Multi-modal data storage control device and method of intelligent agent

Abstract

The invention discloses a multi-modal data storage control device and method for an intelligent agent, and relates to the technical field of data management, and the method comprises the steps: collecting multi-modal original sensing data, and analyzing the multi-modal original sensing data to obtain a multi-modal working state and a weight table; performing multi-mode event condition triggering judgment on the multi-mode working state, and analyzing the change trend of the multi-mode working state to obtain a storage control event; characteristic extraction and multi-modal mode matching are carried out on the storage control event to obtain an event type identification result, entropy processing is carried out on the event type identification result, an optimal storage path is efficiently selected through a path optimization method such as a Dijkstra algorithm, the storage cost and delay are reduced, and finally an accurate storage state report is generated. And an efficient and safe data storage scheme is provided for agent storage and task management.

Description

Technical Field

[0001] The present invention relates to the technical field of data management, and in particular to a multimodal data storage control device and method for an intelligent body. Background Art

[0002] As a cutting-edge research direction in the field of automation and intelligent control, multi-agent systems are widely used in complex task scenarios such as robot cluster collaboration, unmanned driving formations, intelligent inspections, and distributed perception. Collaborative control is achieved through local information interaction among each agent. Typical control objectives include consistency control, formation control, and path planning. In order to improve the dynamic response capability and robustness of the system, a large number of research results have been carried out around methods such as adaptive control, sliding mode control, robust control, and finite time control, and significant progress has been made at the theoretical level. In response to the modeling errors and external disturbances in multi-agent systems, disturbance estimation and compensation mechanisms have also been widely used. Among them, the extended state observer plays an important role in control systems because it can estimate the total disturbance without an accurate model.

[0003] Although existing control strategies have achieved good results in theory, they still have certain limitations in practical engineering applications. Most ESO-based control methods only achieve asymptotic stability of system errors and are unable to meet the requirements for fast response performance. Existing methods often fail to fully integrate safety constraints such as communication maintenance and collision avoidance during the design process, resulting in difficulty in ensuring the safety of the system in high-density or dynamic environments. Summary of the Invention

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

[0005] Therefore, the present invention provides a multimodal data storage control method for an intelligent body to solve the problems of incomplete state reflection and low data management efficiency caused by single modal processing.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a multimodal data storage control method for an intelligent agent, which includes collecting multimodal raw perception data, analyzing the multimodal raw perception data, and obtaining a multimodal working state and a weight table; By performing multimodal event condition triggering judgment on the multimodal working state, analyzing the changing trend of the multimodal working state, and obtaining the storage control event; By performing feature extraction and multimodal pattern matching on storage control events, event type recognition results are obtained, and entropy value processing is performed on the event type recognition results to generate priority labels; Combine the event type recognition results with the weight table, dynamically assign priorities, and obtain a storage path pairing table; A storage completion status report is obtained by using a path optimization method based on the storage-path pairing table.

[0007] As a preferred solution of the multimodal data storage control method of the intelligent agent of the present invention, wherein: collecting multimodal original perception data, analyzing the multimodal original perception data, and obtaining the multimodal working state and weight table include the following steps: The agent collects multimodal raw perception data through its built-in sensors, extracts multimodal feature information, analyzes the multimodal raw perception data, identifies task status and perception quality, and forms a multimodal working state. A weighted evaluation algorithm is used to evaluate the multimodal working status and generate a corresponding weight table.

[0008] As a preferred solution of the multimodal data storage control method of the intelligent body of the present invention, wherein: by performing multimodal event condition triggering judgment on the multimodal working state, analyzing the change trend of the multimodal working state, and obtaining the storage control event, the method includes the following steps: The multimodal features in the multimodal working state are sampled at fixed time intervals to form a time series of feature changes corresponding to the multimodal state, and the first-order difference method is used to calculate the rate of change of the multimodal features. Setting a change rate trigger threshold based on the change rate of multimodal features; When the rate of change of the multimodal feature is greater than the change rate trigger threshold, a multimodal event is obtained; Based on the sliding window cumulative analysis method, the multimodal working status is analyzed to determine the direction and magnitude of change and calculate the event score; Generate storage trigger thresholds through statistical analysis of event scores; When the event score exceeds the storage trigger threshold, a storage control event is generated.

[0009] As a preferred solution of the multimodal data storage control method of the intelligent agent of the present invention, wherein: by extracting features of storage control events and matching multimodal patterns, an event type recognition result is obtained, including the following steps: Extracting storage control event feature information from the storage control event using a feature vector encoding method; Based on the characteristic information of storage control events, the support vector machine classification method is used to identify multimodal patterns; The storage control event feature information and multimodal patterns are matched using the cosine similarity method to obtain the event type recognition result.

[0010] As a preferred solution of the multimodal data storage control method of the intelligent agent of the present invention, wherein: entropy processing is performed on the event type recognition result to generate a priority label, including the following steps: Use the histogram statistical analysis method to classify and count the event type identification results, and calculate the probability of each type of event occurring; Calculate the entropy value based on the probability of occurrence of each type of event; The entropy value threshold interval is set based on the entropy value to divide the priority labels.

[0011] As a preferred solution of the multimodal data storage control method of the intelligent agent of the present invention, wherein: combining the event type recognition result and the weight table, dynamically assigning priorities, and obtaining a storage path pairing table, the following steps are included: The weighted linear fusion method is used to combine the event type recognition results and the weight table to calculate the priority score of multimodal events; According to the priority scores of multimodal events, multimodal events are divided into high priority, medium priority and low priority; Based on the local storage capacity of the agent and the priority of the multimodal event, a storage path pair table is allocated for each multimodal event according to the storage path allocation strategy.

[0012] As a preferred solution of the multimodal data storage control method of the intelligent agent of the present invention, wherein: using a path optimization method based on the storage path pairing table to obtain a storage status report, the following steps are included: Building a weighted graph based on agent local storage nodes and edge nodes; The Dijkstra algorithm is used to search for the optimal storage path on the weighted graph, and the storage path is used to generate a storage status report.

[0013] In a second aspect, the present invention provides a multimodal data storage control device for an intelligent agent, comprising: a data acquisition module for collecting multimodal raw perception data, analyzing the multimodal raw perception data, and obtaining a multimodal working state and a weight table; The control event module determines the triggering conditions of multimodal event by analyzing the changing trend of multimodal working state and obtains storage control events; The priority module extracts features from storage control events and performs multimodal pattern matching to obtain event type recognition results, performs entropy processing on the event type recognition results, and generates priority labels; The pairing table module combines the event type recognition results with the weight table, dynamically assigns priorities, and obtains the storage path pairing table; The status reporting module uses the path optimization method based on the storage path pairing table to obtain the storage status report In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the multimodal data storage control method of an intelligent agent as described in the first aspect of the present invention.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the multimodal data storage control method of an intelligent agent as described in the first aspect of the present invention.

[0015] The beneficial effects of the present invention are as follows: by collecting multimodal perception data and analyzing and generating multimodal working status and weight tables, a comprehensive evaluation of different task states is achieved, and event type recognition results and entropy value processing are combined to dynamically assign priorities. Through path optimization methods such as the Dijkstra algorithm, the optimal storage path is efficiently selected, storage costs and delays are reduced, and an accurate storage status report is finally generated, providing an efficient and secure data storage solution for the storage and task management of intelligent agents. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 Flowchart of a multimodal data storage control method for an intelligent agent.

[0018] Figure 2 Schematic diagram of the multimodal data storage control device of the energy body.

[0019] Figure 3 Flowchart for allocating storage path pair tables for multimodal events.

[0020] Figure 4 Flowchart of the results identified for event types. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0022] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0024] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a multimodal data storage control method for an intelligent agent, comprising the following steps: S1. Collect multimodal original perception data, analyze the multimodal original perception data, and obtain a multimodal working state and weight table.

[0025] S1.1. Collect multimodal raw perception data through the built-in sensors of the intelligent agent, extract multimodal feature information, analyze the multimodal raw perception data, identify the task status and perception quality, and form a multimodal working state.

[0026] Furthermore, the intelligent body's built-in sensors (such as vision, hearing, temperature and acceleration sensors) collect multimodal raw perception data in real time, including environmental parameters, object position, motion status and sound signals. Signal processing and feature extraction algorithms are used to preprocess data of different modalities and extract various types of feature information, such as image edge features, audio spectrum features, and temperature change trends. Multimodal features are fused through multimodal fusion technology to finally form a multimodal working state that reflects the current operating state of the intelligent body.

[0027] S1.2. Use a weighted evaluation algorithm to evaluate the multimodal working status and generate a corresponding weight table.

[0028] Furthermore, a weighted evaluation algorithm is used to combine the multimodal working state, calculate the weighted score, evaluate the quality of the multimodal working state, and generate a weight table of the multimodal working state weights. S2. Perform multimodal event condition triggering judgment on the multimodal working state, analyze the change trend of the multimodal working state, and obtain a storage control event.

[0029] S2.1. Sample the multimodal features in the multimodal working state at fixed time intervals to form a feature change time series corresponding to the multimodal state, and use the first-order difference method to calculate the multimodal feature change rate.

[0030] Furthermore, by sampling the multimodal features in the multimodal working state at fixed time intervals, it is possible to ensure that the state changes of each mode are accurately captured within a continuous time window. The sampled multimodal features will serve as the basic data for subsequent analysis, reflecting the changes of each modal feature over time. The sampled data is processed using the first-order difference method, which calculates the change rate by calculating the difference in the characteristic values ​​between two consecutive time points. For the multimodal features of each mode, the multimodal feature change rate is obtained by calculating the difference between each pair of adjacent time points and dividing it by the time interval.

[0031] S2.2. Set a change rate trigger threshold based on the multimodal feature change rate.

[0032] Furthermore, by statistically analyzing the distribution of feature change rates, a change rate trigger threshold is set. The change rate trigger threshold represents that within a specific time window, when the feature change rate exceeds the change rate trigger threshold, it can be considered that a significant state change has occurred, thereby triggering the corresponding multimodal event. The change rate trigger threshold can be dynamically adjusted based on multiple factors to set the change rate trigger threshold.

[0033] S2.3. When the rate of change of the multimodal feature is greater than the change rate trigger threshold, a multimodal event is obtained.

[0034] Furthermore, when the rate of change of the multimodal feature exceeds the change rate trigger threshold, it is determined that a multimodal event has occurred. The occurrence of this multimodal event indicates that the change in the multimodal working state has reached sufficient significance and has a significant impact on the behavior or task objectives of the intelligent agent. The event will be recorded immediately for further processing and analysis. The multimodal event can trigger operations such as storage control, priority adjustment or path optimization. Through this event triggering mechanism based on the feature change rate, it is possible to sensitively respond to dynamic changes in the environment and tasks and obtain multimodal events.

[0035] S2.4, Based on the sliding window cumulative analysis method, analyze the multimodal working status, determine the change direction and amplitude, and calculate the event score. Specifically, the expression is: ; in, For the moment The event score, is the multimodal weight, is the multimodal trigger frequency regularization value, is the frequency adjustment coefficient, For the moment The range of change below.

[0036] S2.5. Generate a storage trigger threshold through statistical analysis of event scores.

[0037] Furthermore, through statistical analysis of event scores, the central tendency (such as mean, median) and dispersion (such as standard deviation) of event scores are identified, and an event score threshold is set. Events above the event score threshold and greater than the event score are considered high-value events, and a storage trigger threshold is automatically generated. Event scores above the event score threshold and equal to the event score are considered high-value events and can be selectively cached or compressed for storage; event scores above the event score threshold and less than the event score are considered high-value events and are filtered out or only temporarily processed.

[0038] S2.6. If the event score exceeds the storage trigger threshold, a storage control event is generated.

[0039] Furthermore, when the event score exceeds the preset storage trigger threshold, the event will be automatically determined to be a high-priority event and storage control will be required. The trigger threshold is set based on the statistical characteristics of historical data and can be dynamically adjusted according to environmental changes. Once the event score exceeds the storage trigger threshold, a corresponding storage control event will be generated.

[0040] S3. Obtain event type recognition results by performing feature extraction and multimodal pattern matching on storage control events.

[0041] S3.1. Use a feature vector encoding method to extract storage control event feature information from the storage control event.

[0042] Furthermore, various types of storage control event feature information (such as timestamps, change amplitudes, frequencies, etc.) in storage control events are converted into standardized numerical forms, and different types of storage control event feature information are mapped into a high-dimensional space according to specific rules to ensure that each event can be represented by a set of numerical vectors. Feature vector encoding not only retains the temporal characteristics of events, but also captures the relationships between events. Through the analysis of feature vectors, it can better provide accurate input data for subsequent pattern recognition and classification. The feature vector encoding method provides a basis for subsequent multimodal pattern matching, so that the features of storage control events can be effectively compared and identified with other patterns to obtain storage control event feature information.

[0043] S3.2. Based on the storage control event feature information, a support vector machine classification method is used to identify multimodal patterns.

[0044] Furthermore, according to the classification principle of support vector machine, the storage control event feature information is divided into corresponding multimodal pattern categories. By maximizing the event feature information between different categories, the accuracy of event type classification is improved. At the same time, the kernel function is used to map the input event features to a high-dimensional space in order to process the nonlinear relationship between features. Each storage control event can be accurately attributed to a specific multimodal pattern based on its characteristics, and the multimodal pattern can be identified.

[0045] S3.3. Match the storage control event feature information and the multimodal pattern using the cosine similarity method to obtain the event type recognition result.

[0046] Furthermore, the cosine similarity method is used to match the storage control event feature information with the multimodal pattern. By calculating the cosine value of the angle between the feature information of the storage control event, the similarity between the feature information of the storage control event and the multimodal pattern is measured. The more consistent the direction of the feature information of the storage control event and the multimodal pattern in the feature space, the higher the similarity. According to the multimodal pattern corresponding to the highest cosine similarity, the current storage control event is classified into the corresponding event type, thereby accurately obtaining the event type recognition result. S4. Perform entropy processing on the event type recognition results to generate priority labels.

[0047] S4.1. Use the histogram statistical analysis method to classify and count the event type identification results, and calculate the probability of each type of event. Specifically, the expression is: ; in, For the The probability of the current event type occurring, is the total number of event types, For the The frequency of occurrence of historical event types, For the The frequency of occurrence of the current event type, is the index of the current event type, The index of the historical event type.

[0048] S4.2. Calculate the entropy value based on the probability of occurrence of each type of event. The expression is: ; in, For the The entropy value corresponding to the current event type.

[0049] Furthermore, after completing the statistics of the probability of occurrence of each event type, the corresponding entropy value is calculated based on the probability of occurrence of each type of event, which is used to measure the uncertainty of the event type. When the probability of an event is low, the entropy value is high, reflecting a strong randomness, while the entropy value of an event with a high probability of occurrence is low, indicating a high certainty of occurrence, through the calculation of the entropy value.

[0050] S4.3. Prioritize labels by setting entropy threshold intervals based on entropy values.

[0051] Furthermore, event types are divided into different priority labels. Events with lower entropy values ​​(i.e., events are more certain to occur and are of higher importance) are classified as high priority, events with entropy values ​​in the middle range are classified as medium priority, and events with higher entropy values ​​(i.e., events are more random and less urgent) are classified as low priority. By dividing the entropy value intervals, we can achieve a refined classification of events of different importance.

[0052] S5. Combine the event type recognition result and the weight table, dynamically assign priorities, and obtain a storage path pairing table.

[0053] S5.1. Use the weighted linear fusion method to combine the event type recognition results and the weight table to calculate the priority score of the multimodal event.

[0054] Furthermore, after obtaining the event type recognition results and the weight table, the weighted linear fusion method is used to combine the event type recognition results and the weight table. For each multimodal event, linear weighting is performed according to the corresponding event type recognition score and the weight assigned in the weight table, and the set fusion coefficient is used to calculate the comprehensive priority score.

[0055] S5.2. Classify multimodal events into high priority, medium priority, and low priority based on their priority scores.

[0056] Furthermore, based on the priority scores of multimodal events, multimodal events are graded and processed into high-priority events, medium-priority events, and low-priority events. High-priority events usually have higher timeliness requirements, medium-priority events may involve routine but important interactive tasks, and low-priority events are usually background or time-insensitive information. Through the grading mechanism, computing and communication resources can be dynamically allocated to ensure that high-priority event tasks are processed first.

[0057] S5.3. Based on the local storage capacity of the agent and the priority of the multimodal event, a storage path pair table is allocated for each multimodal event according to the storage path allocation strategy.

[0058] Furthermore, the storage priority is determined according to each priority label. High-priority events are allocated to the local high-speed storage area first, medium-priority events select the ordinary storage area based on the remaining capacity, and low-priority events are transferred to the edge node storage. When the local capacity is insufficient, some low-priority events are migrated to external nodes to ensure that critical data is stored efficiently first, thereby improving the flexibility of data management and the utilization of storage resources.

[0059] S6. Using a path optimization method based on the storage-path pairing table, obtain a storage status report.

[0060] S6.1. Establish a weighted graph based on the agent's local storage nodes and edge nodes.

[0061] Furthermore, by utilizing the local storage nodes and edge nodes of the intelligent body, the connection between the local storage nodes and the edge nodes represents the interaction path of the data, and the interaction path of each edge data reflects the performance indicators such as the cost, delay or capacity of data transmission between different nodes, and a weighted graph can be established.

[0062] S6.2. Use the Dijkstra algorithm to search for the optimal storage path on the weighted graph, and generate a storage status report based on the storage path.

[0063] Furthermore, after the weighted graph is established, the Dijkstra algorithm is used to search for the shortest path. The Dijkstra algorithm calculates the shortest path from the source node (such as the local storage node) to each target node (such as the edge node or the remote cloud) and selects the optimal storage path based on the weight of each edge in the weighted graph (representing factors such as storage cost and transmission delay). The Dijkstra algorithm can ensure that data is stored at the lowest cost and can give priority to low-latency, high-bandwidth paths for storage. After the path search is completed, a storage status report is generated.

[0064] This embodiment also provides a multimodal data storage control device for an intelligent agent, comprising: a data acquisition module for collecting multimodal raw perception data, analyzing the multimodal raw perception data, and obtaining a multimodal working state and a weight table; The control event module determines the triggering conditions of multimodal event by analyzing the changing trend of multimodal working state and obtains storage control events; The priority module extracts features from storage control events and performs multimodal pattern matching to obtain event type recognition results, performs entropy processing on the event type recognition results, and generates priority labels; The pairing table module combines the event type recognition results with the weight table, dynamically assigns priorities, and obtains the storage path pairing table; The status reporting module obtains a storage status report using a path optimization method based on the storage-path pairing table.

[0065] This embodiment also provides a computer device suitable for the multimodal data storage control method of an intelligent body, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the multimodal data storage control method of an intelligent body proposed in the above embodiment.

[0066] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0067] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the multimodal data storage control method for implementing an intelligent agent as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0068] In summary, the present invention realizes a comprehensive evaluation of different task states by collecting multimodal perception data and analyzing and generating multimodal working states and weight tables. It combines event type recognition results and entropy processing, dynamically assigns priorities, and efficiently selects the optimal storage path through path optimization methods such as the Dijkstra algorithm, thereby reducing storage costs and delays, and ultimately generating accurate storage status reports, providing an efficient and secure data storage solution for the storage and task management of intelligent agents.

[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A multimodal data storage control method for an intelligent agent, characterized by: include, Collecting multimodal original perception data, analyzing the multimodal original perception data, and obtaining a multimodal working state and weight table; By performing multimodal event condition triggering judgment on the multimodal working state, analyzing the changing trend of the multimodal working state, and obtaining the storage control event; By performing feature extraction and multimodal pattern matching on storage control events, event type recognition results are obtained, and entropy value processing is performed on the event type recognition results to generate priority labels; Combine the event type recognition results with the weight table, dynamically assign priorities, and obtain a storage path pairing table; A storage completion status report is obtained by using a path optimization method based on the storage-path pairing table.

2. The multimodal data storage control method of an intelligent agent according to claim 1, wherein: Collecting multimodal original perception data, analyzing the multimodal original perception data, and obtaining the multimodal working state and weight table include the following steps: The agent collects multimodal raw perception data through its built-in sensors, extracts multimodal feature information, analyzes the multimodal raw perception data, identifies task status and perception quality, and forms a multimodal working state. A weighted evaluation algorithm is used to evaluate the multimodal working status and generate a corresponding weight table.

3. The multimodal data storage control method of an intelligent agent according to claim 2, wherein: By performing multimodal event condition triggering judgment on the multimodal working state, analyzing the change trend of the multimodal working state, and obtaining the storage control event, the following steps are included: The multimodal features in the multimodal working state are sampled at fixed time intervals to form a time series of feature changes corresponding to the multimodal state, and the first-order difference method is used to calculate the rate of change of the multimodal features. Setting a change rate trigger threshold based on the change rate of multimodal features; When the rate of change of the multimodal feature is greater than the change rate trigger threshold, a multimodal event is obtained; Based on the sliding window cumulative analysis method, the multimodal working status is analyzed to determine the direction and magnitude of change and calculate the event score; Generate storage trigger thresholds through statistical analysis of event scores; When the event score exceeds the storage trigger threshold, a storage control event is generated.

4. The multimodal data storage control method of an intelligent agent according to claim 3, wherein: By performing feature extraction and multimodal pattern matching on storage control events, event type recognition results are obtained, including the following steps: Extracting storage control event feature information from the storage control event using a feature vector encoding method; Based on the characteristic information of storage control events, the support vector machine classification method is used to identify multimodal patterns; The storage control event feature information and multimodal patterns are matched using the cosine similarity method to obtain the event type recognition result.

5. The multimodal data storage control method of an intelligent agent according to claim 4, wherein: The entropy value of the event type recognition result is processed to generate a priority label, which includes the following steps: Use the histogram statistical analysis method to classify and count the event type identification results, and calculate the probability of each type of event occurring; Calculate the entropy value based on the probability of occurrence of each type of event; The entropy value threshold interval is set based on the entropy value to divide the priority labels.

6. The multimodal data storage control method of an intelligent agent according to claim 5, wherein: Combine the event type identification results with the weight table, dynamically assign priorities, and obtain the storage path pairing table, including the following steps: The weighted linear fusion method is used to combine the event type recognition results and the weight table to calculate the priority score of multimodal events; According to the priority scores of multimodal events, multimodal events are divided into high priority, medium priority and low priority; Based on the local storage capacity of the agent and the priority of the multimodal event, a storage path pair table is allocated for each multimodal event according to the storage path allocation strategy.

7. The multimodal data storage control method of an intelligent agent according to claim 6, wherein: Based on the storage path pairing table, a path optimization method is used to obtain a storage status report, including the following steps: Building a weighted graph based on agent local storage nodes and edge nodes; The Dijkstra algorithm is used to search for the optimal storage path on the weighted graph, and the storage path is used to generate a storage status report.

8. A multimodal data storage control device for an intelligent agent, based on the multimodal data storage control method for an intelligent agent according to any one of claims 1 to 7, characterized in that: include, The data acquisition module collects and analyzes the multimodal original perception data to obtain the multimodal working status and weight table; The control event module determines the triggering conditions of multimodal event by analyzing the changing trend of multimodal working state and obtains storage control events; The priority module extracts features from storage control events and performs multimodal pattern matching to obtain event type recognition results, performs entropy processing on the event type recognition results, and generates priority labels; The pairing table module combines the event type recognition results with the weight table, dynamically assigns priorities, and obtains the storage path pairing table; The status reporting module obtains a storage status report using a path optimization method based on the storage-path pairing table.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the multimodal data storage control method for an intelligent agent according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the multimodal data storage control method for an intelligent agent according to any one of claims 1 to 7 are implemented.

Images