Environment sensing system based on multi-agent interaction

Abstract

The invention discloses an environment perception system based on multi-agent interaction, and the system comprises an environment perception data obtaining unit which is used for obtaining multi-agent environment perception data; the data feature vector construction unit is used for acquiring environmental perception standard data, performing noise elimination processing on the perception standard data, constructing a data feature vector and acquiring standard associated data from the environmental perception standard data; the perception analysis model construction unit is used for clustering the data feature vectors according to the standard associated data to obtain a to-be-trained data set, and performing multi-agent interaction training on the initial model by using the to-be-trained data set to obtain a multi-agent perception analysis model; and the perception analysis unit is used for carrying out hierarchical processing on the environment perception data to obtain perception feedback data, and inputting the perception feedback data into the multi-agent perception analysis model to obtain a dynamic perception result. And the survivability of the intelligent agent in a complex environment is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an environment perception system based on multi-agent interaction. Background Art

[0002] Current environmental perception technologies mainly rely on a single or limited number of sensor devices to acquire and process physical parameters in the environment. These systems usually use fixed sensing points, such as cameras, temperature sensors, humidity sensors, etc., to collect data and analyze it through preset algorithms. However, this traditional approach has significant limitations when facing complex and dynamic environments.

[0003] Because traditional perception systems lack intelligent processing capabilities, they cannot effectively respond to and adjust to the changing factors in the environment. This defect is particularly evident in complex environments with multiple targets and multiple variables. For example, in disaster monitoring, intelligent traffic management, or complex industrial production environments, environmental conditions often change rapidly, and single or preset perception methods are difficult to cope with diverse scene changes, which can easily lead to wrong decisions or slow responses.

[0004] Existing technologies are insufficient in terms of adaptability, dynamics, and intelligent processing of environmental perception. As an emerging technology that simulates complex social behaviors and adaptive systems, multi-agent systems provide a possibility to solve this problem. However, current multi-agent systems are mainly focused on the fields of simulation and simulation, lacking an organic combination with actual environmental perception. In the existing technology, there is no mature solution that can fully utilize the collaboration and interaction of multiple agents to build a dynamic and efficient environmental perception system that can adapt to complex environments.

[0005] Therefore, there is an urgent need for an environmental perception system based on multi-agent interaction. Summary of the invention

[0006] The present invention provides an environment perception system based on multi-agent interaction to solve the above-mentioned problems existing in the prior art.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] An environment perception system based on multi-agent interaction, comprising:

[0009] An environment perception data acquisition unit is used to acquire multi-agent environment perception data, where the environment perception data includes environment feature parameters and agent perception parameters;

[0010] A data feature vector construction unit is used to obtain environmental perception standard data, perform noise elimination processing on the perception standard data, construct a data feature vector, and obtain standard associated data from the environmental perception standard data;

[0011] Constructing a perception analysis model unit, which is used to cluster data feature vectors according to standard associated data to obtain a data group to be trained, and to perform multi-agent interactive training on the initial model using the data group to be trained to obtain a multi-agent perception analysis model;

[0012] The perception analysis unit is used to perform hierarchical processing on environmental perception data, obtain perception feedback data, input the perception feedback data into the multi-agent perception analysis model, and obtain dynamic perception results.

[0013] Wherein, the environment perception data acquisition unit includes:

[0014] The sensing result acquisition module is used to control multiple agents to simultaneously start sensing operations in a preset environment, output initial environmental sensing data, and determine the sensing results of each agent regarding the current environmental sensing data. The sensing results include data collection accuracy, collection frequency, and corresponding environmental characteristic parameters.

[0015] The environment perception data acquisition module is used to match and analyze the sensing results of each intelligent agent with the preset perception model, determine new environment perception tasks and assign them to the intelligent agent, and generate new environment perception data;

[0016] A feature extraction module is used to extract features from new environmental perception data, obtain new environmental feature parameters and agent perception parameters, and determine multiple new environmental perception tasks until the rate of change of the environmental feature parameters is less than a preset threshold, or the current perception task is determined to be the last task;

[0017] The perception task completion module is used to control multiple intelligent agents to perform perception operations in another preset environment, and repeatedly execute the steps from the initial environment perception data output to the current environment feature parameter change rate being less than the preset threshold until the perception tasks of all preset environments are completed.

[0018] The data feature vector construction unit includes:

[0019] A preprocessing operation module is used to preprocess the acquired environmental perception standard data to eliminate incomplete or abnormal data, input the preprocessed environmental perception standard data into the noise elimination module, perform noise elimination processing on the environmental perception standard data, and obtain denoised data;

[0020] An initial data feature vector construction module is used to extract feature parameters from the denoised data according to preset rules to construct an initial data feature vector;

[0021] Constructing a second data feature vector module, which is used to input the initial data feature vector into the feature screening module, and screening the initial data feature vector according to the set screening logic to obtain an optimized data feature vector;

[0022] Construct a third data feature vector module, which is used to obtain associated parameters from the optimized data feature vector, obtain standard associated data associated with the optimized data feature vector in combination with the set logical relationship, input the standard associated data into the feature construction module, further process the standard associated data, and generate the final data feature vector.

[0023] Among them, obtaining a multi-agent perception analysis model includes:

[0024] Performing clustering processing on the data feature vectors to generate multiple clustering results;

[0025] Determine a data group to be trained based on the clustering result, and output the data group to be trained;

[0026] Input the data set to be trained into the initial model and perform multi-agent interactive training;

[0027] During the multi-agent interactive training process, the perception data of each agent is obtained and analyzed in real time;

[0028] Based on the perception data of each agent and the feedback data in the interactive training, update the parameters of the initial model and output the updated model parameters;

[0029] Based on the updated model parameters, a multi-agent perception analysis model is generated and output;

[0030] Verify the multi-agent perception analysis model and adjust the parameters of the multi-agent perception analysis model based on the verification results;

[0031] In the adjusted multi-agent perception analysis model, multi-agent interaction training is performed again until the preset convergence conditions are reached, and the final multi-agent perception analysis model is output.

[0032] The dynamic perception results are obtained, including:

[0033] Perform hierarchical processing on the perception data to generate multiple sets of perception features at different levels;

[0034] Based on a preset feature extraction algorithm, multiple key features are extracted from the perception feature set, and perception feedback data is generated;

[0035] Input the perception feedback data into the multi-agent perception analysis model, triggering the multi-agent perception analysis model to process the perception feedback data;

[0036] During the processing, the analysis results of each agent are obtained in real time, and multiple sub-level perception results are generated based on the analysis results;

[0037] Integrate the sub-level perception results to generate preliminary dynamic perception results;

[0038] Based on the preset dynamic adjustment rules, the preliminary dynamic perception results are dynamically adjusted to obtain the final dynamic perception results.

[0039] Among them, new environmental perception tasks are determined and assigned to agents, including:

[0040] Determine, based on the sensing result database, a plurality of second sensing results historically generated by the agent;

[0041] extracting a plurality of first perception features of the second sensing result;

[0042] Obtain a preset perception model library, match the first perception feature with the second perception feature in the perception model library, and if the match is consistent, use the matched second sensing result as the third sensing result, and at the same time, obtain the perception task corresponding to the matched second perception feature and associate it with the corresponding intelligent agent;

[0043] Acquire the historical environment item corresponding to the third sensing result, and at the same time, acquire the sensing generation time corresponding to the third sensing result;

[0044] Determine from the sensing result database a plurality of fourth sensing results corresponding to the historical environmental item and generated within a preset time period before and after the generation time;

[0045] Inputting all fourth sensing results into a preset task analysis model, obtaining at least one new perception task output by the task analysis model, and accumulating and calculating the perception tasks associated with the intelligent agent to obtain a task sum;

[0046] If the perception task is the first task and / or the perception task is the fourth task in the second task excluding the third task and / or the total of the tasks is greater than or equal to the preset total threshold of the first task, the agent perception task allocation fails;

[0047] Otherwise, obtaining a preset environment perception task process record library, and determining a plurality of environment perception task process records corresponding to the perception project from the environment perception task process record library;

[0048] Input all environmental perception task process records into a preset importance analysis model to obtain the importance of the importance analysis model output;

[0049] Obtain a second task sum threshold corresponding to the importance;

[0050] If the total of the tasks is greater than or equal to the second task total threshold, the agent perceives that the task assignment has failed;

[0051] Otherwise, the agent perceives that the task assignment is successful.

[0052] The acquired environmental perception standard data is preprocessed, including:

[0053] Determine a preliminary processing result from the environmental perception standard data, and extract multiple data features from the preliminary processing result;

[0054] Determine whether the data features meet the preset integrity standards. If so, execute: further process the preliminary processing results, remove redundant data, and output the processed environmental perception standard data;

[0055] If not, the incomplete or abnormal data features are marked and removed from the environmental perception standard data;

[0056] It is determined again whether the remaining data features in the environmental perception standard data after the elimination meet the preset integrity standard. If so, further processing is performed on the data features and the final processing result is output;

[0057] If not, repeat the process: remove incomplete or abnormal data from the environment perception standard data set until the data features meet the preset integrity standards or the size of the environment perception standard data reaches the preset minimum threshold.

[0058] Among them, multiple clustering results are generated, including:

[0059] Obtaining a data feature vector set, where the data feature vector set includes multiple data feature vectors;

[0060] Performing a preprocessing step on the data feature vector set, the preprocessing step including normalizing the data feature vector;

[0061] According to a preset clustering algorithm, the similarity between each data feature vector and other data feature vectors is determined, and according to a similarity matrix, the data feature vector set is divided into a plurality of preliminary clustering results;

[0062] According to the preset optimization strategy, the preliminary clustering results are optimized, and the optimization includes readjusting the clustering boundaries to improve the overall consistency of the clustering results and generate the final clustering results;

[0063] Determine whether the final clustering result meets the preset clustering evaluation criteria. If so, output the final clustering result.

[0064] Among them, multiple perceptual feature sets of different levels are generated, including:

[0065] Acquire a perception data set, where the perception data set includes a plurality of perception data points;

[0066] Divide the perception data set into several preliminary perception levels according to preset hierarchical processing rules, where the hierarchical processing rules are determined based on the characteristic dimensions, data density and relevance of the perception data;

[0067] Performing hierarchical optimization on the preliminary perception hierarchies, which includes adjusting hierarchical boundaries and merging similar hierarchies to ensure the rationality of hierarchical division and generate multiple optimized perception hierarchies;

[0068] Based on the data features of each perception level, feature sets of the perception level are extracted respectively, and the feature sets are generated by a preset feature extraction algorithm;

[0069] The extracted perceptual feature sets are combined in hierarchical order to generate multiple perceptual feature sets at different levels.

[0070] Wherein, obtaining a set of perception data includes:

[0071] Get request information for sensing data;

[0072] According to the perception data acquisition mechanism, identifying whether the perception data meets the preset conditions, wherein the perception data acquisition mechanism is used to configure the perception data acquisition strategy;

[0073] When the perception data meets the preset conditions, the relevance of the current perception data points is determined based on the data association analysis mechanism;

[0074] When the perception data points are relevant, the perception data points that meet the conditions are aggregated to form a perception data set.

[0075] Compared with the prior art, the present invention has the following advantages:

[0076] An environment perception system based on multi-agent interaction includes: an environment perception data acquisition unit for acquiring multi-agent environment perception data, the environment perception data includes environment characteristic parameters and agent perception parameters; a data feature vector construction unit for acquiring environment perception standard data, performing noise elimination processing on the perception standard data, constructing data feature vectors, and acquiring standard associated data from the environment perception standard data; a perception analysis model construction unit for clustering the data feature vectors according to the standard associated data to obtain a data group to be trained, performing multi-agent interaction training on the initial model with the data group to be trained, and obtaining a multi-agent perception analysis model; a perception analysis unit for hierarchical processing of the environment perception data, acquiring perception feedback data, and inputting the perception feedback data into the multi-agent perception analysis model to obtain dynamic perception results. This not only improves the survivability of the agent in a complex environment, but also improves the safety and operation efficiency of the overall system.

[0077] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the present invention.

[0078] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0080] Figure 1 This is a structural diagram of an environment perception system based on multi-agent interaction in an embodiment of the present invention;

[0081] Figure 2 This is a structural diagram of an environment perception data acquisition unit in an embodiment of the present invention;

[0082] Figure 3 This is a structural diagram of a data feature vector unit in an embodiment of the present invention. DETAILED DESCRIPTION

[0083] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0084] The embodiment of the present invention provides an environment perception system based on multi-agent interaction, including:

[0085] An environment perception data acquisition unit is used to acquire multi-agent environment perception data, where the environment perception data includes environment feature parameters and agent perception parameters;

[0086] A data feature vector construction unit is used to obtain environmental perception standard data, perform noise elimination processing on the perception standard data, construct a data feature vector, and obtain standard associated data from the environmental perception standard data;

[0087] Constructing a perception analysis model unit, which is used to cluster data feature vectors according to standard associated data to obtain a data group to be trained, and to perform multi-agent interactive training on the initial model using the data group to be trained to obtain a multi-agent perception analysis model;

[0088] The perception analysis unit is used to perform hierarchical processing on environmental perception data, obtain perception feedback data, input the perception feedback data into the multi-agent perception analysis model, and obtain dynamic perception results.

[0089] The working principle of the above technical solution is: first, through the environmental perception data acquisition unit, the characteristic parameters of the environment in which the multi-agents are located and the perception parameters of the agents themselves are collected. These data include the physical properties of the environment (such as temperature, humidity, light, etc.) and the state information of the agents (such as position, speed, direction, etc.). This information is the basis for subsequent analysis and processing.

[0090] After data collection, the system standardizes the collected environmental perception data by constructing data feature vector units. This process includes noise elimination to ensure data accuracy and consistency. On this basis, the system generates data feature vectors and extracts standard associated data related to the environment to provide support for subsequent clustering and model training.

[0091] Next, the system uses standard association data to perform cluster analysis on the data feature vectors and extracts the data groups to be trained. These data groups are used to train a multi-agent perception analysis model. Through multiple trainings, the model can learn and simulate the interaction between agents and the interaction between agents and the environment, thereby improving the system's ability to perceive and understand the environment.

[0092] Finally, the system processes the real-time environmental perception data in layers through the perception analysis unit to obtain perception feedback data. Then, the system inputs these feedback data into the trained multi-agent perception analysis model to obtain dynamic perception results of the current environment. These results can be used to guide the further actions of the agent or provide decision support.

[0093] Assume that in an intelligent transportation system, multiple autonomous vehicles are driving on the road as intelligent agents. The system first collects environmental information around the vehicle (such as road conditions, weather conditions, etc.) and the vehicle's driving status (such as speed, direction). After noise processing, these data are used to train a perception analysis model. In the actual operation of the system, when road conditions or weather changes, the system can perceive and analyze these changes in real time, and provide accurate perception results to each vehicle, thereby helping the vehicle make safe driving decisions. For example, when driving in foggy weather, the system can recommend vehicles to reduce speed and maintain a safe distance through real-time perception analysis to prevent traffic accidents.

[0094] The beneficial effects of the above technical solution are: through noise elimination and standardization processing, the system can ensure the accuracy of the data, thereby improving the reliability of perception and analysis results; data feature vectors and cluster analysis enable the system to effectively process large amounts of environmental data and quickly build efficient perception analysis models; the multi-agent interactive training process enables the system to have the ability to adapt to different environmental changes, and maintain a high level of perception accuracy even in complex or dynamic environments.

[0095] In another embodiment, the environment perception data acquisition unit includes:

[0096] The sensing result acquisition module is used to control multiple agents to simultaneously start sensing operations under a preset environment, output initial environmental sensing data, and determine the sensing results of each agent regarding the current environmental sensing data. The sensing results include data collection accuracy, collection frequency, and corresponding environmental characteristic parameters.

[0097] The environment perception data acquisition module is used to match and analyze the sensing results of each intelligent agent with the preset perception model, determine new environment perception tasks and assign them to the intelligent agent, and generate new environment perception data;

[0098] A feature extraction module is used to extract features from new environmental perception data, obtain new environmental feature parameters and agent perception parameters, and determine multiple new environmental perception tasks until the rate of change of the environmental feature parameters is less than a preset threshold, or the current perception task is determined to be the last task;

[0099] The perception task completion module is used to control multiple intelligent agents to perform perception operations in another preset environment, and repeatedly execute the steps from the initial environment perception data output to the current environment feature parameter change rate being less than the preset threshold until the perception tasks of all preset environments are completed.

[0100] The working principle of the above technical solution is as follows: the sensing result acquisition module is started first, which instructs multiple agents to start performing perception operations in a preset environment. These agents are equipped with various sensors, such as cameras, microphones, temperature sensors, etc., which collect information in the environment according to predetermined parameters (such as data collection accuracy and collection frequency); the data collected by each agent is sent back to the sensing result acquisition module, which processes and analyzes the data and identifies environmental characteristic parameters such as temperature, humidity, light intensity, etc.; the environmental perception data acquisition module then intervenes, which compares and matches the data collected by the agent with the preset perception model. This step is to determine whether the environment has changed or whether the agent needs to perform a new perception task. If the environment has changed or there is a new perception requirement, the module will determine the new environmental perception task and assign it to the corresponding agent.

[0101] The feature extraction module processes the new environmental perception data in depth and extracts key features. These features include not only environmental parameters but also the perception performance parameters of the agent. The module continuously monitors the rate of change of environmental feature parameters. Once the rate of change is less than the preset threshold, indicating that the environment is stable or the expected perception depth has been reached, the module will determine that the current perception task is the last task.

[0102] Finally, the perception task module will guide the agent to repeat the above process in another preset environment. This process will continue to cycle until the perception tasks of all preset environments are completed.

[0103] The beneficial effects of the above technical solution are: the intelligent agent can efficiently collect environmental data according to preset parameters to ensure the accuracy and timeliness of the data. Through the preset perception model, the intelligent agent can adapt to environmental changes and dynamically adjust the perception tasks, thereby improving the flexibility and accuracy of environmental monitoring. The feature extraction module can effectively extract key information from a large amount of perception data, reduce the amount of data processing, and speed up the recognition of environmental changes. Completing the perception task module ensures that the intelligent agent can systematically and comprehensively cover all preset perception tasks, avoid perception blind spots, and ensure the integrity of environmental monitoring.

[0104] In another embodiment, constructing a data feature vector unit includes:

[0105] A preprocessing operation module is used to preprocess the acquired environmental perception standard data to eliminate incomplete or abnormal data, input the preprocessed environmental perception standard data into the noise elimination module, perform noise elimination processing on the environmental perception standard data, and obtain denoised data;

[0106] An initial data feature vector construction module is used to extract feature parameters from the denoised data according to preset rules to construct an initial data feature vector;

[0107] Constructing a second data feature vector module, which is used to input the initial data feature vector into the feature screening module, and screen the initial data feature vector according to the set screening logic to obtain an optimized data feature vector;

[0108] Construct a third data feature vector module, which is used to obtain associated parameters from the optimized data feature vector, obtain standard associated data associated with the optimized data feature vector in combination with the set logical relationship, input the standard associated data into the feature construction module, further process the standard associated data, and generate the final data feature vector.

[0109] The working principle of the above technical solution is as follows: The main purpose of the preprocessing operation module is to ensure the quality and integrity of the input data. It uses a series of algorithms and rules to identify and eliminate incomplete or abnormal data. This process may include steps such as data cleaning, data verification, and data filling. The preprocessed data is considered "clean" and can be used for subsequent processing flows.

[0110] The function of the noise removal module is to remove noise from the preprocessed data to improve the credibility and accuracy of the data. Noise comes from many reasons, such as sensor errors, data transmission errors, etc. By using filtering algorithms, smoothing algorithms, etc., the denoising module can reduce the impact of these noises and generate smoother and more reliable data.

[0111] The goal of the module to construct the initial data feature vector is to extract useful information from the denoised data and convert it into a feature vector. These feature parameters may include environmental variables such as temperature, humidity, and light intensity. The preset rules guide how to extract these feature parameters from the data and construct the feature vector.

[0112] The initial feature vector is further processed by the Build Second Data Feature Vector module and the Build Third Data Feature Vector module. The Feature Screening module optimizes the initial data feature vector according to the set screening logic and retains only the most important features. This helps reduce the complexity of the data and may improve the efficiency of subsequent analysis and processing. Finally, by combining the optimized data feature vector and specific logical relationships, standard associated data is extracted and further processed in the Feature Building module to generate the final data feature vector.

[0113] The beneficial effects of the above technical solution are: ensuring the quality and accuracy of the data. The preprocessing operation module removes incomplete or abnormal data, thereby reducing the impact of erroneous data on the final analysis results. The denoising module improves the credibility of the data, making decisions based on the data more reliable. By constructing feature vectors, the data is converted into a format that is easier to process and analyze. The optimization process of the feature screening module helps to extract the most critical information, which may improve the efficiency of analysis and reduce the consumption of computing resources. Finally, by generating the final data feature vector, the entire system is able to process and interpret the data more efficiently, providing users with valuable information and insights.

[0114] In another embodiment, obtaining a multi-agent perception analysis model includes:

[0115] Performing clustering processing on the data feature vectors to generate multiple clustering results;

[0116] Determine a data group to be trained based on the clustering result, and output the data group to be trained;

[0117] Input the data set to be trained into the initial model and perform multi-agent interactive training;

[0118] During the multi-agent interactive training process, the perception data of each agent is obtained and analyzed in real time;

[0119] Based on the perception data of each agent and the feedback data in the interactive training, update the parameters of the initial model and output the updated model parameters;

[0120] Based on the updated model parameters, a multi-agent perception analysis model is generated and output;

[0121] Verify the multi-agent perception analysis model and adjust the parameters of the multi-agent perception analysis model based on the verification results;

[0122] In the adjusted multi-agent perception analysis model, multi-agent interaction training is performed again until the preset convergence conditions are reached, and the final multi-agent perception analysis model is output.

[0123] The working principle of the above technical solution is as follows: clustering the data feature vectors is to group similar data points together, so as to find natural clusters or patterns in the data. This step is usually implemented using algorithms such as K-means, DBSCAN, etc. The clustering results generate multiple cluster centers, each center represents a data group. Based on the clustering results, the data group to be trained can be determined. These data groups are selected from the clusters, and they contain more similar features and behavior patterns, so they are selected as important data for training the model. The data group to be trained is then input into the initial model. In the multi-agent interactive training phase, the models learn how to better understand and predict data through simulated interactions. This training method can improve the generalization ability and interactivity of the model. During the training process, each agent will generate perception data. These data reflect the performance and behavior of the model during the training process. Real-time analysis of these perception data can help monitor the training progress and performance of the model. According to the perception data of each agent and the feedback data in the interactive training, the parameters of the model are updated. This process optimizes the model performance by adjusting the internal parameters of the model. The updated model parameters will generate a new model version.

[0124] The generated multi-agent perception and analysis model needs to be validated to ensure that its performance meets expectations. Validation may include using a dataset that was not involved in training to test the accuracy and generalization ability of the model. Based on the validation results, the model parameters may need to be further adjusted to improve the quality of the model. Repeat the multi-agent interactive training in the adjusted model until the model reaches the preset convergence condition. Convergence means that the performance of the model is no longer significantly improved, or the parameter changes of the model tend to be stable. The final multi-agent perception and analysis model is output for actual perception and analysis tasks.

[0125] The beneficial effects of the above technical solution are: the learning efficiency and accuracy of the model can be improved through the interaction of multiple agents, clustering processing helps to identify key features in the data, and multi-agent interactive training promotes the model's adaptability and learning ability, and real-time analysis of perception data ensures the monitorability and transparency of the training process. By continuously updating model parameters and verifying model performance, the resulting multi-agent perception analysis model has higher accuracy and better generalization ability. This method is particularly suitable for complex and dynamic environments because it can self-adjust and improve to adapt to changing data and conditions.

[0126] In another embodiment, obtaining a dynamic perception result includes:

[0127] Perform hierarchical processing on the perception data to generate multiple sets of perception features at different levels;

[0128] Based on a preset feature extraction algorithm, multiple key features are extracted from the perception feature set, and perception feedback data is generated;

[0129] Input the perception feedback data into the multi-agent perception analysis model, triggering the multi-agent perception analysis model to process the perception feedback data;

[0130] During the processing, the analysis results of each agent are obtained in real time, and multiple sub-level perception results are generated based on the analysis results;

[0131] Integrate the sub-level perception results to generate preliminary dynamic perception results;

[0132] Based on the preset dynamic adjustment rules, the preliminary dynamic perception results are dynamically adjusted to obtain the final dynamic perception results.

[0133] The working principle of the above technical solution is as follows: hierarchical processing of perceptual data is to understand and utilize data from different levels of abstraction. This step usually involves decomposing the data into multiple levels, each level capturing different aspects or details of the data. For example, in image processing, the hierarchical process may include raw pixel data, edge detection, and feature recognition at a higher level. After hierarchical processing, a preset feature extraction algorithm is applied to the perceptual feature set at each level to identify and extract key features. These key features are considered to be crucial for understanding the data and building models. Perceptual feedback data is a collection of these extracted features, which represents important information about the environment or scene.

[0134] The sensory feedback data is then fed into a multi-agent sensory analysis model. This model consists of multiple agents, each responsible for processing a portion of the data and interacting to share information and experience. The model processes the sensory feedback data to identify patterns, make predictions, or generate decisions. During the processing, each agent generates its analysis results in real time. These results reflect the model's processing and interpretation of the sensory feedback data. Based on these analysis results, multiple sub-level sensory results can be generated, which are more specific and detailed, providing deeper insights into the original data.

[0135] The sub-level perception results are then integrated to generate preliminary dynamic perception results. These results integrate the analysis and interpretation of multiple agents and provide a basis for subsequent decisions and behaviors. Finally, the preliminary dynamic perception results are adjusted based on preset dynamic adjustment rules. These rules involve weight allocation, threshold setting or feedback mechanism to ensure the accuracy and adaptability of the perception results. After dynamic adjustment, the final dynamic perception results are obtained, which can be used to guide the behavior of the agent or further decision-making process.

[0136] The beneficial effects of the above technical solution are: improving the accuracy and real-time performance of perception analysis through the collaborative work of multiple agents. Hierarchical processing and feature extraction algorithms help identify key information from complex data, while the multi-agent perception analysis model enhances its analysis capabilities through interaction and sharing of information. The analysis results of each agent are obtained in real time and sub-level perception results are generated, so that the data can be mined and understood more deeply. The final dynamic perception results generated by integrating and dynamically adjusting these results provide accurate perception information for the agent, enabling it to better adapt to the environment and perform tasks.

[0137] In another embodiment, determining a new environment perception task and assigning it to an agent includes:

[0138] Determine, based on the sensing result database, a plurality of second sensing results historically generated by the agent;

[0139] extracting a plurality of first perception features of the second sensing result;

[0140] Obtain a preset perception model library, match the first perception feature with the second perception feature in the perception model library, and if the match is consistent, use the matched second sensing result as the third sensing result, and at the same time, obtain the perception task corresponding to the matched second perception feature and associate it with the corresponding intelligent agent;

[0141] Acquire the historical environment item corresponding to the third sensing result, and at the same time, acquire the sensing generation time corresponding to the third sensing result;

[0142] Determine from the sensing result database a plurality of fourth sensing results corresponding to the historical environmental item and generated within a preset time period before and after the generation time;

[0143] Inputting all fourth sensing results into a preset task analysis model, obtaining at least one new perception task output by the task analysis model, and accumulating and calculating the perception tasks associated with the intelligent agent to obtain a task sum;

[0144] If the perception task is the first task and / or the perception task is the fourth task in the second task excluding the third task and / or the total of the tasks is greater than or equal to the preset total threshold of the first task, the agent perception task allocation fails;

[0145] Otherwise, obtaining a preset environment perception task process record library, and determining a plurality of environment perception task process records corresponding to the perception project from the environment perception task process record library;

[0146] Input all environmental perception task process records into a preset importance analysis model to obtain the importance of the importance analysis model output;

[0147] Obtain a second task sum threshold corresponding to the importance;

[0148] If the total of the tasks is greater than or equal to the second task total threshold, the agent perceives that the task allocation has failed;

[0149] Otherwise, the agent perceives that the task assignment is successful.

[0150] The working principle of the above technical solution is as follows: historical data analysis, obtaining the second sensory result generated by the agent in history from the sensory result database, extracting the first sensory feature of the second sensory result, this step uses historical data to provide a basis for subsequent analysis. Model matching, matching the first sensory feature with the second sensory feature in the sensory model library, and determining the third sensory result and the corresponding sensory task if the match is successful. This step determines the appropriate sensory task through model matching. Environmental analysis, obtaining the historical environmental item and generation time corresponding to the third sensory result, determining the fourth sensory result in the relevant time period, this step analyzes the environmental background of the sensory result. Task analysis, inputting the fourth sensory result into the task analysis model, obtaining a new sensory task, and accumulating and calculating the sum of sensory tasks associated with the agent. This step generates a new task based on the environmental analysis results. Task screening, preliminary screening is performed based on the task type and the task sum, this step filters out unsuitable tasks. Importance analysis, inputting the environmental perception task process record into the importance analysis model, obtaining the importance and the corresponding task sum threshold, this step further analyzes the importance of the task. Final judgment, comparing the task sum with the threshold, deciding whether to pass the task allocation, this step makes the final task allocation decision.

[0151] The beneficial effects of the above technical solution are: automatically assigning tasks using historical data and multiple models; ensuring that tasks match the capabilities of the agent through multiple screenings; dynamically adjusting task assignments based on the environment and historical records; preventing the agent from having too many tasks through task sum control; considering the importance of tasks and giving priority to important tasks; adjusting various thresholds and models as needed; and recording detailed task assignment processes for subsequent analysis.

[0152] In another embodiment, preprocessing the acquired environment perception standard data includes:

[0153] Determine a preliminary processing result from the environmental perception standard data, and extract multiple data features from the preliminary processing result;

[0154] Determine whether the data features meet the preset integrity standards. If so, execute: further process the preliminary processing results, remove redundant data, and output the processed environmental perception standard data;

[0155] If not, the incomplete or abnormal data features are marked and removed from the environmental perception standard data;

[0156] It is determined again whether the remaining data features in the environmental perception standard data after the elimination meet the preset integrity standard. If so, further processing is performed on the data features and the final processing result is output;

[0157] If not, repeat the process: remove incomplete or abnormal data from the environment perception standard data set until the data features meet the preset integrity standards or the size of the environment perception standard data reaches the preset minimum threshold.

[0158] The working principle of the above technical solution is: determine the preliminary processing results from the environmental perception standard data, and extract multiple data features in the preliminary processing results; judge whether the data features meet the preset integrity standards; judge whether the data features meet the preset integrity standards, if so, execute: further process the preliminary processing results, remove redundant data, and output the processed environmental perception standard data; if not, mark the incomplete or abnormal data features, and eliminate the incomplete or abnormal data features from the environmental perception standard data; again judge whether the remaining data features in the environmental perception standard data after elimination meet the preset integrity standards, if so, perform further processing on the data features, and output the final processing results; if not, repeat the execution: eliminate incomplete or abnormal data from the environmental perception standard data set until the data features meet the preset integrity standards or the size of the environmental perception standard data reaches the preset minimum threshold.

[0159] The beneficial effects of the above technical solution are: through multiple integrity checks and data cleaning, the quality of the final output data is ensured. Different processing paths are automatically selected according to the integrity of data features; incomplete or abnormal data features can be identified and eliminated to improve data reliability; redundant data can be removed to improve data efficiency when the integrity standards are met; the data set is continuously optimized through cyclic processing until the expected standards are met; the integrity standards and minimum data thresholds can be adjusted as needed. While ensuring data quality, sufficient data volume is retained through minimum threshold control.

[0160] In another embodiment, a plurality of clustering results are generated, including:

[0161] Obtaining a data feature vector set, the data feature vector set includes multiple data feature vectors;

[0162] Performing a preprocessing step on the data feature vector set, the preprocessing step including normalizing the data feature vector;

[0163] According to a preset clustering algorithm, the similarity between each data feature vector and other data feature vectors is determined, and according to a similarity matrix, the data feature vector set is divided into a plurality of preliminary clustering results;

[0164] According to the preset optimization strategy, the preliminary clustering results are optimized, and the optimization includes readjusting the clustering boundaries to improve the overall consistency of the clustering results and generate the final clustering results;

[0165] Determine whether the final clustering result meets the preset clustering evaluation criteria. If so, output the final clustering result.

[0166] The working principle of the above technical solution is as follows: obtain a data feature vector set, which contains multiple data feature vectors; perform preprocessing on the data feature vector set, mainly including normalization processing, which is a very important step that can eliminate the dimensional differences between different features and make subsequent clustering more accurate; use a preset clustering algorithm to calculate the similarity between data feature vectors, and divide the data feature vector set into multiple preliminary clustering results based on the obtained similarity matrix. According to the preset optimization strategy, the preliminary clustering results are optimized, which includes readjusting the clustering boundaries to improve the overall consistency of the clustering results, and finally generating optimized clustering results.

[0167] Determine whether the final clustering result meets the preset clustering evaluation criteria. If so, output the final clustering result.

[0168] The beneficial effects of the above technical solution are: ensuring the comparability between different features through normalization processing; selecting different clustering algorithms and optimization strategies as needed; improving the quality of clustering results through optimization processing; ensuring the reliability of clustering results using preset evaluation criteria; being applicable to various types of data sets and clustering requirements; and adding more preprocessing steps or optimization strategies as needed.

[0169] In another embodiment, a plurality of perceptual feature sets at different levels are generated, including:

[0170] Acquire a perception data set, where the perception data set includes a plurality of perception data points;

[0171] Divide the perception data set into several preliminary perception levels according to preset hierarchical processing rules, where the hierarchical processing rules are determined based on the characteristic dimensions, data density and relevance of the perception data;

[0172] Performing hierarchical optimization on the preliminary perception hierarchies, which includes adjusting hierarchical boundaries and merging similar hierarchies to ensure the rationality of hierarchical division and generate multiple optimized perception hierarchies;

[0173] Based on the data features of each perception level, feature sets of the perception level are extracted respectively, and the feature sets are generated by a preset feature extraction algorithm;

[0174] The extracted perceptual feature sets are combined in hierarchical order to generate multiple perceptual feature sets at different levels.

[0175] The working principle of the above technical solution is as follows: first, a data set containing multiple perception data points is collected; according to the preset hierarchical processing rules, the entire perception data set is divided into several preliminary perception levels. These rules usually take into account the characteristic dimensions of the perception data (such as color, shape, size, etc.), data density (the distribution of data points in space or time) and correlation (the degree of association between data points).

[0176] Optimize the initial perception levels, including adjusting the boundaries of the levels and merging similar levels. The purpose of optimization is to ensure the rationality of the level division so that each level can more accurately reflect the inherent structure and characteristics of the data. The optimized perception levels should better reflect the essential differences of the data. Based on each optimized perception level, use the preset feature extraction algorithm to extract the feature set of the level. These feature sets contain the key information of the data at the level and can represent the data characteristics of the level.

[0177] The feature sets extracted at each level are combined in hierarchical order to form multiple perceptual feature sets at different levels. These sets can be used for subsequent data analysis, pattern recognition or decision making.

[0178] According to the preset hierarchical processing rules, the perception data set is divided into several preliminary perception levels, including:

[0179] Classifying the perception data into multiple dimension sets according to the feature dimensions;

[0180] Obtain data density information in the perception data set, classify each dimension set according to the data density information, and use the classified data set as preliminary perception level data;

[0181] According to the relevance of the preliminary perception level data, the data is further segmented, and the segmented data is stored as the final perception level data;

[0182] The dimension sets in each perception level data and the classified data density information are matched one by one to generate the perception level corresponding data and build a hierarchical processing structure.

[0183] The beneficial effect of the above technical solution is: useful features are extracted from complex perception data, which are used for higher-level tasks such as classification, prediction or control. The efficiency and accuracy of feature extraction can be improved through stratification and optimization.

[0184] According to the preset hierarchical processing rules, the perception data set is divided into several preliminary perception levels, including:

[0185] In another embodiment, obtaining a set of perception data includes:

[0186] Get request information for sensing data;

[0187] According to the perception data acquisition mechanism, identifying whether the perception data meets the preset conditions, wherein the perception data acquisition mechanism is used to configure the perception data acquisition strategy;

[0188] When the perception data meets the preset conditions, the relevance of the current perception data points is determined based on the data association analysis mechanism;

[0189] When the perception data points are relevant, the perception data points that meet the conditions are aggregated to form a perception data set.

[0190] The working principle of the above technical solution is as follows: the system receives a request to obtain sensory data. This usually involves specifying parameters such as data type, source, time range, etc. The system identifies whether the sensory data meets the requirements based on the preset sensory data acquisition strategy. This includes filtering out data that does not meet quality standards, collecting only specific types of data, or limiting data collection by time, etc.

[0191] For sensory data that meets the preset conditions, the system uses a data association analysis mechanism to determine the correlation between these data points, which involves time series analysis, spatial analysis, or pattern recognition algorithms.

[0192] Through association analysis, the system determines which perception data points are related, that is, there is a certain connection or pattern between them; when it is determined that there is a correlation between the perception data points, the system aggregates these data points to form a perception data set, which can be used for further analysis or decision-making.

[0193] The beneficial effect of the above technical solution is: valuable information can be extracted from a large amount of perception data. In this way, the system can process and utilize the perception data more efficiently, thereby improving the quality and speed of decision-making.

[0194] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of the same technology, the present invention is also intended to include these changes and variations.

Claims

1. An environment perception system based on multi-agent interaction, characterized in that: include: An environment perception data acquisition unit is used to acquire multi-agent environment perception data, where the environment perception data includes environment feature parameters and agent perception parameters; A data feature vector construction unit is used to obtain environmental perception standard data, perform noise elimination processing on the perception standard data, construct a data feature vector, and obtain standard associated data from the environmental perception standard data; Constructing a perception analysis model unit, which is used to cluster data feature vectors according to standard associated data to obtain a data group to be trained, and to perform multi-agent interactive training on the initial model using the data group to be trained to obtain a multi-agent perception analysis model; The perception analysis unit is used to perform hierarchical processing on environmental perception data, obtain perception feedback data, input the perception feedback data into the multi-agent perception analysis model, and obtain dynamic perception results.

2. The multi-agent interaction-based environment perception system according to claim 1, characterized in that: The environment perception data acquisition unit includes: The sensing result acquisition module is used to control multiple agents to simultaneously start sensing operations in a preset environment, output initial environmental sensing data, and determine the sensing results of each agent regarding the current environmental sensing data. The sensing results include data collection accuracy, collection frequency, and corresponding environmental characteristic parameters. The environment perception data acquisition module is used to match and analyze the sensing results of each intelligent agent with the preset perception model, determine new environment perception tasks and assign them to the intelligent agent, and generate new environment perception data; A feature extraction module is used to extract features from new environmental perception data, obtain new environmental feature parameters and agent perception parameters, and determine multiple new environmental perception tasks until the rate of change of the environmental feature parameters is less than a preset threshold, or the current perception task is determined to be the last task; The perception task completion module is used to control multiple intelligent agents to perform perception operations in another preset environment, and repeatedly execute the steps from the initial environment perception data output to the current environment feature parameter change rate being less than the preset threshold until the perception tasks of all preset environments are completed.

3. The multi-agent interaction-based environment perception system according to claim 1, characterized in that: The unit for constructing data feature vector includes: A preprocessing operation module is used to preprocess the acquired environmental perception standard data to eliminate incomplete or abnormal data, input the preprocessed environmental perception standard data into the noise elimination module, perform noise elimination processing on the environmental perception standard data, and obtain denoised data; An initial data feature vector construction module is used to extract feature parameters from the denoised data according to preset rules to construct an initial data feature vector; Constructing a second data feature vector module, which is used to input the initial data feature vector into the feature screening module, and screen the initial data feature vector according to the set screening logic to obtain an optimized data feature vector; Construct a third data feature vector module, which is used to obtain associated parameters from the optimized data feature vector, obtain standard associated data associated with the optimized data feature vector in combination with the set logical relationship, input the standard associated data into the feature construction module, further process the standard associated data, and generate the final data feature vector.

4. The environment perception system based on multi-agent interaction according to claim 1 is characterized in that: Get multi-agent perception analysis models, including: Performing clustering processing on the data feature vectors to generate multiple clustering results; Determine a data group to be trained based on the clustering result, and output the data group to be trained; Input the data set to be trained into the initial model and perform multi-agent interactive training; During the multi-agent interactive training process, the perception data of each agent is obtained and analyzed in real time; Based on the perception data of each agent and the feedback data in the interactive training, update the parameters of the initial model and output the updated model parameters; Based on the updated model parameters, a multi-agent perception analysis model is generated and output; Verify the multi-agent perception analysis model and adjust the parameters of the multi-agent perception analysis model based on the verification results; In the adjusted multi-agent perception analysis model, multi-agent interaction training is performed again until the preset convergence conditions are reached, and the final multi-agent perception analysis model is output.

5. The multi-agent interactive environment perception system according to claim 1, characterized in that: Get dynamic perception results, including: Perform hierarchical processing on the perception data to generate multiple sets of perception features at different levels; Based on a preset feature extraction algorithm, multiple key features are extracted from the perception feature set, and perception feedback data is generated; Input the perception feedback data into the multi-agent perception analysis model, triggering the multi-agent perception analysis model to process the perception feedback data; During the processing, the analysis results of each agent are obtained in real time, and multiple sub-level perception results are generated based on the analysis results; Integrate the sub-level perception results to generate preliminary dynamic perception results; Based on the preset dynamic adjustment rules, the preliminary dynamic perception results are dynamically adjusted to obtain the final dynamic perception results.

6. The environment perception system based on multi-agent interaction according to claim 2 is characterized in that: Identify new environmental perception tasks and assign them to agents, including: Determine, based on the sensing result database, a plurality of second sensing results historically generated by the agent; extracting a plurality of first perception features of the second sensing result; Obtain a preset perception model library, match the first perception feature with the second perception feature in the perception model library, and if the match is consistent, use the matched second sensing result as the third sensing result, and at the same time, obtain the perception task corresponding to the matched second perception feature and associate it with the corresponding intelligent agent; Acquire the historical environment item corresponding to the third sensing result, and at the same time, acquire the sensing generation time corresponding to the third sensing result; Determine from the sensing result database a plurality of fourth sensing results corresponding to the historical environmental item and generated within a preset time period before and after the generation time; Inputting all fourth sensing results into a preset task analysis model, obtaining at least one new perception task output by the task analysis model, and accumulating and calculating the perception tasks associated with the intelligent agent to obtain a task sum; If the perception task is the first task and / or the perception task is the fourth task in the second task excluding the third task and / or the total of the tasks is greater than or equal to the preset total threshold of the first task, the agent perception task allocation fails; Otherwise, obtaining a preset environment perception task process record library, and determining a plurality of environment perception task process records corresponding to the perception project from the environment perception task process record library; Input all environmental perception task process records into a preset importance analysis model to obtain the importance of the importance analysis model output; Obtain a second task sum threshold corresponding to the importance; If the total of the tasks is greater than or equal to the second task total threshold, the agent perceives that the task assignment has failed; Otherwise, the agent perceives that the task assignment is successful.

7. The environment perception system based on multi-agent interaction according to claim 3 is characterized in that: Preprocess the acquired environmental perception standard data, including: Determine a preliminary processing result from the environmental perception standard data, and extract multiple data features from the preliminary processing result; Determine whether the data features meet the preset integrity standards. If so, execute: further process the preliminary processing results, remove redundant data, and output the processed environmental perception standard data; If not, the incomplete or abnormal data features are marked and removed from the environmental perception standard data; It is determined again whether the remaining data features in the environmental perception standard data after the elimination meet the preset integrity standard. If so, further processing is performed on the data features and the final processing result is output; If not, repeat the process: remove incomplete or abnormal data from the environment perception standard data set until the data features meet the preset integrity standards or the size of the environment perception standard data reaches the preset minimum threshold.

8. The multi-agent interaction-based environment perception system according to claim 4, characterized in that: Generates multiple clustering results, including: Obtaining a data feature vector set, where the data feature vector set includes multiple data feature vectors; Performing a preprocessing step on the data feature vector set, the preprocessing step including normalizing the data feature vector; According to a preset clustering algorithm, the similarity between each data feature vector and other data feature vectors is determined, and according to a similarity matrix, the data feature vector set is divided into a plurality of preliminary clustering results; According to the preset optimization strategy, the preliminary clustering results are optimized, and the optimization includes readjusting the clustering boundaries to improve the overall consistency of the clustering results and generate the final clustering results; Determine whether the final clustering result meets the preset clustering evaluation criteria. If so, output the final clustering result.

9. The environment perception system based on multi-agent interaction according to claim 5, characterized in that: Generate multiple perceptual feature sets at different levels, including: Acquire a perception data set, where the perception data set includes a plurality of perception data points; Divide the perception data set into several preliminary perception levels according to preset hierarchical processing rules, where the hierarchical processing rules are determined based on the characteristic dimensions, data density and relevance of the perception data; Performing hierarchical optimization on the preliminary perception hierarchies, which includes adjusting hierarchical boundaries and merging similar hierarchies to ensure the rationality of hierarchical division and generate multiple optimized perception hierarchies; Based on the data features of each perception level, feature sets of the perception level are extracted respectively, and the feature sets are generated by a preset feature extraction algorithm; The extracted perceptual feature sets are combined in hierarchical order to generate multiple perceptual feature sets at different levels.

10. The environment perception system based on multi-agent interaction according to claim 9, characterized in that: Get the perception data set, including: Get request information for sensing data; According to the perception data acquisition mechanism, identifying whether the perception data meets the preset conditions, wherein the perception data acquisition mechanism is used to configure the perception data acquisition strategy; When the perception data meets the preset conditions, the relevance of the current perception data points is determined based on the data association analysis mechanism; When the perception data points are relevant, the perception data points that meet the conditions are aggregated to form a perception data set.

Images