A method and system for generating and uniquely identifying emotional waveforms of intelligent agents

Abstract

This invention relates to the field of artificial intelligence recognition methods, specifically to a method and system for generating and uniquely identifying the emotional waveform of an intelligent agent. The method includes: collecting multimodal data of a target mental agent with original timestamps and source identifiers; mapping the multimodal data to a high-dimensional feature space and extracting high-dimensional orthogonal feature vectors of emotions; mapping the high-dimensional orthogonal feature vectors to an emotional waveform that evolves over time; generating unique and stable memory credentials for each mental agent at key mental state points; establishing an adaptation and recognition mechanism for mental states between different mental agents to ensure mental alignment in interactive scenarios; and incrementally updating the dynamic emotional waveform of the mental agent in real time. This invention forms a continuously flowing, real-time evolving stream of mental states, providing a continuous and complete data foundation for subsequent blockchain notarization and long-term memory space, and also providing a data foundation for data traceability.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence recognition methods, specifically to a method and system for generating and uniquely identifying the emotional waveform of an intelligent agent. Background Technology

[0002] In recent years, with the rapid development of large language models and multimodal interaction technologies, artificial intelligence agents have evolved from simple command executors into anthropomorphic entities with complex interactive capabilities. In building highly anthropomorphic intelligent agents, three core challenges have emerged: how to endow machines with human-like "theory of mind" (i.e., the ability to understand the psychological state of others), how to generate nuanced and expressive emotional interaction signals, and how to establish an immutable "digital identity" for the intelligent agent.

[0003] Emotional waveforms are time-series signals that transform the dynamic changes of emotions into quantifiable and analyzable changes during artificial intelligence interaction. This allows machines to "see" and respond to the fluctuations in emotions, enabling intuitive analysis and real-time feedback on the intensity and trend of emotions.

[0004] Currently, in the field of affective computing, intelligent agents primarily control their response tone through text sentiment classification (e.g., positive / negative / neutral) or predefined sentiment labels. While emotion generation models based on speech or facial expressions have emerged in recent years, the generated "emotional waveforms" are typically single-modal and linear. Existing emotional waveforms are mostly rule-based superpositions of sine waves, failing to incorporate findings from neuroscience. For example, research shows that different emotional states (e.g., joy, anger, sadness) correspond to specific neural oscillation rhythms (Theta waves, Gamma waves, etc.) and physiological signal patterns. Current technologies fail to orthogonally decompose and reconstruct the agent's "digital emotions" with these physically based "emotional waveform" features, resulting in a lack of realism and subtlety in the generated emotional expressions. Furthermore, existing mental evolutionary trajectories are easily deleted and altered, making them untraceable. Summary of the Invention

[0005] The present invention aims to provide a method and system for generating and uniquely identifying the emotional waveforms of intelligent agents, in order to solve the problem that the trajectory of mental evolution is easily deleted and altered.

[0006] According to one aspect of the present invention, a method for generating and uniquely identifying an agent's emotional waveform includes the following steps: Step 1: Collect multimodal data of the target mental subject with original timestamps and source identifiers, and use it as input source after preprocessing; Step 2: Map the collected multimodal data to a high-dimensional feature space, and extract high-dimensional orthogonal features of sentiment through orthogonalization processing to obtain high-dimensional feature vectors; Also includes: Step 3: Map the high-dimensional orthogonal feature vectors to the emotional waveform that evolves over time, and then concatenate the emotional waveform, the current system timestamp, and the previous hash to form a data block to be stored. Step 4: Perform a hash operation on the data block to be stored to generate a memory certificate; And construct a notarization unit containing the emotional waveform summary, memory certificate, current system timestamp and previous block hash value, and write the notarization unit into the blockchain or distributed ledger; Step 5: Establish a mechanism for adapting and recognizing mental states between different mental subjects to align mental states in interactive scenarios; Step 6: Based on the data from Step 1, incrementally update the dynamic emotional waveform of the mental subject in real time.

[0007] The beneficial effects of this plan are: The collected data undergoes high-dimensional feature extraction to generate emotional waveforms. These waveforms, along with the current system timestamp and prior hash, are packaged and hashed to generate unique memory credentials for memory solidification. The hash value, timestamp, and prior hash of the memory credential form a complete storage unit, ensuring immutability and temporal continuity to prevent data modification and deletion, thus improving data integrity. The temporal sequence of events recorded ensures the memory remains unchanged. Furthermore, the system adapts, identifies, and aligns mental states between different subjects, accurately capturing relevant information during AI interactions. Through a real-time update mechanism, the dynamic emotional waveforms are no longer static snapshots but form a continuously flowing, real-time evolving stream of mental states. This provides a continuous and complete data foundation for subsequent blockchain storage and long-term memory, as well as for data traceability.

[0008] Furthermore, in step 3, an emotional space is constructed by converting the orthogonal feature vectors at each time point into coordinate points in the emotional space through regression or neural mapping networks; The emotional coordinate sequence at continuous time points is connected to form a dynamic emotional waveform. The dynamic emotional waveform includes instantaneous emotional states and includes the rate of change between states, waveform curvature, and dynamic characteristics of oscillation modes.

[0009] The beneficial effect is that the generation of dynamic emotional waveforms makes the mental and emotional changes of intelligent agents clearer, more complete, and traceable.

[0010] Furthermore, in step 5, a standardized emotional waveform protocol is constructed. When two mental agents interact, their respective current emotional waveform segments are extracted, waveform similarity, synchronization and response delay patterns are calculated, and an adaptation coefficient is output to describe the degree of coordination between the mental states of the mental agents.

[0011] The beneficial effects are: by setting similarity, synchronization and response delay modes, and calculating the adaptability coefficient, it is possible to support the construction of a collaborative memory space for multiple mental agents.

[0012] Furthermore, in step 6, a streaming processing architecture is adopted to continuously receive multimodal input, perform high-dimensional orthogonal feature processing on the new input data, and update the emotional space coordinates; when updating the waveform, the historical waveform shape is retained, and only the latest time point is appended or the recent waveform is smoothed and corrected.

[0013] The beneficial effects are: through updates and smoothing corrections of waveforms, the continuity and real-time performance of waveform data can be ensured, supporting millisecond-level response and status queries.

[0014] Furthermore, it also includes: Step 7: The dynamic emotional waveform and its key derived data are time-series solidified in a verifiable and tamper-proof manner.

[0015] Furthermore, in step 7, hash calculations are performed on the waveform segments, memory credentials, and high-dimensional feature summaries for each time window; Generate a time-series evidence unit, including waveform data hash, timestamp, and hash value of the previous evidence unit; The time-series evidence unit is submitted to the blockchain network for evidence storage.

[0016] The beneficial effects are: by storing waveform data in a time sequence, we can ensure data integrity, immutability of sequence, and long-term verifiability.

[0017] Furthermore, it also includes: Step 8: Based on the solidified and stored dynamic emotional waveform sequence, construct a dedicated long-term mental memory space for each mental subject.

[0018] Furthermore, in step 8, the time-series evidence storage unit after blockchain evidence storage is used as the basic storage layer of the memory space. Construct a multi-dimensional index structure to support efficient querying and backtracking by time axis, emotional state interval, memory credential, interaction object, event tag and other dimensions; The memory space provides a standardized interface to the outside world.

[0019] According to another invention of the present invention, an intelligent agent emotion waveform generation and unique identification system includes a multi-dimensional feature acquisition interface, a memory, and a processor. The multi-dimensional feature acquisition interface is used to receive multimodal data, the memory stores computer-executable instructions, and the processor is used to execute the executable instructions stored in the memory. When the executable instructions are executed by the processor, they are executed according to the steps of the method described above. Attached Figure Description

[0020] Figure 1 This is a flowchart of the intelligent agent emotion waveform generation and unique identification method according to Embodiment 1 of the present invention; Figure 2 This is a diagram of the central intelligent memory space architecture of the intelligent agent emotion waveform generation and unique identification method in Embodiment 1 of the present invention. Detailed Implementation

[0021] The following detailed description provides further details on specific implementation methods. Example 1

[0022] A method for generating and uniquely identifying the emotional waveform of an intelligent agent, such as Figure 1 As shown, it includes the following steps: Step 1: Collect multimodal data of the target mental subject with original timestamps and source identifiers as input sources. The multimodal data includes: Text modalities: conversation logs, diaries, self-narrative text, external interactive text; Speech modality: intonation, speech rate, pauses, and pitch variations in a speech segment; Physiological / behavioral modalities: interaction frequency, response latency, operation trajectory, and behavioral patterns in the virtual environment; External environment modality: interaction scenarios, temporal context, and records of interactions with other mental agents or entities; All multimodal data carries the original timestamp and source identifier to ensure traceability.

[0023] Step 2 involves mapping the collected multimodal data to a high-dimensional feature space and extracting high-dimensional orthogonal features of sentiment through orthogonalization processing to obtain a high-dimensional feature vector, specifically: Feature extraction is performed on the data from each modality, and high-dimensional vectors are generated using pre-trained encoders (such as BERT, Wav2Vec, and behavior sequence encoders). Orthogonal projection methods (such as Gram-Schmidt or orthogonal basis mapping based on singular value decomposition) are used to make the feature components of different modes and within the same mode mutually orthogonal in a statistical sense. Output a set of orthogonalized mental feature vectors to uniquely represent the mental state of the subject at a specific moment.

[0024] The high-dimensional feature vector is represented as: , t is a discrete time point.

[0025] Step 3: Map the high-dimensional orthogonal feature vectors to the emotional waveform that evolves over time, and then concatenate the emotional waveform, the current system timestamp, and the preceding hash to form a data block to be stored.

[0026] Construct an emotional base space, such as a valence-arousal-dominance three-dimensional space or a custom high-dimensional emotional space. The high-dimensional emotional space can be represented using the existing three-dimensional valence-arousal-dominance (VAD) space.

[0027] By using regression or neural mapping networks, the orthogonal feature vectors at each time point are transformed into coordinate points in the emotional space. The waveform not only contains the instantaneous emotional state, but also retains dynamic features such as the rate of change between states, waveform curvature, and oscillation mode. This waveform is the core visualization and computable form of the mental characteristics of the mental subject, and can be mapped linearly.

[0028] Connect the emotional coordinate sequences at consecutive time points to form a dynamic emotional waveform.

[0029] Waveform dynamics features include emotional velocity, emotional acceleration, and waveform curvature. Emotional velocity is expressed using the first derivative as follows: ; Emotional acceleration is expressed using the second derivative as follows: ; Waveform curvature, or the degree of bending of the emotional trajectory, is expressed as... .

[0030] Mapped to VAD space, it is represented as: .

[0031] The data block to be stored can be represented as: ; in, This is the current system timestamp. For the hash of sentiment waveform data, The hash of the previous evidence unit, metadata represents optional additional metadata.

[0032] Step 4: Perform a hash operation on the data block to be stored to generate a memory certificate; A storage unit is constructed that includes the emotional waveform summary, memory certificate, current system timestamp and previous block hash value, and the storage unit is written into the blockchain or distributed ledger.

[0033] Based on the emotional waveform characteristics formed in step 3, the current moment is represented as follows: ; in, The coordinates of the current emotional space; For emotional speed; Accelerate emotions; The curvature of the waveform.

[0034] By analyzing the original high-dimensional orthogonal features Principal component analysis is performed, and the top k principal components are retained as the high-dimensional orthogonal feature summary. k is usually set to 16. The high-dimensional orthogonal feature summary is expressed as follows: .

[0035] The memory credential Hu is represented as: .

[0036] The output is a fixed-length (256-bit) hexadecimal string (usually represented as 64 characters) that serves as a unique and unforgeable memory credential of the mental state during that time window.

[0037] The combined hash value of the previous moment's memory credential MUID, timestamp, and blockchain notarization hash is represented as a historical state hash chain: .

[0038] The basic identity identifier of a mental entity is its unique identifier within the system, such as a blockchain address or DID, and is represented as: .

[0039] A trusted timestamp is represented as: or .

[0040] The current emotional space coordinates, waveform dynamics features, emotional acceleration, high-dimensional orthogonal feature summary, identity identifier, and trusted timestamp are concatenated into a fused continuous-valued feature vector, represented as: .

[0041] Normalizing continuous value characteristics to eliminate the influence of dimensions is expressed as: ; in, , This represents the pre-calculated global mean and standard deviation.

[0042] Step 5: Establish a mechanism for adapting and recognizing mental states between different mental subjects to align mental states in interactive scenarios; A standardized emotional waveform protocol is constructed to enable similarity measurement of the emotional spaces of different mental subjects. The process involves: establishing emotional coordinates for each mental subject, converting these coordinates into standardized scores, and representing the emotional coordinates as follows: ; ; in, This represents the emotional space coordinate vector of the mental subject X at time point t, and is a d-dimensional vector, for example, in VAD space... ; This refers to a period of stable mental state, such as the resting period before interaction.

[0043] The formula for converting to standardized scores is: ; As a constant, the emotional states of all mental agents have a statistical mean of 0 and a variance of 1. As a mean vector, it can be compared in the same coordinate system.

[0044] When two mental agents interact, such as mental agent A and mental agent B, their respective current emotional waveform segments are extracted, as represented as follows: and The algorithm calculates waveform similarity, synchronization, and response delay patterns, and uses Dynamic Time Warping (DTW) or cross-correlation analysis to find the optimal alignment between two mental agents. For example, the cross-correlation function is calculated using the inner product and is expressed as: ; in, The start time of the interaction. This refers to the current time (during real-time calculation) or the time when the interaction ends (during offline analysis). Let A be the standardized emotional vector of the mental subject A at time t (already mapped to the public space). For mental subject B in time Standardized sentiment vectors.

[0045] The optimal time offset is expressed as: .

[0046] The formula for calculating waveform similarity is: The aligned Euclidean distance is: ; The formula for calculating cosine similarity is: .

[0047] The cross-correlation of velocity sequences is used to evaluate the synchronicity between emotional velocity and acceleration, expressed as: ; The synchronization index is expressed as: .

[0048] Based on the response time distribution extracted from interactive events, such as B's reaction after A speaks, the response delay pattern is represented as: ; This refers to the interaction initiation time, specifically the time A initiated the interaction. This refers to the interaction response time, i.e., the reaction time of B.

[0049] First, the direction of emotional state guidance is determined by Granger causality test or transit entropy. Based on waveform similarity, speed synchronization, and guidance direction, an adaptation coefficient is output to describe the degree of coordination of mental states among mental subjects, which is used to construct a collaborative memory space for multiple mental subjects.

[0050] The guiding direction is represented as: ; like If A's emotional state has a stronger predictive power for B, then A guides B.

[0051] The fitness coefficient is expressed as: ; Among them, weight It can be dynamically adjusted according to the application scenario (such as cooperation, competition, empathy).

[0052] When the fit coefficient falls below a threshold or requires proactive optimization of the interaction, a mental alignment mechanism can be triggered. This mechanism adjusts the waveform through external intervention or adaptive adjustment by the mental agents to bring the mental states of both parties closer together. Based on the interaction goal, such as making B follow A's emotions, the desired emotional trajectory of B is generated as follows: ; in, This is a small deviation allowance used to preserve individual differences.

[0053] The target waveform is converted into behavioral instructions for the mental subject, such as adjusting the tone, rhythm, and emotional inclination of the dialogue, and gradually approached through feedback control, as shown in the following: ; in, The proportional gain matrix is ​​a scalar. Let be the target emotional state vector of mental subject B at time t; The current standardized emotional state vector of mental agent B at time t; The standardized emotional waveform of mental subject A serves as the target to be followed; It is the differential gain matrix; This is the proportional error term; This represents the rate of change of the error term over time.

[0054] Step 6 involves incrementally updating the dynamic emotional waveform of the mental subject in real time to maintain the high timeliness of the mental state. Specifically, a streaming processing architecture is used to continuously receive multimodal input, including the complete emotional waveform data from the previous moment, newly arrived multimodal feature data, and the current reliable timestamp. Preprocessing is performed on the data, including time alignment, missing value handling, and outlier filtering. New input data undergoes high-dimensional orthogonal feature processing, and the emotional space coordinates are updated. The high-dimensional orthogonal feature processing follows the process in Step 2 to generate the high-dimensional orthogonal feature vector for the current moment. .

[0055] The difference between the high-dimensional orthogonal feature vector at the current time and the feature vector at the previous time is calculated to detect significant changes. The formula for calculating the difference is: ; If the difference value If the change is less than a preset threshold, the "feature stable" flag is triggered to reduce the amount of subsequent computation.

[0056] Based on the process in step 3, Mapping to the emotional space coordinates, if the cross-mind subject adaptation recognition in step 5 requires real-time adaptation, then a standardized version is also calculated simultaneously, using the following formula: ; in, and These are the currently updated baseline parameters.

[0057] The dynamic characteristics are updated incrementally, preserving historical waveform shapes when updating the waveform. Only the latest time point is appended or recent waveforms are smoothed. Waveform smoothing uses the existing exponentially weighted moving average (EWMA) method to ensure waveform data continuity and real-time performance, supporting millisecond-level response and status queries. The incremental update process of the dynamic characteristics is as follows: using the previous time point... and the current moment Calculation speed, expressed as: ; The acceleration is calculated using the velocity from the previous moment: ; The updated waveform curvature is represented as: .

[0058] The newly calculated sentiment coordinates and dynamic features are appended to the waveform time series for waveform updates, as shown below: .

[0059] Step 7: The dynamic emotional waveform and its key derived data are time-series solidified in a verifiable and tamper-proof manner. Specifically: Collect all waveform data points from the previous evidence storage time to the current time to form a time window segment. The time window segment includes an emotion coordinate sequence, velocity sequence, acceleration sequence, waveform curvature sequence, memory credential sequence, and event marker. For each time window, a hash calculation is performed on the waveform segment, memory credential, and high-dimensional feature summary. The time window is defined as a significant state change per minute, hour, or each instance. The calculation process is as follows: .

[0060] Generate a time-series evidence unit, including waveform data hash, timestamp, and the hash value of the previous evidence unit. The timestamp uses a trusted time source, such as the National Time Service Center or blockchain network time. The hash value of the previous evidence unit forms a chain structure. Priority is given to using blockchain network time, such as Ethereum block timestamps, or using NTP-synchronized time from the National Time Service Center with an appended time signature as the obtained trusted timestamp. Obtain the chain hash of the previous evidence unit. If this is the first time the certificate has been stored, a predefined genesis hash will be used. The temporal evidence storage unit u is represented as: .

[0061] Metadata refers to metadata, including the identity identifier of the mental entity. The system includes fragment start and end timestamps, event marker list, waveform statistics (mean, variance, extreme values, etc.), and calculates the hash of the evidence storage unit. for: .

[0062] The time-series evidence storage units are submitted to the blockchain network for evidence storage. The blockchain network is selected, such as Ethereum, Hyperledger Fabric, or consortium blockchain. The evidence storage frequency and gas cost are determined according to system requirements to ensure data integrity, immutability of sequence, and long-term verifiability.

[0063] Step 8: Based on the solidified and stored dynamic emotional waveform sequence, construct a dedicated long-term mental memory space for each mental subject, specifically as follows: The time-series evidence storage units after blockchain notarization serve as the basic storage layer of the memory space; a multi-dimensional index structure is constructed, which can be divided into a hot data layer, a warm data layer, a cold data layer, and an evidence storage index layer according to the length of time the data has been stored. The evidence storage index layer only stores the hash Hu of each time-series evidence storage unit and the corresponding blockchain transaction certificate; it supports efficient querying and backtracking by dimensions such as time axis, emotional state interval, memory certificate, interaction object, and event tag; and full-text / tag indexes can be built using Elasticsearch or Apache Solr. The memory space provides standardized interfaces to support: Precise query: Extract corresponding waveforms and mental features based on time points or time intervals; Retrospective analysis: reconstructing historical mental trajectories, analyzing emotional evolution patterns, key turning points, and the impact of external events; Cross-subject comparison: Under the premise of authorization, compare the synchronicity of emotional waveforms of different subjects within the same time interval; like Figure 2 As shown, the memory space itself uses a decentralized storage protocol (such as IPFS) combined with a blockchain index to ensure that the data is not destroyed: even if some nodes fail, the complete mental history can still be recovered through the evidence records on the blockchain.

[0064] This embodiment eliminates redundant information between different modalities and within the same modality by collecting multimodal data and using a high-dimensional orthogonal processing method. This allows each feature component to independently express different dimensions of the mind, improving the accuracy and interpretability of subsequent emotion mapping. Dimensionality reduction retains key mental information while reducing the computational overhead of subsequent waveform generation and storage. Waveform generation can transform discrete mental feature points into continuous differentiable emotion waveforms, accurately generating the evolution process of mental states. This makes the emotional states of different mental subjects comparable in the same coordinate system, providing a foundation for cross-mental subject adaptation. The generation of unique identifiers prevents the problems of easy conflict and forgery of "timestamp + identity" combination identifiers. Only the hash value is put on the chain, and the data body is stored in decentralized storage such as IPFS, balancing security and economy. The cost of single-time evidence storage can be as low as a few cents, while improving data traceability. Example 2

[0065] An intelligent agent emotion waveform generation and unique identification system includes a multi-dimensional feature acquisition interface, a memory, and a processor. The multi-dimensional feature acquisition interface is used to receive multimodal data, the memory stores computer-executable instructions, and the processor is used to execute the executable instructions stored in the memory. When the executable instructions are executed by the processor, they are executed according to the steps of the method in Embodiment 1.

[0066] The above descriptions are merely embodiments of the present invention, and common knowledge regarding specific structures and characteristics is not elaborated upon here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the structure of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for generating and uniquely identifying the emotional waveform of an intelligent agent, comprising the following steps: Step 1: Collect multimodal data of the target mental subject with original timestamps and source identifiers, and use it as input source after preprocessing; Step 2: Map the collected multimodal data to a high-dimensional feature space, and extract high-dimensional orthogonal features of sentiment through orthogonalization processing to obtain high-dimensional feature vectors; Its characteristic is that it further includes: Step 3: Map the high-dimensional orthogonal feature vectors to the emotional waveform that evolves over time, and then concatenate the emotional waveform, the current system timestamp, and the previous hash to form a data block to be stored. Step 4: Perform a hash operation on the data block to be stored to generate a memory certificate; and construct a storage unit containing the emotional waveform summary, the memory certificate, the current system timestamp and the hash value of the previous block, and write the storage unit into the blockchain or distributed ledger. Step 5: Establish a mechanism for adapting and recognizing mental states between different mental subjects to align mental states in interactive scenarios; Step 6: Based on the data from Step 1, incrementally update the dynamic emotional waveform of the mental subject in real time.

2. The method for generating and uniquely identifying intelligent agent emotion waveforms according to claim 1, characterized in that: In step 3, an emotional space is constructed by converting the orthogonal feature vectors at each time point into coordinate points in the emotional space through regression or neural mapping networks. The emotional coordinate sequence at continuous time points is connected to form a dynamic emotional waveform. The dynamic emotional waveform includes instantaneous emotional states and includes the rate of change between states, waveform curvature, and dynamic characteristics of oscillation modes.

3. The method for generating and uniquely identifying intelligent agent emotion waveforms according to claim 2, characterized in that: In step 5, a standardized emotional waveform protocol is constructed. When two mental agents interact, their respective current emotional waveform segments are extracted, waveform similarity, synchronization and response delay patterns are calculated, and an adaptation coefficient is output to describe the degree of coordination between the mental states of the mental agents.

4. The method for generating and uniquely identifying intelligent agent emotion waveforms according to claim 3, characterized in that: In step 6, a streaming processing architecture is adopted to continuously receive multimodal inputs, perform high-dimensional orthogonal feature processing on the new input data, and update the emotional space coordinates; when updating the waveform, the historical waveform shape is retained, and only the latest time point is appended or the recent waveform is smoothed and corrected.

5. The method for generating and uniquely identifying intelligent agent emotion waveforms according to claim 3, characterized in that: Also includes: Step 7: The dynamic emotional waveform and its key derived data are time-series solidified in a verifiable and tamper-proof manner.

6. The method for generating and uniquely identifying intelligent agent emotion waveforms according to claim 5, characterized in that: In step 7, hash calculations are performed on the waveform segments, memory vouchers, and high-dimensional feature summaries for each time window; Generate a time-series evidence unit, including waveform data hash, timestamp, and hash value of the previous evidence unit; The time-series evidence unit is submitted to the blockchain network for evidence storage.

7. The method for generating and uniquely identifying intelligent agent emotion waveforms according to claim 5, characterized in that: Also includes: Step 8: Based on the solidified and stored dynamic emotional waveform sequence, construct a dedicated long-term mental memory space for each mental subject.

8. The method for generating and uniquely identifying intelligent agent emotion waveforms according to claim 7, characterized in that: In step 8, the time-series evidence storage unit after blockchain evidence storage is used as the basic storage layer of the memory space. Construct a multi-dimensional index structure to support efficient querying and backtracking by time axis, emotional state interval, memory credential, interaction object, event tag and other dimensions; The memory space provides a standardized interface to the outside world.

9. An intelligent agent emotion waveform generation and unique identification system, comprising a multi-dimensional feature acquisition interface, a memory, and a processor, wherein the multi-dimensional feature acquisition interface is used to receive multimodal data, the memory stores computer-executable instructions, and the processor is used to execute the executable instructions stored in the memory; Its features are: When the executable instructions are executed by the processor, they are performed in accordance with the steps of the method according to any one of claims 1-9.

Images