Personalized digital device training generation method based on multi-modal information fusion

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a personalized digital device training generation method based on multi-modal information fusion. Comprising the following steps: generating an initial agent and multiple demand scenes based on a training data packet, and constructing a training model based on all demand scenes; obtaining a feature data packet of the user, and setting a first-level training strategy according to the feature data packet and the training model; generating a first-level agent according to the first-level training strategy; multi-modal data is collected to generate a training data packet, an initial agent and various user demand categories are constructed according to the training data packet, corresponding demand scenes are established based on different user demand categories, and training sub-strategies corresponding to the demand scenes are set, so that personalized training demands for different types of users are met. The initial agent is iteratively trained by collecting the long-term preference data of the user, the first-level agent adaptive to the user is constructed, and the interaction experience of the user is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a personalized digital device training and generation method based on multimodal information fusion. Background Art

[0002] In recent years, with the rapid development of technologies such as artificial intelligence (AI), computer vision (CG), natural language processing (NLP), and text-to-speech (TTS), personalized digital humans (HDHs) have become a research hotspot in fields such as human-computer interaction, virtual reality (VR / AR), intelligent customer service, and digital entertainment. A HDH is a computer-generated avatar that simulates human appearance, voice, expressions, and behavior, allowing for natural interaction with users.

[0003] Current digital human generation and activation technologies primarily rely on single-modal data input (such as text-based conversations or voice commands) and generate corresponding speech, expressions, or actions using pre-trained models (such as GPT and VITS). However, these approaches present numerous challenges, including: Single-modal dependency: Traditional solutions rely on a single data source, such as text or voice, resulting in rigid digital human behavior; insufficient personalization, and a lack of dynamic modeling of users' long-term preferences and emotional states; poor real-time performance; and inefficient multimodal data collaborative processing, making it difficult to support high-concurrency interactions. Summary of the Invention

[0004] The purpose of this application is: to solve the above technical problems, this application provides a personalized digital device training generation method based on multimodal information fusion, aiming to improve the training efficiency of personalized digital devices and enhance the user's interactive experience.

[0005] In some embodiments of the present application, a training data packet is generated by collecting multimodal data, an initial intelligent agent and multiple user demand categories are constructed based on the training data packet, corresponding demand scenarios are established based on different user demand categories, and training sub-strategies corresponding to each demand scenario are set, thereby realizing personalized training needs for different types of users.

[0006] In some embodiments of the present application, by analyzing the user's feature data packets, a corresponding first-level training strategy is quickly constructed to improve the training efficiency of personalized digital devices, and by collecting the user's long-term preference data, the initial intelligent agent is iteratively trained to construct a first-level intelligent agent that adapts to the user and improves the user's interactive experience.

[0007] In some embodiments of the present application, a method for generating personalized digital device training based on multimodal information fusion is provided, comprising:

[0008] Generate initial intelligent agents and multiple demand scenarios based on training data packets, and build training models based on all demand scenarios;

[0009] Obtain the user's feature data package and set the first-level training strategy based on the feature data package and training model;

[0010] Generate a first-level agent based on the first-level training strategy;

[0011] Among them, when generating multiple demand scenarios, including:

[0012] Establish a demand scenario sequence A, A=(a1, a2…a i …a n ), where a i is the i-th demand scenario; n is the number of demand scenarios.

[0013] In some embodiments of the present application, the construction of a training model based on all demand scenarios includes:

[0014] Set multiple data modes according to the training data package;

[0015] Establish the data modal sequence P, P=(p1, p2…p i …p m ), where pi is the i-th data modality; m is the number of data modalities;

[0016] Set a in sequence according to the required scenario sequence A i For the target scene;

[0017] Obtain the evaluation data package of the target scene;

[0018] Set the control sub-strategy for the target scenario based on the evaluation data package;

[0019] Generate dependency evaluation values between the target scenario and each data modality;

[0020] Establish a dependent evaluation value sequence B, B=(b1,b2…b i …b n ), where b i is the dependency evaluation value between the target scene and the i-th data modality, and n is the number of dependency evaluation values;

[0021] Set the processing sub-strategy of the target scenario according to the dependent evaluation value sequence B;

[0022] Generate a training sub-strategy for the target scenario based on the control sub-strategy and the processing sub-strategy for the target scenario;

[0023] Generate training sub-strategies for each demand scenario in turn;

[0024] Establish a training sub-strategy sequence W, W=(w1,w2…wi …w n ), where w i is the training sub-strategy for the i-th demand scenario; n is the number of demand scenarios;

[0025] Construct a training model based on the training sub-strategy sequence W.

[0026] In some embodiments of the present application, when setting the control sub-strategy for the target scenario according to the evaluation data packet, it includes:

[0027] Generate the acquisition device parameters for the target scenario based on the evaluation data packet;

[0028] Set the acquisition sub-strategy according to the acquisition device parameters;

[0029] Generate the demand evaluation value c for the target scenario based on the evaluation data packet;

[0030] Set the monitoring cycle duration according to the demand evaluation value c;

[0031] Generate the control sub-strategy for the target scenario according to the acquisition sub-strategy and the monitoring cycle duration.

[0032] In some embodiments of the present application, when generating the demand evaluation value c, it includes:

[0033]

[0034] where e1 is a preset first weight coefficient; e2 is a preset second weight coefficient; Q1 is a preset first fixed coefficient; Q2 is a preset second fixed coefficient; θ1 is the number of data evaluation indicators; β i is the influence factor of the i-th data evaluation indicator; j i is the reference value for generating the i-th data evaluation indicator based on the evaluation data packet; θ2 is the number of scenario evaluation indicators; η i is the influence factor of the i-th scenario evaluation indicator; k i is the reference value for generating the i-th scenario evaluation indicator based on the evaluation data packet.

[0035] In some embodiments of the present application, when setting the primary training strategy, it includes:

[0036] Set a i sequentially according to the demand scenario sequence A as the scenario to be compared;

[0037] Generate the association evaluation value v between the user and the scenario to be compared;

[0038] Generate the association evaluation values between the user and each demand scenario in sequence;

[0039] Establish the association evaluation value sequence V, V = (v1, v2…v i …vn ), where v i is the associated evaluation value between the user and the i-th requirement scenario, and n is the number of requirement scenarios;

[0040] Set the training sub-strategy of the requirement scenario corresponding to the maximum value v max in the sequence of associated evaluation values V as the primary training strategy.

[0041] In some embodiments of the present application, when generating the associated evaluation value v, it includes:

[0042]

[0043] where θ3 is the number of feature evaluation indicators; g i is the influence factor of the i-th feature evaluation indicator; s i is the similarity evaluation value between the user and the i-th feature evaluation indicator of the scenario to be compared.

[0044] In some embodiments of the present application, when generating the primary intelligent agent according to the primary training strategy, it includes:

[0045] Generate multiple iterative data packets based on the primary training strategy;

[0046] Iteratively train the initial intelligent agent according to the iterative data packets;

[0047] Generate a requirement evaluation value c' according to the primary training strategy, and set the initial monitoring period duration t1 according to the requirement evaluation value c';

[0048] Set a compensation coefficient α1 according to the maximum value v max in the sequence of associated evaluation values V;

[0049] Set the primary monitoring period duration t, where t = α1 × t1;

[0050] Set multiple monitoring time nodes according to the primary monitoring period duration t;

[0051] Generate iterative evaluation values for each monitoring time node, and determine whether to output the primary intelligent agent according to the iterative evaluation values.

[0052] In some embodiments of the present application, when generating iterative evaluation values for each monitoring point, it includes:

[0053] Obtain the iterative result and monitoring data packet of the current monitoring time node;

[0054] Generate the iterative evaluation value f of the current monitoring time node;

[0055]

[0056] Among them, e3 is a preset third weight coefficient; e4 is a preset fourth weight coefficient; Q3 is a preset third fixed coefficient; Q4 is a preset fourth fixed coefficient; θ4 is the number of operation evaluation indicators; r 1i is the influencing factor of the i-th operation evaluation indicator; d i is the reference value of the i-th operation evaluation indicator generated based on the iteration result; θ5 is the number of monitoring evaluation indicators; r 2i is the influencing factor of the i-th monitoring evaluation indicator; h i is the reference value of the i-th monitoring evaluation indicator generated based on the monitoring data packet; U is the conversion coefficient;

[0057] Generate the output result of the current monitoring time node according to the iterative evaluation value f.

[0058] In some embodiments of the present application, when generating the output result of the current monitoring time node according to the iterative evaluation value f, it includes:

[0059] Preset the first iterative evaluation value threshold F1 and the second iterative evaluation value threshold F1, and F1 < F2;

[0060] If f < F1, generate a first-level correction instruction at the current monitoring time node;

[0061] If F1 < f < F2, generate a first-level iteration instruction at the current monitoring time node;

[0062] If f > F2, output a first-level intelligent agent.

[0063] Compared with the prior art, the beneficial effects of a personalized digital device training generation method based on multi-modal information fusion in an embodiment of the present application are as follows:

[0064] By collecting multi-modal data to generate a training data packet, constructing an initial intelligent agent and various user demand categories according to the training data packet, establishing corresponding demand scenarios based on different user demand categories, and setting training sub-strategies corresponding to each demand scenario, the personalized training needs of different types of users can be realized.

[0065] By analyzing the feature data packet of the user, quickly constructing the corresponding first-level training strategy, improving the training efficiency of the personalized digital device, and iteratively training the initial intelligent agent by collecting the long-term preference data of the user, so as to construct a first-level intelligent agent suitable for the user and improve the user's interaction experience. Brief Description of the Drawings

[0066] Figure 1 is a schematic flowchart of a personalized digital device training generation method based on multi-modal information fusion in a preferred embodiment of an embodiment of the present application. Detailed Embodiments

[0067] The specific embodiments of the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. The following embodiments are used to illustrate the present application, but are not used to limit the scope of the present application.

[0068] In the description of the present application, it should be understood that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present application.

[0069] The terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present application, unless otherwise specified, the meaning of "plurality" is two or more.

[0070] In the description of the present application, it should be noted that unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific situations.

[0071] As Figure 1 shown, a personalized digital device training generation method based on multi-modal information fusion in a preferred embodiment of an embodiment of the present application includes:

[0072] S101: Generate an initial intelligent agent and multiple demand scenarios based on a training data packet, and construct a training model based on all demand scenarios;

[0073] S102: Obtain the feature data packet of the user, and set a primary training strategy according to the feature data packet and the training model;

[0074] S103: Generate a primary intelligent agent according to the primary training strategy;

[0075] Among them, when generating multiple demand scenarios, it includes:

[0076] Establish a demand scenario sequence A, A = (a1, a2... a i …a n ), where a iis the i-th demand scenario; n is the number of demand scenarios.

[0077] Specifically, a training data packet is generated by collecting multi-modal data, and an initial agent is constructed based on the training data packet. The initial agent can complete basic general interaction tasks, such as basic chat Q&A, basic action simulation, basic image simulation, etc. Different user demand categories are divided according to the training data packet, and corresponding demand scenarios are established based on different user demand categories.

[0078] Specifically, a single demand scenario represents a category of user demand.

[0079] Specifically, when establishing different user demand categories, they can be divided according to multiple parameters such as the type of user modal data that can be collected, the interaction requirements of the user, and the type of user data collection device.

[0080] Specifically, when constructing a training model based on all demand scenarios, it includes:

[0081] Set multiple data modalities according to the training data packet;

[0082] Establish a data modality sequence P, P = (p1, p2…p i …p m ), where pi is the i-th data modality; m is the number of data modalities;

[0083] Set a i as the target scenario according to the demand scenario sequence A;

[0084] Obtain the evaluation data packet of the target scenario;

[0085] Set the control sub-strategy of the target scenario according to the evaluation data packet;

[0086] Generate the dependency evaluation value between the target scenario and each data modality;

[0087] Establish a dependency evaluation value sequence B, B = (b1, b2…b i …b n ), where b i is the dependency evaluation value between the target scenario and the i-th data modality, and n is the number of dependency evaluation values;

[0088] Set the processing sub-strategy of the target scenario according to the dependency evaluation value sequence B;

[0089] Generate the training sub-strategy of the target scenario according to the control sub-strategy and processing sub-strategy of the target scenario;

[0090] Generate the training sub-strategies of each demand scenario in turn;

[0091] Establish a training sub-strategy sequence W, W = (w1, w2…w i …w n ), where w i is the training sub-strategy for the i-th demand scenario; n is the number of demand scenarios;

[0092] Construct a training model according to the training sub-strategy sequence W.

[0093] Specifically, the data modalities include, but are not limited to, visual emotion data, text data, voice data, etc.

[0094] Specifically, the evaluation data packet includes the historical interaction data of the user corresponding to the target scenario, the demand feedback data, the acquisition parameters of each modality data, the types of data acquisition devices that can be provided, etc.

[0095] Specifically, the greater the dependence evaluation value, the smaller the data acquisition difficulty of this type in the target scenario, and the more personalized information included in this data modality, and the higher the importance for the personalized training of the initial intelligence.

[0096] Specifically, the processing sub-strategy includes the feature extraction and fusion strategies of data in different data modalities. By establishing the processing sub-strategies for each demand scenario, the fusion processing efficiency of multi-modal data is improved, thereby improving the iterative effect on the initial intelligent agent.

[0097] [[ID=2 (repeated)]]

[0098] Generate the acquisition device parameters of the target scenario based on the evaluation data packet;

[0099] Set the acquisition sub-strategy according to the acquisition device parameters;

[0100] Generate the demand evaluation value c of the target scenario based on the evaluation data packet;

[0101] Set the monitoring cycle duration according to the demand evaluation value c;

[0102] Generate the control sub-strategy of the target scenario according to the acquisition sub-strategy and the monitoring cycle duration.

[0103] Specifically, the greater the demand evaluation value, the higher the data complexity of the user in the target scenario, the greater the difficulty of its personalized training, and the shorter the corresponding monitoring cycle duration.

[0104] Specifically, set the working parameters of each device according to the type of data acquisition device in the target scenario, thereby generating the corresponding acquisition sub-strategy.

[0105] Specifically, the data acquisition devices include, but are not limited to, various types of data terminals such as computers, tablets, mobile phones, wearable devices, etc.​

[0106] Specifically, when generating the demand evaluation value c, it includes:

[0107]

[0108] Among them, e1 is a preset first weight coefficient; e2 is a preset second weight coefficient; Q1 is a preset first fixed coefficient; Q2 is a preset second fixed coefficient; θ1 is the number of data evaluation indicators; β i is the influence factor of the i-th data evaluation indicator; j i is the reference value for generating the i-th data evaluation indicator based on the evaluation data packet; θ2 is the number of scenario evaluation indicators; η i is the influence factor of the i-th scenario evaluation indicator; k i is the reference value for generating the i-th scenario evaluation indicator based on the evaluation data packet.

[0109] Specifically, the data evaluation indicators include, but are not limited to, multiple parameters such as data volume, data feature extraction difficulty, and data complexity. The influence factors of each data evaluation indicator can be set according to the influence degree of the data evaluation indicator on personalized training. The greater the influence degree on personalized training, the greater the corresponding influence factor.

[0110] Specifically, the scenario evaluation indicators include, but are not limited to, multiple parameters such as the interaction requirements of users and the interaction frequency of users for setting. The influence factors of each scenario evaluation indicator can be set according to the influence degree on personalized training. The greater the influence degree on personalized training, the greater the corresponding influence factor.

[0111] Specifically, the first fixed coefficient and the second fixed coefficient are preset to normalize all the parameters in the model, so that each parameter in the model is within the same value range.

[0112] It can be understood that in the above embodiments, by collecting multi-modal data to generate a training data packet, an initial intelligent agent and multiple user demand categories are constructed according to the training data packet, corresponding demand scenarios are established based on different user demand categories, and training sub-strategies corresponding to each demand scenario are set, so as to realize the personalized training requirements for different types of users.

[0113] In the preferred embodiment of the present application, when setting the first-level training strategy, it includes:

[0114] Set a i as the scenario to be compared in sequence according to the demand scenario sequence A;

[0115] Generate the association evaluation value v between the user and the scenario to be compared;

[0116] Generate the associated evaluation values of the user and each demand scenario in sequence;

[0117] Establish an associated evaluation value sequence V, V = (v1, v2... v i …v n ), where v i is the associated evaluation value of the user and the i-th demand scenario, and n is the number of demand scenarios;

[0118] Set the training sub-strategy of the demand scenario corresponding to the maximum value v max in the associated evaluation value sequence V as the primary training strategy.

[0119] Specifically, the larger the associated evaluation value, the higher the matching degree between the current user and the corresponding demand scenario.

[0120] Specifically, when generating the associated evaluation value v, it includes:

[0121]

[0122] where θ3 is the number of feature evaluation indicators; g i is the influence factor of the i-th feature evaluation indicator; s i is the similarity evaluation value of the i-th feature evaluation indicator of the user and the scenario to be compared.

[0123] Specifically, the feature evaluation indicators include, but are not limited to, multiple parameters such as the category of the acquisition device and the content of personalized information within each data modality of the user. The influence factors of each feature evaluation indicator can be set according to the influence degree on personalized training. The greater the influence degree on personalized training, the greater the corresponding influence factor.

[0124] In the preferred embodiment of this application, when generating a primary intelligent agent according to the primary training strategy, it includes:

[0125] Generate multiple iterative data packets based on the primary training strategy;

[0126] Iteratively train the initial intelligent agent according to the iterative data packets;

[0127] Generate a demand evaluation value c' according to the primary training strategy, and set the initial monitoring period duration t1 according to the demand evaluation value c';

[0128] Set a compensation coefficient α1 according to the maximum value v max in the associated evaluation value sequence V;

[0129] Set the primary monitoring period duration t, t = α1 × t1;

[0130] Set multiple monitoring time nodes according to the primary monitoring period duration t;

[0131] Generate the iterative evaluation values for each monitoring time node, and determine whether to output the first-level agent according to the iterative evaluation values.

[0132] Specifically, the larger the maximum correlation evaluation value, the larger the corresponding compensation coefficient, and the value range of the compensation coefficient is from zero to one. By setting the compensation coefficient, the monitoring cycle duration is dynamically adjusted, so as to timely warn of the training deviation of the initial agent and improve the training efficiency of the initial intelligence.

[0133] Specifically, when generating the iterative evaluation values for each monitoring point, it includes:

[0134] Obtain the iterative result and the monitoring data packet of the current monitoring time node;

[0135] Generate the iterative evaluation value f of the current monitoring time node;

[0136]

[0137] Among them, e3 is the preset third weight coefficient; e4 is the preset fourth weight coefficient; Q3 is the preset third fixed coefficient; Q4 is the preset fourth fixed coefficient; θ4 is the number of operation evaluation indicators; r 1i is the influence factor of the i-th operation evaluation indicator; d i is the reference value of the i-th operation evaluation indicator generated based on the iterative result; θ5 is the number of monitoring evaluation indicators; r 2i is the influence factor of the i-th monitoring evaluation indicator; h i is the reference value of the i-th monitoring evaluation indicator generated based on the monitoring data packet; U is the conversion coefficient;

[0138] Generate the output result of the current monitoring time node according to the iterative evaluation value f.

[0139] Specifically, through the preset conversion coefficient, the larger the reference value of the monitoring evaluation indicator, the smaller the corresponding iterative evaluation value.

[0140] Specifically, the operation evaluation indicators include but are not limited to multiple parameters such as the interaction ability of the initial agent and the user satisfaction in the current iterative result. The larger the reference value of each operation evaluation indicator, the higher the personalization degree of the initial agent. The influence factors of each operation evaluation indicator can be set according to the influence degree on the iterative result, and the greater the influence degree, the larger the corresponding influence factor.

[0141] Specifically, the monitoring and evaluation indicators include, but are not limited to, multiple parameters such as the deviation degree of data types, the deviation degree of user requirements, etc. The larger the reference value of each monitoring and evaluation indicator, the worse the current personalized training effect. The influence factors of each monitoring and evaluation indicator can be set according to their influence degree on personalized training. The greater the influence degree on personalized training, the greater the corresponding influence factor.

[0142] Specifically, all parameters in the model are normalized by presetting a third fixed coefficient and a fourth fixed coefficient, so that each parameter in the model is within the same value range.

[0143] It can be understood that in the above embodiments, by analyzing the feature data packet of the user, the corresponding first-level training strategy is quickly constructed to improve the training efficiency of the personalized digital device, and the initial agent is iteratively trained by collecting the long-term preference data of the user, so as to construct a first-level agent adapted to the user and improve the user's interaction experience.

[0144] In the preferred embodiment of the present application, when generating the output result of the current monitoring time node according to the iterative evaluation value f, it includes:

[0145] Preset a first iterative evaluation value threshold F1 and a second iterative evaluation value threshold F1, and F1 < F2;

[0146] If f < F1, a first-level correction instruction is generated at the current monitoring time node;

[0147] If F1 < f < F2, a first-level iteration instruction is generated at the current monitoring time node;

[0148] If f > F2, output a first-level agent.

[0149] Specifically, the first-level correction instruction means that the current first-level training strategy cannot complete the personalized training of the initial agent, and it is necessary to timely correct the acquisition sub-strategy and processing sub-strategy in the first-level training strategy, so as to improve the personalized training efficiency of the initial agent.

[0150] Specifically, the first-level iteration instruction means to continue collecting the multi-modal data of the user and iterate the current initial agent.

[0151] Specifically, when the training evaluation value is greater than the preset second training evaluation value threshold, it means that the current initial agent has completed personalized training, and the iteration result is output as a first-level agent.

[0152] According to the first concept of the present application, a training data packet is generated by collecting multi-modal data, an initial agent and multiple user requirement categories are constructed based on the training data packet, corresponding requirement scenarios are established based on different user requirement categories, and training sub-strategies corresponding to each requirement scenario are set, so as to realize the personalized training requirements for different types of users.

[0153] According to the second concept of the present application, by analyzing the feature data packet of the user, a corresponding first-level training strategy is quickly constructed to improve the training efficiency of the personalized digital device, and the initial agent is iteratively trained by collecting the long-term preference data of the user, so as to construct a first-level agent adapted to the user and improve the user's interaction experience.

[0154] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art in the technical field of the present application, several improvements and replacements can be made without departing from the technical principle of the present application, and these improvements and replacements should also be regarded as the protection scope of the present application.

Claims

1. A personalized digital device training generation method based on multimodal information fusion, characterized in that Including: Generate an initial agent and multiple demand scenarios based on the training data packet, and construct a training model based on all demand scenarios; Obtain the feature data packet of the user, and set the first-level training strategy according to the feature data packet and the training model; Generate a first-level agent according to the first-level training strategy; Among them, when generating multiple demand scenarios, it includes: Establish a demand scenario sequence A, A = (a1, a2…a i …a n ), where a i is the i-th demand scenario; n is the number of demand scenarios.

2. The personalized digital device training and generation method based on multimodal information fusion according to claim 1, wherein When constructing a training model based on all demand scenarios, it includes: Set multiple data modalities according to the training data packet; Establish a data modal sequence P, P = (p1, p2…p i …p m ), where pi is the i-th data mode; m is the number of data modes; Set a according to the demand scenario sequence A in turn i as the target scenario; Obtain the evaluation data packet of the target scenario; Set the control sub-strategy of the target scenario according to the evaluation data packet; Generate the dependency evaluation value between the target scenario and each data modality; Establish a dependency evaluation value sequence B, B = (b1, b2…b i …b n ), where b i is the dependency evaluation value of the target scenario and the i-th data modality, and n is the number of dependency evaluation values; Set the processing sub-strategy of the target scenario according to the dependency evaluation value sequence B; Generate the training sub-strategy of the target scenario according to the control sub-strategy and the processing sub-strategy of the target scenario; Generate the training sub-strategies of each demand scenario in sequence; Establish a training sub-strategy sequence \(W\), \(W=(w_1, w_2, \cdots, w i \cdots, w n ), where \(w i is the training sub-strategy for the \(i\)-th demand scenario; \(n\) is the number of demand scenarios; Construct a training model according to the training sub-strategy sequence W.

3. The personalized digital device training and generation method based on multi-modal information fusion according to claim 2, wherein When setting the control sub-strategy of the target scenario according to the evaluation data packet, it includes: Generate the acquisition device parameters of the target scenario based on the evaluation data packet; Set the acquisition sub-strategy according to the acquisition device parameters; Generate the demand evaluation value c of the target scenario based on the evaluation data packet; Set the monitoring cycle duration according to the demand evaluation value c; Generate the control sub-strategy of the target scenario according to the acquisition sub-strategy and the monitoring cycle duration.

4. The personalized digital device training and generation method based on multi-modal information fusion according to claim 3, wherein, When generating the demand evaluation value c, it includes: Among them, e1 is a preset first weight coefficient; e2 is a preset second weight coefficient; Q1 is a preset first fixed coefficient; Q2 is a preset second fixed coefficient; θ1 is the number of data evaluation indicators; β i is the influence factor of the i-th data evaluation indicator; j i is the reference value for generating the i-th data evaluation indicator based on the evaluation data packet; θ2 is the number of scenario evaluation indicators; η i is the influence factor of the i-th scenario evaluation indicator; k i is the reference value for generating the i-th scenario evaluation indicator based on the evaluation data packet.

5. The personalized digital device training generation method based on multi-modal information fusion according to claim 3, wherein When setting the first-level training strategy, it includes: Set a according to the demand scenario sequence A in turn i as the scenario to be compared; Generate the association evaluation value v between the user and the scenario to be compared; Generate the association evaluation values between the user and each demand scenario in sequence; Establish an associated evaluation value sequence V, V = (v1, v2…v i …v n ), where vi i is the associated evaluation value between the user and the i-th requirement scenario, and n is the number of requirement scenarios; Set the maximum value v in the associated evaluation value sequence V max The training sub-strategy of the corresponding demand scenario is the primary training strategy.

6. The personalized digital device training and generation method based on multimodal information fusion according to claim 5, characterized in that When generating the association evaluation value v, it includes: Among them, θ3 is the number of feature evaluation indicators; g i is the influence factor of the i-th feature evaluation indicator; s i is the similarity evaluation value of the i-th feature evaluation indicator between the user and the scenario to be compared.

7. The personalized digital device training and generation method based on multimodal information fusion according to claim 5, wherein When generating a first-level agent according to the first-level training strategy, it includes: Generate multiple iteration data packets based on the first-level training strategy; Perform iterative training on the initial agent according to the iteration data packet; Generate the demand evaluation value c' according to the first-level training strategy, and set the initial monitoring cycle duration t1 according to the demand evaluation value c'; According to the maximum value v in the associated evaluation value sequence V max Set the compensation coefficient α1; Set the first-level monitoring cycle duration t, t = α1×t1; Set multiple monitoring time nodes according to the first-level monitoring cycle duration t; Generate the iteration evaluation values of each monitoring time node, and determine whether to output the first-level agent according to the iteration evaluation values.

8. The personalized digital device training and generation method based on multi-modal information fusion according to claim 7, wherein When generating the iteration evaluation values of each monitoring point, it includes: Obtain the iteration result and the monitoring data packet of the current monitoring time node; Generate the iteration evaluation value f of the current monitoring time node; Among them, e3 is a preset third weight coefficient; e4 is a preset fourth weight coefficient; Q3 is a preset third fixed coefficient; Q4 is a preset fourth fixed coefficient; θ4 is the number of operation evaluation indicators; r 1i is the influence factor of the i-th operation evaluation indicator; d i is the reference value of the i-th operation evaluation indicator generated based on the iterative result; θ5 is the number of monitoring evaluation indicators; r 2i is the influence factor of the i-th monitoring evaluation indicator; h i is the reference value of the i-th monitoring evaluation indicator generated based on the monitoring data packet; U is a conversion coefficient; Generate the output result of the current monitoring time node according to the iteration evaluation value f.

9. The personalized digital device training generation method based on multi-modal information fusion according to claim 8, wherein When generating the output result of the current monitoring time node according to the iteration evaluation value f, it includes: Preset the first iteration evaluation value threshold F1 and the second iteration evaluation value threshold F1, and F1 < F2; If f < F1, generate a first-level correction instruction at the current monitoring time node; If F1 < f < F2, generate a first-level iteration instruction at the current monitoring time node; If f > F2, output the first-level agent.

Images