longest browsing time, maximum number of views, maximum consultation frequency, maximum number of comparisons with other products, maximum total number of comparison items of all historically purchased products of the user, the browsing time, view number, consultation frequency, number of comparisons with other products, total number of comparison information items of each unpurchased product of the user, and the consultation initiation initiative index, determine the calculated reference browsing time, calculated reference view number, calculated reference consultation frequency, calculated reference comparison number with other products, and total number of calculated reference comparison information items of each unpurchased product of the user;

[0037] Based on the user's browsing time, viewing times, consultation frequency, comparison times with other products, total number of comparison information items, and consultation initiation initiative index for each unpurchased product, the purchase intention index of each unpurchased product is calculated:

[0038]

[0039] Where ρ is the purchase intention index of the unpurchased product, t 0·scan is the reference browsing time for users to calculate the unpurchased products, t scan·max The longest browsing time of all historically purchased products of the user, n 0·ov The number of times users have viewed products that they have not purchased, n ov·max The maximum number of views for all historical purchases of a user, f 0·oc is the frequency of users’ reference consultation for products they have not purchased, f oc·max The maximum consultation frequency of all historical purchases of the user, n 0·oc The number of times users compare unpurchased products with other products, n oc·max The maximum number of comparisons with other products for all historically purchased products of the user, n 0·iifc is the total number of comparison information items of the user for the unpurchased products, n iifc·max The maximum total number of comparison items for all historical purchases of the user, Ac oci Initiate proactive index for users’ inquiries about products they have not purchased.

[0040] Preferably, the AI ​​digital human autonomous learning method based on deep learning uses the latest collected operation and maintenance data of the application equipment to perform model optimization training on the equipment prediction operation and maintenance model of the AI ​​digital human to obtain an optimized equipment prediction operation and maintenance model, including:

[0041] Based on the manual operation and maintenance data in the latest collected operation and maintenance data of the application equipment, the historical operation data and operation performance data between each group of adjacent manual operation and maintenance records and the manual operation and maintenance data of the latest manual operation and maintenance record in the corresponding group of adjacent manual operation and maintenance records are extracted from the latest collected operation and maintenance data of the application equipment;

[0042] The historical operation data and operation performance data between each group of adjacent manual operation and maintenance records and the manual operation and maintenance data of the latest manual operation and maintenance record in the corresponding group of adjacent manual operation and maintenance records are taken as a single training sample. All training samples are input into the equipment predictive operation and maintenance model of AI digital human for model optimization training to obtain an optimized equipment predictive operation and maintenance model.

[0043] The present invention provides an AI digital human autonomous learning system based on deep learning, which is used to execute any one of the AI ​​digital human autonomous learning methods based on deep learning in Embodiments 1 to 9, including:

[0044] The data collection module is used to collect user interaction behavior data and application equipment operation and maintenance data in real time under the application scenario;

[0045] The autonomous learning module is used to optimize the training of the built-in functional model of AI digital human by utilizing the latest collected user interaction behavior data and operation and maintenance data of application equipment to obtain the autonomous learning results of AI digital human.

[0046] The beneficial effects of the present invention compared to the prior art are: real-time collection of user interaction behavior data and equipment operation and maintenance data, timely acquisition of the latest and most authentic information, and provision of rich and timely materials for the learning of AI digital humans. Optimizing and training the built-in functional models using the latest data can help AI digital humans quickly adapt to changing application scenarios and user needs. The autonomous learning results obtained can improve the quality and effect of the interaction between AI digital humans and users, and provide services that are more in line with user expectations. It can enhance the AI ​​digital human's understanding and response capabilities to the operating conditions of application equipment, and improve the stability and reliability of the system. It helps to continuously improve the functions of AI digital humans, making them more adaptable and competitive.

[0047] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in this application document.

[0048] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0050] Figure 1 This is a flow chart of an AI digital human autonomous learning method based on deep learning in an embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram of the built-in functional modules of the AI ​​digital human in the embodiment of the present invention;

[0052] Figure 3 Schematic diagram of an AI digital human autonomous learning system based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0054] Embodiment 1:

[0055] The present invention provides an AI digital human autonomous learning method based on deep learning, referring to Figure 1 ,include:

[0056] S1: Real-time collection of user interaction behavior data and application equipment operation and maintenance data in application scenarios;

[0057] S2: Utilize the latest collected user interaction behavior data and application equipment operation and maintenance data to optimize the built-in functional model of AI digital human and obtain the autonomous learning results of AI digital human.

[0058] In this embodiment, the user interaction behavior data is in the scenario of a vending machine, such as the conversation between the user and the little man on the vending machine screen, such as "buy a bottle of Coke", the user's response to product recommendations, whether the introduction of certain products is clicked, and operations in fun questions and answers and mini-games.

[0059] In this embodiment, the operation and maintenance data of the application device refers to the operation status of the vending machine, such as the temperature of the machine, whether the refrigeration system is normal, whether the delivery port is blocked, and the real-time changes in the inventory quantity of the goods.

[0060] In this embodiment, the result of AI digital human autonomous learning may be that the digital human on the vending machine can more accurately judge user needs, such as more accurately identifying new and old users, recommending products that are more in line with their possible needs to new users, and recommending products that are more in line with their purchasing habits to old users. It may also be smarter in inventory management, such as more timely and accurate reminders to replenishers to put goods on and off shelves and replace them. Or in terms of fault detection, it can detect potential faults faster and more accurately.

[0061] The beneficial effects of the above technologies are: real-time collection of user interaction behavior data and equipment operation and maintenance data, timely acquisition of the latest and most authentic information, and provision of rich and timely materials for AI digital human learning. Optimizing and training the built-in functional models using the latest data can help AI digital humans quickly adapt to changing application scenarios and user needs. The autonomous learning results obtained can improve the quality and effectiveness of AI digital humans’ interactions with users, and provide services that better meet user expectations. It can enhance AI digital humans’ understanding and response capabilities to the operating conditions of application equipment, and improve the stability and reliability of the system. It helps to continuously improve the functions of AI digital humans, making them more adaptable and competitive.

[0062] Embodiment 2:

[0063] On the basis of Example 1, the AI ​​digital human autonomous learning method based on deep learning, S1: real-time collection of user interaction behavior data in application scenarios and operation and maintenance data of application devices, including:

[0064] Collect user voice interaction behavior data, user face interaction behavior data, user touch interaction behavior data, and user online and offline interaction behavior data in real time during each complete interaction process in the application scenario to obtain user interaction behavior data in the application scenario;

[0065] Collect historical operation data, operation performance data and manual operation and maintenance data of application equipment in real time as operation and maintenance data of application equipment.

[0066] In this embodiment, the application scenario refers to the specific context and environment in which a specific technology, product or service is used. For example, the AI ​​digital human for a vending machine is the entire context in which a user performs a shopping operation in front of the vending machine, including interaction with the digital human, selecting products, and completing payment.

[0067] In this embodiment, the complete interaction process is the entire coherent process from the user's initial contact with the system to the completion of a specific task or achievement of a certain goal, and the end of the interaction with the system. For example, a user walks to a vending machine, a digital human pops up, the user expresses the purchase demand through voice, the digital human makes recommendations and guides, the user completes the selection and payment, takes the product and leaves. This series of continuous actions constitutes a complete interaction process.

[0068] In this embodiment, the user voice interaction behavior data refers to the data generated by the user communicating with the system through voice, including the specific content of the speech, the voice tone, the speech speed, etc. For example, if a user says to a vending machine "I want a bottle of Coke, ice cold", this sentence and the tone, speech speed, etc. are the user voice interaction behavior data.

[0069] In this embodiment, the user facial interaction behavior data is data obtained by identifying and analyzing the user's facial expressions, features, etc. For example, the camera of a vending machine captures the user's surprised expression when seeing the recommended product or the impatient expression due to the long waiting time. These expression information is the user facial interaction behavior data.

[0070] In this embodiment, the user touch interaction behavior data refers to the relevant data generated by the user directly touching the device screen or operation interface, such as the touch position, duration, strength, etc. For example, if a user clicks on a certain product icon on the vending machine screen for a long time to express a strong desire to buy, or quickly slides the screen to view more products, the position, duration, strength, etc. of these touch operations are the user touch interaction behavior data.

[0071] In this embodiment, the user online and offline interactive behavior data covers all relevant data of the user's interaction with the system online (such as through mobile applications, websites) and offline (in front of physical devices). For example, a user views the product introduction of a vending machine online but does not place an order, and directly purchases the product in front of a physical vending machine offline, or receives a coupon online and uses it offline. These online and offline operations and related records are the user's online and offline interactive behavior data.

[0072] In this embodiment, the historical operation data of the application device refers to the record data about the past operation of the device, including the operation time, number of times, various parameters in the operation, etc. For example, the daily start-up time and shutdown time of the vending machine in the past, the working hours and temperature settings of the refrigeration system in the past month, the number of shipments and the types of goods shipped each time, etc.

[0073] In this embodiment, the operating performance data of the application device refers to the performance data displayed by the device during operation, such as refrigeration efficiency, network connection stability, screen response speed, etc. For example, whether the refrigeration system of the vending machine can keep the goods in a low temperature state under different ambient temperatures, whether the network connection is stable to ensure timely update of product information and smooth payment, whether the screen responds quickly to user touch operations, and other performance data.

[0074] In this embodiment, the manual operation and maintenance data of the application equipment refers to the data generated by the staff in maintaining and managing the equipment, such as the time of maintenance, the record of parts replacement, the cycle of cleaning and maintenance, etc. For example, the staff of the vending machine repaired the machine on a certain day and recorded the specific problem and solution of the repair, or the time and operation record of regularly cleaning and replenishing the vending machine, which are the manual operation and maintenance data of the application equipment.

[0075] The beneficial effects of the above technologies are: comprehensive collection of various types of user interaction behavior data, including voice, face, touch, and online and offline interactions, which can provide a more comprehensive and in-depth understanding of user behavior patterns and needs, and provide rich and multi-dimensional information for digital human learning. Real-time collection ensures the timeliness of data, enabling AI digital humans to respond and adapt to the latest user interaction trends in a timely manner. Detailed collection of operation and maintenance data of application equipment, covering historical operation data, performance data, and manual operation and maintenance data, helps to comprehensively evaluate the status of equipment and provide a more comprehensive reference for the learning and optimization of digital humans. Rich user interaction behavior data and equipment operation and maintenance data can improve AI digital humans' understanding and response capabilities for complex application scenarios. It helps to improve the service quality and intelligence level of AI digital humans, and provide users with a more personalized and efficient interactive experience.

[0076] Embodiment 3:

[0077] On the basis of Example 2, the AI ​​digital human autonomous learning method based on deep learning collects user voice interaction behavior data, user face interaction behavior data, user touch interaction behavior data, and user online and offline interaction behavior data in real time during each complete interaction process in the application scenario to obtain user interaction behavior data in the application scenario, including:

[0078] Collect user voice interaction behavior data, user face interaction behavior data, user touch interaction behavior data, and user online and offline interaction behavior data in real time during each complete interaction process in the application scenario;

[0079] Generate the identity information of each complete interaction process based on the user's facial interaction behavior data and the user's online and offline interaction behavior data during each complete interaction process;

[0080] The identity information of each complete interaction process is bound and stored with the user voice interaction behavior data, user face interaction behavior data, user touch interaction behavior data and user online and offline interaction behavior data of the corresponding complete interaction process to obtain the user interaction behavior data of each complete interaction process in the application scenario.

[0081] In this embodiment, the identity information of the complete interaction process refers to the relevant information that can identify the user's identity determined by analyzing and processing the user's facial interaction behavior data and the user's online and offline interaction behavior data during each complete interaction between the user and the application scenario (such as a vending machine).

[0082] For example, in the vending machine scenario, by analyzing the user's facial features, online payment account number, or offline membership card information, it can be determined whether the user is Zhang San or Li Si. This Zhang San or Li Si identification is the identity information of the entire interaction process. It helps to distinguish and classify the interaction behavior data of different users, so as to provide personalized services for each user and analyze their specific behavior patterns.

[0083] The beneficial effects of the above technologies are: comprehensive and real-time collection of a variety of user interaction behavior data, ensuring the integrity and comprehensiveness of the data, and providing rich materials for subsequent analysis. Generate the identity information of each complete interaction process, and bind and store it with other interaction behavior data, so as to achieve accurate tracing and analysis of user interaction behavior. It helps to better understand the user's behavior habits and preferences in different scenarios, and provide strong support for the personalized learning of digital people. It can improve the accuracy of identifying user needs, so as to provide services that are more in line with the user's identity and characteristics. The binding storage method facilitates data management and call, and improves the efficiency and accuracy of data processing.

[0084] Embodiment 4:

[0085] On the basis of Example 3, the AI ​​digital human autonomous learning method based on deep learning generates the identity information of each complete interaction process based on the user's facial interaction behavior data and the user's online and offline interaction behavior data in each complete interaction process, including:

[0086] Extract the valid facial data of each complete interaction process from the user facial interaction behavior data of each complete interaction process, and extract the system login information or payment information of each complete interaction process from the user online and offline interaction behavior data of each complete interaction process;

[0087] Based on the valid facial data and system login information or payment information of each complete interaction process, the identity information of each complete interaction process is generated.

[0088] In this embodiment, effective facial data refers to key facial feature data extracted from the user's facial interaction behavior data during each complete interaction process and can be used to accurately identify and distinguish the user's identity;

[0089] For example: In a vending machine scenario, effective facial data may be unique and stable feature data such as the user's facial contour, eye distance, nose shape, etc.

[0090] In this embodiment, the system login information is information such as account number and password entered by the user for identity verification and login when using the relevant system.

[0091] For example: For example, the mobile phone number and password entered by the user in the online application of the vending machine, or the card number when using a membership card.

[0092] In this embodiment, payment information refers to relevant data generated when a user makes a transaction payment, including payment method, payment account number, payment amount, etc.

[0093] For example: At a vending machine, the user uses their WeChat account when paying with WeChat, or their bank card number when paying with a bank card.

[0094] In this embodiment, based on the valid facial data and system login information or payment information of each complete interaction process, the identity information of each complete interaction process is generated. This means that the facial feature data extracted from each interaction that can accurately reflect the user's identity, as well as the user's system login account, password or account used for payment, are comprehensively utilized to determine the identity of the specific user to whom this interaction process belongs.

[0095] For example, the vending machine analyzes the user's facial features and combines the user's mobile phone number used to log in to the online application or the specific payment account used for payment to ultimately determine that the interaction belongs to the user Zhang San. The result "Zhang San" determined here is the identity information of the entire interaction process.

[0096] The beneficial effects of the above technology are: extracting effective facial data from facial interaction behavior data, and extracting system login or payment information from online and offline interaction behavior data, thereby improving the accuracy of obtaining key information for identity recognition. Generating identity information based on effective facial data and specific login or payment information enhances the accuracy and reliability of identity recognition. It helps to more accurately associate user interaction behaviors with specific identities, providing a more accurate basis for personalized services and learning. It can effectively protect user privacy, extract only necessary information related to identity generation, and reduce the risk of data leakage. It provides a solid foundation for AI digital humans to more accurately understand user needs and optimize service strategies, thereby improving user experience.

[0097] Embodiment 5:

[0098] On the basis of Example 1, the AI ​​digital human autonomous learning method based on deep learning, S2: using the latest collected user interaction behavior data and application equipment operation and maintenance data, optimize the built-in function model of the AI ​​digital human to obtain the AI ​​digital human autonomous learning results, including:

[0099] Use the latest collected user interaction behavior data to optimize and train the product recommendation model and product demand prediction model currently built into the AI ​​digital human to obtain optimized product recommendation model and product demand prediction model;

[0100] Use the latest collected operation and maintenance data of application equipment to optimize the equipment prediction operation and maintenance model of AI digital human to obtain an optimized equipment prediction operation and maintenance model;

[0101] Among them, reference Figure 2 The built-in functional models of AI digital humans include: product recommendation model, product demand prediction model, and equipment prediction and operation model.

[0102] In this embodiment, the product recommendation model is a model that can recommend products that the user may be interested in or need based on the user's behavior data and characteristics.

[0103] For example: In a vending machine, the product recommendation model will recommend similar or related products to users based on their past purchase records, browsing behavior, etc.

[0104] In this embodiment, the product demand prediction model is a model used to predict users' future demand for different products.

[0105] For example: By analyzing the sales trend, seasonal factors, and changes in the surrounding environment of a certain commodity over a period of time, the demand for this commodity in the future can be predicted.

[0106] In this embodiment, the optimized product recommendation model and product demand prediction model are product recommendation models and product demand prediction models that have been trained and improved with the latest collected user interaction behavior data, and can perform product recommendations and demand predictions more accurately.

[0107] For example: After optimization, the product recommendation model can more accurately recommend breakfast items to users who come in the morning, and the product demand forecasting model can more accurately predict that the demand for beverages will increase on weekends.

[0108] In this embodiment, the equipment prediction operation and maintenance model is a model used to predict the operation status and maintenance requirements of application equipment (such as vending machines).

[0109] For example, it is possible to predict how long in the future a certain component of a vending machine may fail, or when the inventory of a certain product needs to be replenished.

[0110] In this embodiment, the optimized equipment prediction operation and maintenance model is an improved equipment prediction operation and maintenance model trained using the latest application equipment operation and maintenance data, which can more effectively predict the operation and maintenance status of the equipment.

[0111] For example, after optimization, it can more accurately predict a week in advance that a vending machine’s refrigeration system needs maintenance, or predict three days in advance that a popular product is about to be out of stock.

[0112] The beneficial effects of the above technologies are as follows: optimizing and training the product recommendation model and the product demand prediction model respectively can improve the accuracy of AI digital humans in recommending products to users and predicting user needs, thereby better meeting the personalized needs of users. Optimizing and training the equipment prediction and operation and maintenance model can help improve the AI ​​digital human's ability to predict and maintain the operating status of application equipment, ensure the stable operation of the system, and reduce the impact of equipment failures. Targeted optimization is carried out for different built-in functional models to give full play to the advantages of each model and improve the overall service quality and efficiency. The optimized model can provide users with better quality and more tailored services to their needs, and enhance user satisfaction and trust in AI digital humans. It helps AI digital humans to continuously adapt to new data and changing environments, and maintain their competitiveness and practicality in application scenarios.

[0113] Embodiment 6:

[0114] On the basis of Example 5, the AI ​​digital human autonomous learning method based on deep learning uses the latest collected user interaction behavior data to optimize and train the product recommendation model and product demand prediction model currently built into the AI ​​digital human, respectively, to obtain the optimized product recommendation model and product demand prediction model, including:

[0115] Based on the latest collected user interaction behavior data of each complete interaction process, all the final purchased products of the user and the purchase intention index of each product are analyzed;

[0116] Based on the latest collected identity information of each complete interaction process, all final purchased products, and the purchase intention index of each product, the product recommendation training sample set corresponding to the identity information is updated;

[0117] Input the product recommendation training sample set of the updated identity information into the built-in product recommendation model of the AI ​​digital human for model optimization training to obtain an optimized product recommendation model;

[0118] Based on the latest collected identity information of all complete interaction processes, all final purchased products, and the purchase intention index of each product, the product demand prediction training sample set is updated to obtain a new product demand training sample set;

[0119] The latest product demand training sample set is input into the product demand forecasting model built into the AI ​​digital human for model optimization training to obtain an optimized product demand forecasting model.

[0120] In this embodiment, all the final products purchased by the user refer to all the commodities that the user actually decides to purchase and completes in a complete shopping interaction process. For example, in a vending machine, the user may finally purchase a bottle of Coke and a bag of potato chips.

[0121] In this embodiment, the purchase intention index of each product is a numerical indicator used to measure the degree of the user's purchase intention for each product. For example, the user's purchase intention index for cola is 80%, and the purchase intention index for potato chips is 60%. The higher the value, the stronger the purchase intention.

[0122] In this embodiment, the product recommendation training sample set of the identity information is a set of data associated with a specific user identity and used to train the product recommendation model, including the user's purchase history, browsing behavior, etc. For example, for user Zhang San, his training sample set may include his past purchase records of beverages and snacks.

[0123] In this embodiment, the product recommendation training sample set of the updated identity information is input into the product recommendation model built into the AI ​​digital human for model optimization training to obtain an optimized product recommendation model: the training sample set related to the specific user identity updated according to the new user interaction behavior data is put into the product recommendation model for training, thereby obtaining an improved model that can more accurately recommend products. For example: According to Zhang San's latest purchase behavior, his training sample set is updated. After inputting it into the model training, the model can more accurately recommend products that Zhang San may like.

[0124] In this embodiment, the product demand prediction training sample set is a set of data used to train a product demand prediction model, and includes information such as purchase history and purchase intention index of multiple users.

[0125] For example: Contains a large amount of data on users’ purchase status and intentions for various commodities in different time periods.

[0126] In this embodiment, the newly obtained product demand training sample set is input into the product demand forecasting model built into the AI ​​digital human for model optimization training to obtain an optimized product demand forecasting model: the newly obtained product demand forecasting training sample set is put into the product demand forecasting model for training to obtain a product demand forecasting model that can more accurately predict product demand.

[0127] For example: By inputting a sample set containing the latest market changes and user demand trends, the trained model can more accurately predict the demand for a certain product in the future.

[0128] The beneficial effects of the above technologies are: analyzing the user's final purchase product and purchase intention index can deeply understand the user's consumption preferences and demand tendencies, and provide valuable basis for model optimization. Update the product recommendation training sample set according to identity information and purchase status, so that the product recommendation model is more personalized and accurate, and the hit rate and satisfaction of the recommendation are improved. Continuously updating the product demand prediction training sample set and using it for model optimization training can help more accurately predict users' future needs and prepare for services in advance. An optimized product recommendation model can provide users with product recommendations that are more in line with their interests and needs, and enhance users' shopping experience and loyalty. An optimized product demand prediction model can help companies plan product strategies and resource allocation more effectively and improve market competitiveness.

[0129] Embodiment 7:

[0130] On the basis of Example 6, the AI ​​digital human autonomous learning method based on deep learning analyzes all the user's final purchased products and the purchase intention index of each product based on the latest collected user interaction behavior data of each complete interaction process, including:

[0131] Determine all final purchase products of the user based on the latest collected user interaction behavior data of each complete interaction process, and set the purchase intention index of each final purchase product to 1;

[0132] Treat all available products, except for all final purchased products in each complete interaction process, as unpurchased products;

[0133] Based on the most recently collected user interaction behavior data of each complete interaction process, determine the browsing time, viewing times, consultation frequency, comparison times with other products, total number of comparison information items, and consultation initiation initiative index of each unpurchased product by the user in each complete interaction process;

[0134] Based on the browsing time, viewing times, consultation frequency, number of comparisons with other products, total number of comparison information items and consultation initiation initiative index of each unpurchased product during each complete interaction process, the purchase intention index of each unpurchased product is determined.

[0135] In this embodiment, the consultation frequency refers to how often the user consults the system about a certain unpurchased product during each complete interaction process.

[0136] For example, in the vending machine scenario, if a user asks three questions about a chocolate, then the inquiry frequency is three.

[0137] In this embodiment, the number of comparisons with other products indicates the number of times a user compares a certain unpurchased product with other products during each complete interaction process.

[0138] For example, when a user is considering buying a beverage, he compares a certain juice with 5 other beverages. Then the number of comparisons between the juice and other products is 5.

[0139] In this embodiment, the total number of comparison information items is the sum total of various information involved when the user compares the unpurchased product with other products.

[0140] For example, if a user is comparing information about prices, flavors, volumes, etc. of different beverages, the total number of these information items is the total number of compared information items.

[0141] In this embodiment, the consultation initiation initiative index is used to measure the degree of activeness of users in initiating consultations on unpurchased products during each complete interaction process. When the first consultation on a product is initiated by the user, the corresponding consultation initiation initiative index is 1; when the first consultation on a product is initiated after the guidance of the AI ​​digital human, the corresponding consultation initiation initiative index is 0.

[0142] The beneficial effects of the above technology are: setting the purchase intention index of the final purchased product to 1, clarifying the certainty of the purchase behavior, and providing a clear benchmark for subsequent analysis. Determining unpurchased products and comprehensively considering multi-dimensional behavioral data such as user browsing time and number of views can more comprehensively evaluate users' potential interest in unpurchased products. Determining the purchase intention index of unpurchased products based on multi-dimensional behavioral data improves the accuracy and reliability of intention assessment. It helps to more accurately grasp the user's consumption psychology and potential needs, and provide more valuable information for product recommendations and demand forecasting. It can make the learning of AI digital people more in line with actual user behavior, thereby providing services and suggestions that are more in line with user expectations.

[0143] Embodiment 8:

[0144] On the basis of Example 7, the AI ​​digital human autonomous learning method based on deep learning determines the purchase intention index of each unpurchased product based on the browsing time, viewing times, consultation frequency, comparison times with other products, total number of comparison information items, and consultation initiation initiative index of each unpurchased product by the user in each complete interaction process, including:

[0145] Determine the longest browsing time, maximum number of views, maximum consultation frequency, maximum number of comparisons with other products, and maximum total number of comparison items for all historically purchased products of the user;

[0146] Based on the longest browsing time, maximum number of views, maximum consultation frequency, maximum number of comparisons with other products, maximum total number of comparison items of all historically purchased products of the user, the browsing time, view number, consultation frequency, number of comparisons with other products, total number of comparison information items of each unpurchased product of the user, and the consultation initiation initiative index, determine the calculated reference browsing time, calculated reference view number, calculated reference consultation frequency, calculated reference comparison number with other products, and total number of calculated reference comparison information items of each unpurchased product of the user;

[0147] Based on the user's browsing time, viewing times, consultation frequency, comparison times with other products, total number of comparison information items, and consultation initiation initiative index for each unpurchased product, the purchase intention index of each unpurchased product is calculated:

[0148]

[0149] Where ρ is the purchase intention index of the unpurchased product, t 0·scan is the reference browsing time for users to calculate the unpurchased products, t scan·max The longest browsing time of all historically purchased products of the user, n 0·ov The number of times users have viewed products that they have not purchased, n ov·max The maximum number of views for all historical purchases of a user, f 0·oc is the frequency of users’ reference consultation for products they have not purchased, f oc·max The maximum consultation frequency of all historical purchases of the user, n 0·oc The number of times users compare unpurchased products with other products, n oc·max The maximum number of comparisons with other products for all historically purchased products of the user, n 0·iifc is the total number of comparison information items of the user for the unpurchased products, n iifc·max The maximum total number of comparison items for all historical purchases of the user, Ac oci Initiate proactive index for users’ inquiries about products they have not purchased.

[0150] In this embodiment, based on the longest browsing time, maximum number of views, maximum consultation frequency, maximum number of comparisons with other products, maximum total number of comparison items of all historically purchased products of the user, the browsing time, view number, consultation frequency, number of comparisons with other products, total number of comparison information items of each unpurchased product of the user, and the consultation initiation initiative index, the calculated reference browsing time, calculated reference view number, calculated reference consultation frequency, calculated reference comparison number with other products, and total number of calculated reference comparison information items of each unpurchased product of the user are determined, namely:

[0151] When the longest browsing time (or the maximum number of views or the maximum consultation frequency or the maximum number of comparisons with other products or the maximum total number of comparison items) of all the user's historically purchased products is greater than the user's browsing time (or the number of views or the consultation frequency or the number of comparisons with other products or the total number of comparison information items) for each unpurchased product, the user's browsing time (or the number of views or the consultation frequency or the number of comparisons with other products or the total number of comparison information items) for each unpurchased product shall be regarded as the user's calculated reference browsing time (or the calculated reference number of views or the calculated reference consultation frequency or the calculated reference number of comparisons with other products or the total number of calculated reference comparison information items) for each unpurchased product; otherwise, the longest browsing time (or the maximum number of views or the maximum consultation frequency or the maximum number of comparisons with other products or the maximum total number of comparison items) for all the user's historically purchased products shall be regarded as the user's calculated reference browsing time (or the calculated reference number of views or the calculated reference consultation frequency or the calculated reference number of comparisons with other products or the total number of calculated reference comparison information items) for each unpurchased product.

[0152] The beneficial effects of the above technology are: considering multiple factors such as the browsing time, viewing times, consultation frequency, comparison times with other products, total number of comparison information items, and consultation initiation initiative index of users for unpurchased products, it can comprehensively evaluate the user's interest in unpurchased products from multiple perspectives, and improve the accuracy and reliability of the purchase intention index. Combined with the relevant maximum values ​​of all historically purchased products of the user, it provides an effective reference for evaluating the purchase intention of unpurchased products. By comparing the parameters of unpurchased products with historically purchased products, the user's purchase behavior patterns and preferences can be better understood, so as to more accurately predict the user's purchase intention for unpurchased products. The calculation of the purchase intention index is more scientific and reasonable by comprehensively calculating multiple parameters through a specific formula. Each parameter in the formula has a clear physical meaning, and the different parameters are integrated by weighted average, which avoids the one-sidedness of a single factor and improves the credibility and practicality of the purchase intention index. This technical solution provides an effective tool for the autonomous learning of AI digital people, which can help the model better understand the user's purchase intention, so as to formulate more accurate marketing strategies, optimize product recommendations and inventory management decisions. At the same time, it can also provide users with a more personalized shopping experience and improve user satisfaction and loyalty.

[0153] Embodiment 9:

[0154] On the basis of Example 5, the AI ​​digital human autonomous learning method based on deep learning uses the latest collected operation and maintenance data of the application equipment to perform model optimization training on the equipment prediction operation and maintenance model of the AI ​​digital human to obtain an optimized equipment prediction operation and maintenance model, including:

[0155] Based on the manual operation and maintenance data in the latest collected operation and maintenance data of the application equipment, the historical operation data and operation performance data between each group of adjacent manual operation and maintenance records and the manual operation and maintenance data of the latest manual operation and maintenance record in the corresponding group of adjacent manual operation and maintenance records are extracted from the latest collected operation and maintenance data of the application equipment;

[0156] The historical operation data and operation performance data between each group of adjacent manual operation and maintenance records and the manual operation and maintenance data of the latest manual operation and maintenance record in the corresponding group of adjacent manual operation and maintenance records are taken as a single training sample. All training samples are input into the equipment predictive operation and maintenance model of AI digital human for model optimization training to obtain an optimized equipment predictive operation and maintenance model.

[0157] In this embodiment, adjacent manual operation and maintenance records refer to two records that are adjacent to each other in time in a series of manual operation and maintenance records of application devices arranged in chronological order.

[0158] For example: In the operation and maintenance records of a vending machine, assuming that there is manual operation and maintenance on May 1 and May 8 respectively, then these two operation and maintenance records are adjacent manual operation and maintenance records.

[0159] In this embodiment, all training samples are input into the equipment prediction and operation and maintenance model of the AI ​​digital human for model optimization training. Obtaining an optimized equipment prediction and operation and maintenance model means that all individual training samples extracted and sorted according to the above method are put into the equipment prediction and operation and maintenance model of the AI ​​digital human, and the model is adjusted and improved through specific algorithms and processes, so as to obtain an equipment prediction and operation and maintenance model with better performance and more accurate predictions.

[0160] For example, multiple training samples related to adjacent manual operation and maintenance records extracted from the operation and maintenance data of the vending machine are input into the equipment predictive operation and maintenance model for training, so that the model can more accurately predict possible future failures of the vending machine, the goods that need to be replenished, etc., and finally obtain an optimized model.

[0161] The beneficial effects of the above technology are: extracting relevant data between adjacent manual operation and maintenance records from operation and maintenance data can focus on the operation of equipment in critical time periods, and improve the pertinence and effectiveness of data. Treating a specific combination of data as a single training sample makes the training sample more comprehensive and practical, which helps the model learn more accurate operation and maintenance rules. Inputting these training samples for model optimization training can improve the equipment predictive operation and maintenance model's ability to predict the equipment's operating status and the accuracy of operation and maintenance decisions. The optimized equipment predictive operation and maintenance model can detect potential equipment problems in advance, reduce the probability of equipment failures, and improve the stability and reliability of the equipment. It helps to reduce equipment operation and maintenance costs, improve operation and maintenance efficiency, and ensure the normal operation of the entire system.

[0162] Embodiment 10:

[0163] The present invention provides an AI digital human autonomous learning system based on deep learning, which is used to execute any one of the AI ​​digital human autonomous learning methods based on deep learning in Embodiments 1 to 9. Figure 3 ,include:

[0164] The data collection module is used to collect user interaction behavior data and application equipment operation and maintenance data in real time under the application scenario;

[0165] The autonomous learning module is used to optimize the training of the built-in functional model of AI digital human by utilizing the latest collected user interaction behavior data and operation and maintenance data of application equipment to obtain the autonomous learning results of AI digital human.

[0166] The beneficial effects of the above technologies are: real-time collection of user interaction behavior data and equipment operation and maintenance data, timely acquisition of the latest and most authentic information, and provision of rich and timely materials for AI digital human learning. Optimizing and training the built-in functional models using the latest data can help AI digital humans quickly adapt to changing application scenarios and user needs. The autonomous learning results obtained can improve the quality and effectiveness of AI digital humans’ interactions with users, and provide services that better meet user expectations. It can enhance AI digital humans’ understanding and response capabilities to the operating conditions of application equipment, and improve the stability and reliability of the system. It helps to continuously improve the functions of AI digital humans, making them more adaptable and competitive.

[0167] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. The AI ​​digital human autonomous learning method based on deep learning is characterized by: include: S1: Real-time collection of user interaction behavior data and application equipment operation and maintenance data in application scenarios; S2: Utilize the latest collected user interaction behavior data and application equipment operation and maintenance data to optimize the built-in functional model of AI digital human and obtain the autonomous learning results of AI digital human.

2. The AI ​​digital human autonomous learning method based on deep learning according to claim 1 is characterized in that: S1: Real-time collection of user interaction behavior data and application equipment operation and maintenance data in application scenarios, including: Collect user voice interaction behavior data, user face interaction behavior data, user touch interaction behavior data, and user online and offline interaction behavior data in real time during each complete interaction process in the application scenario to obtain user interaction behavior data in the application scenario; Collect historical operation data, operation performance data and manual operation and maintenance data of application equipment in real time as operation and maintenance data of application equipment.

3. The AI ​​digital human autonomous learning method based on deep learning according to claim 2 is characterized in that: Collect user voice interaction behavior data, user face interaction behavior data, user touch interaction behavior data, and user online and offline interaction behavior data in real time during each complete interaction process in the application scenario to obtain user interaction behavior data in the application scenario, including: Collect user voice interaction behavior data, user face interaction behavior data, user touch interaction behavior data, and user online and offline interaction behavior data in real time during each complete interaction process in the application scenario; Generate the identity information of each complete interaction process based on the user's facial interaction behavior data and the user's online and offline interaction behavior data during each complete interaction process; The identity information of each complete interaction process is bound and stored with the user voice interaction behavior data, user face interaction behavior data, user touch interaction behavior data and user online and offline interaction behavior data of the corresponding complete interaction process to obtain the user interaction behavior data of each complete interaction process in the application scenario.

4. The AI ​​digital human autonomous learning method based on deep learning according to claim 3 is characterized in that: Based on the user's facial interaction behavior data and the user's online and offline interaction behavior data in each complete interaction process, the identity information of each complete interaction process is generated, including: Extract the valid facial data of each complete interaction process from the user facial interaction behavior data of each complete interaction process, and extract the system login information or payment information of each complete interaction process from the user online and offline interaction behavior data of each complete interaction process; Based on the valid facial data and system login information or payment information of each complete interaction process, the identity information of each complete interaction process is generated.

5. The AI ​​digital human autonomous learning method based on deep learning according to claim 1 is characterized in that: S2: Utilize the latest collected user interaction behavior data and application equipment operation and maintenance data to optimize the built-in functional model of AI digital human and obtain the autonomous learning results of AI digital human, including: Use the latest collected user interaction behavior data to optimize and train the product recommendation model and product demand prediction model currently built into the AI ​​digital human to obtain optimized product recommendation model and product demand prediction model; Use the latest collected operation and maintenance data of application equipment to optimize the equipment prediction operation and maintenance model of AI digital human to obtain an optimized equipment prediction operation and maintenance model; Among them, the built-in functional models of AI digital humans include: product recommendation model, product demand prediction model, and equipment prediction and operation and maintenance model.

6. The AI ​​digital human autonomous learning method based on deep learning according to claim 5 is characterized in that: The latest collected user interaction behavior data is used to optimize and train the product recommendation model and product demand prediction model currently built into the AI ​​digital human, and the optimized product recommendation model and product demand prediction model are obtained, including: Based on the latest collected user interaction behavior data of each complete interaction process, all the final purchased products of the user and the purchase intention index of each product are analyzed; Based on the latest collected identity information of each complete interaction process, all final purchased products, and the purchase intention index of each product, the product recommendation training sample set corresponding to the identity information is updated; Input the product recommendation training sample set of the updated identity information into the built-in product recommendation model of the AI ​​digital human for model optimization training to obtain an optimized product recommendation model; Based on the latest collected identity information of all complete interaction processes, all final purchased products, and the purchase intention index of each product, the product demand prediction training sample set is updated to obtain a new product demand training sample set; The latest product demand training sample set is input into the product demand forecasting model built into the AI ​​digital human for model optimization training to obtain an optimized product demand forecasting model.

7. The AI ​​digital human autonomous learning method based on deep learning according to claim 6 is characterized in that: Based on the latest collected user interaction behavior data of each complete interaction process, all the user's final purchase products and the purchase intention index of each product are analyzed, including: Determine all final purchase products of the user based on the latest collected user interaction behavior data of each complete interaction process, and set the purchase intention index of each final purchase product to 1; Treat all available products, except for all final purchased products in each complete interaction process, as unpurchased products; Based on the most recently collected user interaction behavior data of each complete interaction process, determine the browsing time, viewing times, consultation frequency, comparison times with other products, total number of comparison information items, and consultation initiation initiative index of each unpurchased product by the user in each complete interaction process; Based on the browsing time, viewing times, consultation frequency, number of comparisons with other products, total number of comparison information items and consultation initiation initiative index of each unpurchased product during each complete interaction process, the purchase intention index of each unpurchased product is determined.

8. The AI ​​digital human autonomous learning method based on deep learning according to claim 7 is characterized in that: Based on the browsing time, viewing times, consultation frequency, number of comparisons with other products, total number of comparison information items, and consultation initiation initiative index of each unpurchased product during each complete interaction process, the purchase intention index of each unpurchased product is determined, including: Determine the longest browsing time, maximum number of views, maximum consultation frequency, maximum number of comparisons with other products, and maximum total number of comparison items for all historically purchased products of the user; Based on the longest browsing time, maximum number of views, maximum consultation frequency, maximum number of comparisons with other products, maximum total number of comparison items of all historically purchased products of the user, the browsing time, view number, consultation frequency, number of comparisons with other products, total number of comparison information items of each unpurchased product of the user, and the consultation initiation initiative index, determine the calculated reference browsing time, calculated reference view number, calculated reference consultation frequency, calculated reference comparison number with other products, and total number of calculated reference comparison information items of each unpurchased product of the user; Based on the user's browsing time, viewing times, consultation frequency, comparison times with other products, total number of comparison information items, and consultation initiation initiative index for each unpurchased product, the purchase intention index of each unpurchased product is calculated: Where ρ is the purchase intention index of the unpurchased product, t 0·scan is the reference browsing time for users to calculate the unpurchased products, t scan·max The longest browsing time of all historically purchased products of the user, n 0·ov The number of times users have viewed products that they have not purchased, n ov·max The maximum number of views for all historical purchases of a user, f 0·oc is the frequency of users’ reference consultation for products they have not purchased, f oc·max The maximum consultation frequency of all historical purchases of the user, n 0·oc The number of times users compare unpurchased products with other products, n oc·max The maximum number of comparisons with other products for all historically purchased products of the user, n 0·iifc is the total number of comparison information items of the user for the unpurchased products, n iifc·max The maximum total number of comparison items for all historical purchases of the user, Ac oci Initiate proactive index for users’ inquiries about products they have not purchased.

9. The AI ​​digital human autonomous learning method based on deep learning according to claim 5 is characterized in that: The latest collected operation and maintenance data of application equipment is used to optimize the equipment prediction operation and maintenance model of AI digital human to obtain an optimized equipment prediction operation and maintenance model, including: Based on the manual operation and maintenance data in the latest collected operation and maintenance data of the application equipment, the historical operation data and operation performance data between each group of adjacent manual operation and maintenance records and the manual operation and maintenance data of the latest manual operation and maintenance record in the corresponding group of adjacent manual operation and maintenance records are extracted from the latest collected operation and maintenance data of the application equipment; The historical operation data and operation performance data between each group of adjacent manual operation and maintenance records and the manual operation and maintenance data of the latest manual operation and maintenance record in the corresponding group of adjacent manual operation and maintenance records are taken as a single training sample. All training samples are input into the equipment predictive operation and maintenance model of AI digital human for model optimization training to obtain an optimized equipment predictive operation and maintenance model.

10. The AI ​​digital human autonomous learning system based on deep learning is characterized by: The method for autonomously learning an AI digital human based on deep learning according to any one of claims 1 to 9 comprises: The data collection module is used to collect user interaction behavior data and application equipment operation and maintenance data in real time under the application scenario; The autonomous learning module is used to optimize the training of the built-in functional model of AI digital human by utilizing the latest collected user interaction behavior data and operation and maintenance data of application equipment to obtain the autonomous learning results of AI digital human.

Images