CN117894311A

CN117894311A - Bionic robot in aspect of voice recognition control

Info

Publication number: CN117894311A
Application number: CN202410225452.3A
Authority: CN
Inventors: 刘志坚; 刘旭; 原俊杰; 潘远娇; 刘贞; 刘运钦; 陈世超; 宋建达; 胡晓伟; 张洪国
Original assignee: Dalian Bo Tao Culture Polytron Technologies Inc
Current assignee: Dalian Bo Tao Culture Polytron Technologies Inc
Priority date: 2024-02-29
Filing date: 2024-02-29
Publication date: 2024-04-16

Abstract

The invention relates to the technical field of voice recognition control, in particular to a bionic robot in the aspect of voice recognition control, which comprises a voice recognition module, a user intention analysis module, a behavior mode recognition module, a security enhancement module, a task priority adjustment module, an abnormal state recognition module, a decision support module and a response execution module. In the invention, a convolutional neural network and a long-short-term memory network are adopted to deeply analyze voice signals, so that the recognition accuracy is improved, a bi-directional encoder represents a slave converter model, text semantics are deeply analyzed, the intention of a user is accurately grasped, a hidden Markov model and a voiceprint recognition technology are adopted, behavior pattern recognition and safety are optimized, a genetic algorithm and an isolated forest algorithm are realized, resource optimization and anomaly detection are realized, the response and stability of a system are enhanced, risk and benefit analysis is performed through a decision tree and a Bayesian network, and the execution accuracy is improved by utilizing a state machine model and a rule engine.

Description

Bionic robot in aspect of voice recognition control

Technical Field

The invention relates to the technical field of voice recognition control, in particular to a bionic robot in the aspect of voice recognition control.

Background

The field of speech recognition control technology has focused on developing and optimizing systems capable of understanding, processing, and executing human speech commands. At the heart of technology in this field is the conversion of speech input into a machine-understandable format, which in turn performs corresponding actions or commands. Including from simple command recognition to complex conversational and interactive understanding. The field combines a plurality of technologies such as voice processing, natural language understanding and the like, and aims to improve the natural interaction capability of the machine, so that the machine can respond to the demands of human users more intelligently.

The bionic robot in the aspect of voice recognition control is a robot integrating voice recognition and control technology, and aims to operate and interact through voice instructions. The method aims to realize a more natural and visual man-machine interaction mode, and allows a user to control the robot through a verbal command, so that the convenience and efficiency of operation are improved. By simulating the behavior and function of a human or animal, the robot is able to exhibit greater flexibility and adaptability in performing certain tasks. The achievement of the effect means that the robot can accurately understand and respond to complex voice commands, perform diversified tasks such as searching, carrying and navigating, and further improve user experience and practicality of the robot.

The bionic robot in the aspect of traditional voice recognition control has obvious defects in the aspects of accuracy and efficiency of voice recognition, accurate analysis of user intention, recognition of behavior patterns, safety verification, task priority adjustment, abnormal state detection, decision support, response execution and the like, the bionic robot in the aspect of traditional voice recognition control depends on a simpler algorithm, lacks support of deep learning technology, causes larger voice recognition error, is difficult to accurately capture the user intention and the behavior patterns, and the safety verification depends on static passwords or a simple identity verification mode, is easy to be threatened by safety, and in the aspects of task scheduling and abnormal state recognition, resource allocation is improper and abnormal state omission is often caused due to the lack of efficient algorithm support, so that the overall performance and user experience of a system are influenced.

Disclosure of Invention

The invention aims to solve the defects in the prior art, and provides a bionic robot in the aspect of voice recognition control.

In order to achieve the above purpose, the present invention adopts the following technical scheme: the bionic robot in the aspect of voice recognition control comprises a voice recognition module, a user intention analysis module, a behavior mode recognition module, a safety enhancement module, a task priority adjustment module, an abnormal state recognition module, a decision support module and a response execution module;

The voice recognition module processes voice signals to extract time-frequency characteristics based on voice input of a user by adopting a convolutional neural network algorithm, captures the dependency relationship in a voice instruction by the processing characteristics through a long-term and short-term memory network, and converts the dependency relationship into a voice instruction in a text format to generate voice text information;

The user intention analysis module analyzes text contents by utilizing a bi-directional encoder representation slave converter model based on voice text information, analyzes instruction meanings and context environments of a user, recognizes user demands and intentions and generates user intention information;

The behavior pattern recognition module is used for analyzing historical interaction data of a user by using a hidden Markov model based on user intention information, recognizing behavior habits and patterns of the user, recognizing abnormal behavior patterns by comparing current behaviors with the historical patterns, and generating a behavior pattern analysis result;

The security enhancement module performs voice identity verification by using a deep neural network based on a behavior pattern analysis result and combining a voiceprint recognition technology and a user behavior pattern, analyzes whether the behavior is consistent with a known security pattern, evaluates potential security threats, dynamically adjusts security measures according to risk levels, and generates a security verification result;

the task priority adjustment module analyzes the current system resource condition and the task emergency degree based on the security verification result, performs priority ordering on a task list to be executed, analyzes the resource allocation efficiency and the timeliness of task completion, dynamically adjusts a task queue and generates a task execution plan;

The abnormal state identification module is used for carrying out real-time monitoring and analysis on data by utilizing an isolated forest algorithm based on a task execution plan and combining data input of multiple sensors of the robot, identifying an abnormal state deviating from a normal behavior mode, evaluating the internal state and external environmental factors of the robot, and generating an abnormal state result;

the decision support module analyzes risks and benefits of the action schemes by using decision trees and Bayesian networks based on abnormal state results, analyzes potential results of the multi-action schemes, and selects matched action strategies by evaluating differential decision paths to generate decision support schemes;

The response execution module adopts a state machine model to determine the operation state and the expected conversion of the current robot based on a decision support scheme, applies a rule engine, matches corresponding execution rules according to the operation state, the safety alarm requirement and the fault recovery requirement, executes response measures, and comprises behavior adjustment, safety alarm emission and fault recovery to generate a decision execution scheme.

As a further aspect of the present invention, the voice text information includes transcription content, confidence score and time stamp, the user intention information includes operation targets, operation objects and operation attributes, the behavior pattern analysis results include regular behavior patterns, abnormal behavior indexes and behavior trend analysis, the security verification results include user authentication states, behavior compliance assessment and security risk levels, the task execution plan includes task sequences, task priorities and predicted completion times, the abnormal state results include abnormal types, abnormal degrees and recommended response measures, the decision support scheme includes recommended action schemes, potential risk analysis and expected effect assessment, and the decision execution scheme includes execution states, execution result assessment.

As a further scheme of the invention, the voice recognition module comprises a feature extraction sub-module, a time sequence analysis sub-module and a text conversion sub-module;

the characteristic extraction submodule is based on voice input of a user, analyzes voice signals by adopting a deep convolutional neural network algorithm, extracts time-frequency characteristics from an original audio signal by constructing a plurality of convolution layers and pooling layers, and generates time-frequency characteristic data;

The time sequence analysis sub-module is used for performing time sequence analysis on the extracted features by utilizing a long-period memory network algorithm based on time-frequency feature data, processing and memorizing long-term dependence information through a gate control mechanism, capturing the structure and the context of a voice instruction and generating a structured voice feature;

the text conversion submodule converts the structured voice features into corresponding text instructions by adopting a sequence-to-sequence learning model based on the structured voice features and through the sequence-to-sequence learning model with an attention mechanism to generate voice text information.

As a further scheme of the invention, the user intention analysis module comprises a semantic analysis sub-module, an intention recognition sub-module and a requirement mapping sub-module;

The semantic analysis submodule adopts a bidirectional encoder to express a slave converter model based on voice text information, performs semantic analysis on text content, learns semantic information through a pre-trained text corpus, analyzes meaning and context in the text, and generates a semantic analysis result;

The intention recognition sub-module is used for analyzing and recognizing the demands of users by applying an intention recognition algorithm in a natural language processing technology based on a semantic analysis result, recognizing the intention of the users by analyzing verbs, nouns and phrases in the semantic analysis result and generating intention information;

The demand mapping sub-module maps the intention of the user to an execution operation by using a logical reasoning algorithm based on the intention information, and generates user intention information by analyzing the relationship between the intention of the user and the robot service and the function and matching the service and the operation.

As a further scheme of the invention, the behavior pattern recognition module comprises a data analysis sub-module, a pattern establishment sub-module and an abnormality detection sub-module;

The data analysis submodule analyzes historical interaction data of the user based on user intention information by adopting a statistical analysis method, identifies the frequency, time distribution and preference mode of the user behavior and generates a statistical analysis result of the user behavior;

The mode establishing submodule is used for modeling the sequence and the conversion probability of the user behavior by adopting a hidden Markov model based on the statistical analysis result of the user behavior, revealing the potential state and the conversion rule of the user behavior by analyzing the conversion probability of the user behavior, establishing the behavior habit and the mode of the user and generating the user behavior mode information;

The abnormality detection submodule identifies deviation between current behaviors and historical behavior patterns by using an abnormality detection algorithm based on user behavior pattern information, quantifies the deviation degree of the behavior patterns, carries out abnormality behavior identification and generates a behavior pattern analysis result.

As a further scheme of the invention, the security enhancement module comprises an identity verification sub-module, a behavior compliance sub-module and a risk rating sub-module;

The identity verification submodule is used for carrying out identity verification on the voice of the user by adopting a deep neural network based on the analysis result of the behavior mode and combining with the voiceprint recognition technology, analyzing the characteristics of the voice, including tone, rhythm and voice texture, and carrying out voiceprint model matching to generate a voice identity verification result;

the behavior compliance submodule analyzes whether the current behavior of the user is consistent with the recorded safety behavior mode or not by adopting a behavior pattern matching algorithm based on the voice identity verification result, and identifies potential non-compliance behaviors by comparing the behavior pattern of the user with a predefined compliance standard to generate a behavior compliance analysis result;

The risk rating submodule evaluates and ranks the potential security threats based on the behavior compliance analysis result by using a risk evaluation model, evaluates the risk level by analyzing the behavior deviation degree, the historical security event and the current operation environment factor, adjusts the security measures and generates a security verification result.

As a further scheme of the invention, the task priority adjustment module comprises an emergency analysis sub-module, a resource matching sub-module and a scheduling strategy sub-module;

The emergency degree analysis submodule evaluates the emergency degree of a plurality of tasks according to a safety verification result by adopting a multi-factor decision analysis method, analyzes the deadline, the relevance and the demand factors of the tasks on resources, distributes the emergency degree score of each task and generates task emergency degree scoring information;

the resource matching submodule analyzes and matches the current system resources by using a linear programming algorithm based on the task urgency scoring information, and the matching task obtains the priority of the required resources by optimizing the resource allocation scheme to generate a resource matching scheme;

The scheduling strategy submodule adopts a genetic algorithm based on a resource matching scheme to perform priority ordering and scheduling strategy optimization on a task list to be executed, and iteratively captures a task execution sequence through simulating a natural selection and genetic mechanism to generate a task execution plan.

As a further scheme of the invention, the abnormal state identification module comprises a state monitoring sub-module, an abnormal analysis sub-module and a result generation sub-module;

The state monitoring submodule is used for carrying out real-time monitoring on the working state and the external environment of the robot by combining the data input of multiple sensors of the robot based on a task execution plan and adopting a time sequence analysis method to generate a state monitoring result;

The abnormal analysis submodule applies an isolation forest algorithm to detect abnormal points of the monitored data based on the state monitoring result, identifies the isolation degree of the data points by constructing a plurality of isolation trees, identifies an abnormal state deviating from a normal behavior mode and generates an abnormal state analysis result;

The result generation submodule evaluates the abnormal state based on the analysis result of the abnormal state by adopting a report generation technology, wherein the abnormal state comprises abnormal type, degree and influence, and provides an abnormal state description and response scheme by integrating the analysis result and the context information to generate the abnormal state result.

As a further scheme of the invention, the decision support module comprises a risk analysis sub-module, a scheme optimization sub-module and a strategy preparation sub-module;

The risk analysis submodule analyzes potential risks and benefits of each action scheme by adopting a decision tree algorithm based on abnormal state results, evaluates results caused by the differential schemes by constructing differential decision paths, performs probabilistic analysis on uncertain factors by combining with a Bayesian network, gives risk scores and benefit expectations to each action scheme, and generates action scheme risk evaluation results;

The scheme optimizing submodule optimizes the action scheme by utilizing a multi-objective optimizing algorithm through pareto front analysis based on the action scheme risk assessment result, identifies action scheme combinations, and generates an optimized action scheme with controllable optimized scheme risk;

The strategy making sub-module adopts a rule engine to make action strategies according to the optimized post-action scheme and the operation capability and the environmental condition of the current robot, and makes execution steps and expected targets by analyzing the content and the implementation conditions of the scheme to generate a decision support scheme.

As a further scheme of the invention, the response execution module comprises an operation implementation sub-module, a safety measure sub-module and a feedback arrangement sub-module;

The operation implementation submodule is based on a decision support scheme, a state machine model is applied, the current operation state and the required conversion of the robot are analyzed, the behavior mode of the robot is adjusted according to the guidance of the scheme, the behavior adjustment and the task starting operation are executed, and an operation adjustment result is generated;

The safety measure submodule adopts a dynamic risk management strategy based on an operation adjustment result, dynamically adjusts safety measures according to action risk levels and current environmental conditions, and generates a safety adjustment scheme by monitoring the operation environment and the state of a robot in real time, sending out a safety alarm and activating a fault recovery program;

The feedback arrangement submodule is based on a safety adjustment scheme, adopts a feedback analysis technology, collects operation implementation and safety adjustment results, evaluates and arranges the execution process and the results, analyzes the success rate of operation and the processing efficiency index of a safety event, makes an improvement opinion and a follow-up scheme, and generates a decision execution scheme.

Compared with the prior art, the invention has the advantages and positive effects that:

According to the invention, by adopting the convolutional neural network and the long-term and short-term memory network to carry out deep analysis on voice signals, the time-frequency characteristics can be extracted efficiently, the dependency relationship in voice instructions can be captured, and then the voice instructions are converted into voice instructions in a text format, the accuracy and efficiency of voice recognition are greatly improved, the bidirectional encoder represents the application of a slave converter model, so that the semantic analysis of text content is deeper, the instruction meaning and the context environment of a user can be accurately analyzed, the user intention can be more accurately mastered, the hidden Markov model is used in behavior pattern recognition, the analysis process of historical interaction data of the user is optimized, the behavior habit and pattern of the user can be effectively identified, the voice pattern recognition technology and the deep neural network are combined, the security enhancement module can provide higher security guarantee in the aspects of identity verification and behavior analysis, the introduction of a genetic algorithm and an isolation forest algorithm are realized, the optimal allocation of resources and the accurate detection of abnormal states are realized on task priority adjustment and abnormal state recognition respectively, the response speed and stability of a system are remarkably improved, the decision support module uses a decision tree and a network, the decision engine is used for carrying out the decision engine, the decision engine is more accurate and the decision engine is used for carrying out the scientific and flexible state analysis, and the flexible state machine is more flexible.

Drawings

FIG. 1 is a flow chart of the present invention;

FIG. 2 is a schematic diagram of a frame of the present invention;

FIG. 3 is a flow chart of a speech recognition module according to the present invention;

FIG. 4 is a flowchart of a user intent analysis module of the present invention;

FIG. 5 is a flow chart of a behavior pattern recognition module according to the present invention;

FIG. 6 is a flow chart of a security enhancement module of the present invention;

FIG. 7 is a flow chart of a task priority adjustment module according to the present invention;

FIG. 8 is a flow chart of an abnormal state identification module according to the present invention;

FIG. 9 is a flow chart of a decision support module according to the present invention;

FIG. 10 is a flow chart of a response execution module of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

In the description of the present invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like indicate orientations or positional relationships based on the orientation or positional relationships shown in the drawings, merely to facilitate describing the present invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention. Furthermore, in the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.

Example 1

Referring to fig. 1 to 2, a bionic robot in terms of voice recognition control includes a voice recognition module, a user intention analysis module, a behavior pattern recognition module, a security enhancement module, a task priority adjustment module, an abnormal state recognition module, a decision support module, and a response execution module;

The voice recognition module processes voice signals to extract time-frequency characteristics based on voice input of a user by adopting a convolutional neural network algorithm, processes the characteristics to capture the dependency relationship in a voice instruction through a long-term and short-term memory network, and converts the dependency relationship into a voice instruction in a text format to generate voice text information;

The behavior pattern recognition module is used for analyzing historical interaction data of a user by using a hidden Markov model based on user intention information, recognizing behavior habits and patterns of the user, recognizing abnormal behavior patterns by comparing current behaviors with historical patterns, and generating a behavior pattern analysis result;

The task priority adjustment module analyzes the current system resource condition and the task emergency degree based on the security verification result, prioritizes a task list to be executed, analyzes the resource allocation efficiency and the timeliness of task completion, dynamically adjusts a task queue and generates a task execution plan;

the decision support module analyzes risks and benefits of the action scheme by using a decision tree and a Bayesian network based on the abnormal state result, analyzes potential results of the multi-action scheme, and selects a matched action strategy by evaluating a differential decision path to generate a decision support scheme;

The response execution module adopts a state machine model to determine the current operation state and expected conversion of the robot based on the decision support scheme, applies a rule engine, matches corresponding execution rules according to the operation state, the safety alarm requirement and the fault recovery requirement, executes response measures, and comprises behavior adjustment, safety alarm emission and fault recovery to generate a decision execution scheme.

The voice text information comprises transcribed content, confidence scores and time stamps, the user intention information comprises operation targets, operation objects and operation attributes, the behavior pattern analysis results comprise conventional behavior patterns, abnormal behavior indexes and behavior trend analysis, the security verification results comprise user identity verification states, behavior compliance assessment and security risk levels, the task execution plans comprise task sequences, task priorities and predicted completion times, the abnormal state results comprise abnormal types, abnormal degrees and recommended response measures, the decision support schemes comprise recommended action schemes, potential risk analysis and expected effect assessment, and the decision execution schemes comprise execution states and execution result assessment.

In the voice recognition module, the voice input of a user is processed by adopting a convolutional neural network algorithm, a plurality of convolutional layers are constructed to extract time-frequency characteristics in voice signals, the characteristics comprise information such as frequency, rhythm, tone and the like of voice, and the deep learning capability of the neural network is utilized to carry out efficient characteristic extraction on complex voice data. The time sequence dependency relationship in the voice instruction is captured through the processing characteristics of the long-term memory network, and the long-term memory network can memorize long-term dependency information by processing time sequence data through the special gating mechanism of the long-term memory network, so that the voice instruction of the user can be accurately identified and understood. The processed features are converted into voice instructions in text format, generating voice text information with transcribed content, confidence scores and time stamps, which provide a basis for subsequent user intent analysis.

In the user intent analysis module, the generated phonetic text information is analyzed from the transducer model by using bi-directional encoder representations. Through the pre-trained bidirectional converter network, the semantics of text content are deeply analyzed, the meaning of user instructions and the context environment are analyzed, and the comprehensive understanding of the requirements and the intentions of users is ensured. And analyzing the complex sentence structure, and identifying key information such as actions, objects, attributes and the like in the complex sentence structure, so that the intention of the user is accurately extracted, and detailed user intention information including operation targets, operation objects and operation attributes is generated. And an accurate user requirement and intention analysis result is provided for the behavior pattern recognition and subsequent security verification of the robot.

In the behavior pattern recognition module, user intention information and historical interaction data of a user are analyzed by using a hidden Markov model. And establishing a probability model of the user behavior by analyzing and calculating the transition probability between the states, so that the behavior habit and mode of the user are effectively identified. By comparing the current behavior with the historically established patterns, the hidden Markov model is able to identify abnormal behavior patterns. Not only is the user behavior habit deeply understood, but also any abnormal behavior can be found in time, and a behavior pattern analysis result comprising a conventional behavior pattern, an abnormal behavior index and a behavior trend analysis is generated, so that the result is important for improving the safety of the system.

In the security enhancement module, voice identity verification is performed based on the behavioral pattern analysis result by combining a voiceprint recognition technology and a deep neural network. The deep neural network is matched with the existing voiceprint model by analyzing the characteristics of the user voice, so that the user identity is effectively verified. And analyzing whether the behavior of the user is consistent with the known security mode, evaluating the potential security threat by using a deep learning technology, and dynamically adjusting the security measures according to the risk level. The system can enhance the security on the two levels of identity verification and behavior analysis, and the generated security verification result comprises the user identity verification state, behavior compliance assessment and security risk level, so that comprehensive security guarantee is provided for the system.

In the task priority adjustment module, the current system resource status and the task urgency are analyzed based on the security verification result by applying a genetic algorithm. The genetic algorithm performs priority ordering on the task list to be executed by simulating the processes of natural selection and genetic mechanism, considers the urgency degree of the task, the resource demand and the expected completion time, and realizes dynamic optimization of the resource allocation efficiency and the timeliness of task completion. The task execution plan generated details the task sequence, the priority of each task, and the predicted completion time, ensuring that the system is able to respond to user demands in the most efficient manner.

In the abnormal state identification module, real-time monitoring and analysis are performed based on a task execution plan by combining data input of multiple sensors of the robot and utilizing an isolated forest algorithm. The isolation forest algorithm isolates abnormal points by constructing a decision tree, and effectively identifies abnormal states which deviate significantly from a normal behavior mode. The method not only comprises the step of monitoring the internal state of the robot, but also comprehensively evaluates the external environment factors, and the generated abnormal state results comprise abnormal types, abnormal degrees and recommended response measures, so that important support is provided for the stable operation of the system.

In the decision support module, the action scheme is subjected to deep risk and benefit analysis based on abnormal state results by using a decision tree and a Bayesian network. Decision trees provide an intuitive way to evaluate the consequences of different mobility schemes, while bayesian networks perform well in handling uncertainty information, allowing the system to make more accurate predictions in the case of incomplete information. By comprehensively considering potential consequences of a plurality of action schemes, the most suitable action strategy is selected, and the generated decision support scheme comprises recommended action schemes, potential risk analysis and expected effect evaluation, so that a scientific decision basis is provided for the robot.

In the response execution module, a decision support scheme is based by employing a state machine model and a rules engine. The state machine model determines the current operating state and the expected state transitions of the robot, and the rules engine matches the corresponding execution rules according to the operating state, the safety alert requirements, and the fault recovery requirements. Not only ensures that the robot can flexibly adjust the behavior according to the current state and environmental conditions, but also ensures that response measures can be quickly taken when safety threat or fault is encountered. The generated decision execution scheme records the execution state and the execution result evaluation in detail, and provides powerful execution support for autonomous operation of the robot.

Referring to fig. 2 and 3, the voice recognition module includes a feature extraction sub-module, a timing analysis sub-module, and a text conversion sub-module;

The feature extraction submodule is based on voice input of a user, the submodule analyzes voice signals by adopting a deep convolutional neural network algorithm, and extracts time-frequency features from original audio signals by constructing a plurality of convolution layers and pooling layers to generate time-frequency feature data;

The time sequence analysis sub-module is used for performing time sequence analysis on the extracted features by utilizing a long-period memory network algorithm based on time-frequency feature data, processing and memorizing long-term dependent information through a gate control mechanism, capturing the structure and the context of a voice instruction and generating a structured voice feature;

The text conversion submodule converts the structured voice features into corresponding text instructions by adopting a sequence-to-sequence learning model based on the structured voice features and through the sequence-to-sequence learning model with the attention mechanism, and generates voice text information.

In the feature extraction sub-module, the voice input of the user is analyzed through a deep convolutional neural network algorithm. The algorithm can automatically learn useful features from the data, and is suitable for processing complex voice signals. Deep convolutional neural networks perform a series of conversions and downsampling on an input raw audio signal by constructing convolutional and pooling layers over multiple layers. Each layer of convolution layer is responsible for extracting features of different levels, from simple audio waveforms to more complex time-frequency features such as pitch and tempo changes, while the pooling layer is used to reduce the dimensionality of the features and enhance the generalization ability of the model. The model can extract time-frequency characteristic data reflecting voice content and characteristics from an original audio signal, the data is converted into a format which is more useful for a subsequent module, and the generated time-frequency characteristic data is key information in the voice signal and provides a basis for realizing accurate voice recognition.

In the time sequence analysis sub-module, time sequence analysis is carried out on the time-frequency characteristic data obtained from the characteristic extraction sub-module through a long-period and short-period memory network algorithm. The long-term and short-term memory network is a special cyclic neural network, and the problem that the traditional cyclic neural network is difficult to process long-term dependence is solved by introducing a gate control mechanism. When processing voice signals, the long-term memory network can effectively memorize and process information which is far apart in time according to the time sequence characteristics of the voice signals, thereby capturing the structure and the context in voice instructions. The input gate, the forget gate and the output gate in the network work cooperatively to decide the storage, updating and output of the information at different time points. The long-term and short-term memory network can extract the features reflecting the time sequence structure of the voice command from the time-frequency feature data, and the generated structured voice features contain rich time sequence information, so that key data is provided for accurately converting the voice command into the text.

In the text conversion sub-module, the structured speech features generated by the time sequence analysis sub-module are converted into corresponding text instructions by a sequence-to-sequence learning model, in particular a sequence-to-sequence learning model that introduces a mechanism of attention. The sequence-to-sequence learning model consists of an encoder that is responsible for encoding the input structured speech features into a fixed-length vector and a decoder that generates a text sequence based on the vector. The introduced attention mechanism enables the model to focus more on the input part most relevant to the current output when generating text, thereby improving the accuracy of the conversion. The model not only can accurately convert the voice command into a text format, but also can reserve important information in voice data, such as mood and emphasis, and the generated voice text information provides an accurate data basis for subsequent user intention analysis, behavior pattern recognition and the like.

Referring to fig. 2 and 4, the user intention analysis module includes a semantic analysis sub-module, an intention recognition sub-module, and a requirement mapping sub-module;

the semantic analysis submodule adopts a bidirectional encoder to express a slave converter model based on the voice text information, carries out semantic analysis on the text content, learns the semantic information through a pre-trained text corpus, analyzes the meaning and the context in the text and generates a semantic analysis result;

The intention recognition sub-module is used for analyzing and recognizing the demands of the user by applying an intention recognition algorithm in a natural language processing technology based on the semantic analysis result, recognizing the intention of the user by analyzing verbs, nouns and phrases in the semantic analysis result and generating intention information;

The demand mapping sub-module maps the intention of the user to an execution operation by using a logical reasoning algorithm based on the intention information, and generates the intention information of the user by analyzing the relationship between the intention of the user and the robot service and the function and matching the service and the operation.

In the semantic parsing sub-module, the phonetic text information is parsed semantically from the transformer model by bi-directional encoder representation. Text content is analyzed deep from the transducer model using the bi-directional encoder representation by taking as input text information converted from speech. The bi-directional encoder represents the slave-transducer model, which is a pre-trained deep learning model, by learning language rules in a large text corpus, the word sense, grammar structure and context in the text can be understood. The bi-directional encoder encodes each term from the transformer, captures its relationship to the term in the context, and extracts the deep semantic information of the text. The generated semantic analysis results contain contextually relevant semantic representations of each term in the text, providing an accurate semantic basis for subsequent intent recognition.

In the intent recognition sub-module, the semantic analysis results are analyzed and recognized by applying an intent recognition algorithm in natural language processing technology. The algorithm is focused on analyzing key words in semantic analysis results, such as verbs, nouns, phrases and the like, and comprehensively considering the meaning of the words in a specific context so as to identify specific requirements and intentions of users. The intention recognition process based on deep learning can accurately extract the intention of the user from complex semantic information, the generated intention information clearly indicates the operation target, the operation object and the operation attribute which the user wants to execute, and clear instructions are provided for the robot to execute accurate operations.

In the demand mapping sub-module, the user's intention is mapped to a specific execution operation by applying a logical reasoning algorithm. The logical reasoning algorithm analyzes the logical relationship between the user intention information and the services and functions which can be provided by the robot, and identifies the service or operation which is most suitable for the user requirement. Through the logic matching process, the algorithm can establish an accurate corresponding relation between the user intention and the robot capability, so that the robot can respond to the instruction of the user properly. The generated user intention information not only contains the original requirements of the user, but also indicates specific services and operations required by the robot to meet the requirements, and provides an accurate operation guide for autonomous execution of the robot.

Referring to fig. 2 and 5, the behavior pattern recognition module includes a data analysis sub-module, a pattern establishment sub-module, and an anomaly detection sub-module;

The data analysis submodule analyzes the historical interaction data of the user based on the user intention information by adopting a statistical analysis method, identifies the frequency, time distribution and preference mode of the user behavior and generates a statistical analysis result of the user behavior;

The abnormality detection submodule identifies deviation between current behavior and historical behavior patterns by using an abnormality detection algorithm based on user behavior pattern information, quantifies the deviation degree of the behavior patterns, carries out abnormality behavior identification and generates a behavior pattern analysis result.

In the data analysis sub-module, historical interaction data of the user is deeply analyzed through a statistical analysis method. And through data arrangement and cleaning, the accuracy of analysis is ensured. And quantitatively analyzing the user behavior frequency, time distribution and preference modes in the data set by using a statistical principle. And analyzing according to the interaction records of the user and the system, such as command input frequency, using time points, preference functions and the like, and adopting statistical methods such as frequency distribution analysis, time sequence analysis and the like to identify the normal state and the characteristics of the user behavior. And generating a user behavior statistical analysis result, wherein the result details the interactive habit of the user, and provides basic data and insight for establishing an accurate user behavior model.

In the pattern building sub-module, the sequence of user actions and transition probabilities are modeled by a hidden Markov model. And using a user behavior statistical analysis result as input, and exploring a transition rule and a potential state between user behaviors by using a hidden Markov model. A hidden markov model is a statistical model that is inferred by assuming that the state of the system is not directly visible, but by the sequence of events observed. The behavior sequence of the user is regarded as an observation sequence, and the behavior habits and patterns of the user are regarded as potential states of the system. The model can reveal potential rules of user behaviors by calculating transition probabilities among different behavior states, and finally generates information containing user behavior habits and modes, and the information is not only helpful for understanding the behavior motivation of the user, but also provides accurate reference standards for subsequent abnormal behavior detection.

In the abnormality detection sub-module, user behavior pattern information is analyzed through an abnormality detection algorithm to identify a current behavior having a significant deviation from a historical behavior pattern. And according to the user behavior pattern information, adopting algorithms such as distance-based anomaly detection, density-based anomaly detection and the like to evaluate the current behavior data of the user, and identifying the anomaly points of the behavior pattern. By comparing the deviation degree of the current behavior and the normal behavior pattern defined in the model, the abnormal index of the behavior pattern is quantified, so that the abnormal behavior is effectively identified and marked. The generated behavior pattern analysis result records the identified abnormal behavior and the deviation degree thereof in detail, provides important basis for safety monitoring and user behavior management of the system, and effectively enhances the response capability and preventive measures of the system to the abnormal behavior.

Referring to fig. 2 and 6, the security enhancement module includes an identity verification sub-module, a behavior compliance sub-module, and a risk rating sub-module;

the identity verification submodule is used for carrying out identity verification on the voice of the user by adopting a deep neural network based on the behavior pattern analysis result and combining with the voiceprint recognition technology, analyzing the characteristics of the voice, including tone, rhythm and voice texture, and carrying out voiceprint model matching to generate a voice identity verification result;

In the identity verification sub-module, the voice of the user is verified by combining a voiceprint recognition technology and a deep neural network algorithm. The method comprises the steps of collecting a sound sample of a user, and extracting characteristics in the sound sample, including sound characteristics such as tone, rhythm and sound texture, by using a deep neural network, particularly a convolutional neural network and a cyclic neural network. The deep learning model establishes a voiceprint model by learning unique patterns in the voice sample, and the voice features of each user are converted into a set of digital representations to form a voiceprint feature library. In the verification stage, the voice sample of the user is processed similarly and matched with the model in the feature library, and identity is verified by calculating a similarity score. The generated voice identity verification result not only contains the information of whether the user is authenticated by the system, but also contains the confidence score of the authentication, thereby providing an efficient and safe user identity verification means for the system.

In the behavior compliance submodule, whether the current behavior of the user is consistent with the recorded safety behavior mode is analyzed through a behavior pattern matching algorithm. Based on the voice identity verification result, the established user behavior pattern library is utilized to monitor and analyze the current behavior of the user in real time. By comparing the current behavior to the compliance behavior patterns stored in the library, the algorithm is able to identify behaviors that deviate from the normal patterns. The analysis process considers multidimensional information such as the types of behaviors, the execution time, the execution frequency and the like, and ensures the comprehensiveness and the accuracy of analysis results. The generated behavior compliance analysis results detail the compliance rating of each behavior, helping the system to determine if there is a potential risk or violation of the user's operation, thereby preventing possible security problems at an early stage.

In the risk rating sub-module, potential security threats are evaluated and ranked by a risk evaluation model. Based on the behavior compliance analysis result, the factors such as the degree of behavior deviation, the correlation with the historical security event, the security condition of the current operation environment and the like are comprehensively analyzed by adopting a quantification method and a risk assessment standard. Through comprehensive assessment of dimensions, the model assigns a risk level to each detected potential threat, thereby determining the urgency and strength of the security measures to be taken. The generated security verification result not only defines the risk level of each potential threat, but also provides the basis for adjusting the security policy for different risk levels for the system, so that the system can dynamically manage and cope with the security threats, and the security of the system and the user is ensured.

Referring to fig. 2 and fig. 7, the task priority adjustment module includes an emergency analysis sub-module, a resource matching sub-module, and a scheduling policy sub-module;

The emergency degree analysis submodule evaluates the emergency degree of a plurality of tasks according to the safety verification result by adopting a multi-factor decision analysis method, analyzes the deadline, the relevance and the demand factors of the tasks on resources, distributes the emergency degree score of each task and generates task emergency degree scoring information;

And in the emergency degree analysis submodule, the emergency degree of each task is comprehensively evaluated through a multi-factor decision analysis method. By collecting various information related to tasks, including factors such as the deadlines of the tasks, the relatedness between the tasks, and the specific demands on resources. And comprehensively considering a plurality of evaluation indexes by utilizing a multi-factor decision analysis method, and distributing an urgency score for each task by constructing a scoring model. The scoring mechanism can accurately reflect the priority of the task, and ensure that the resource is preferentially allocated to the most urgent task. And generating task emergency degree scoring information, providing quantitative basis for subsequent resource matching and task scheduling, and ensuring that the system can efficiently respond to each emergency task.

In the resource matching sub-module, the current system resource is accurately analyzed and matched through a linear programming algorithm. And comprehensively combing available resources in the system based on the task urgency scoring information, and optimizing a resource allocation scheme by using a linear programming algorithm. By establishing a mathematical model, an optimal resource allocation solution is found to meet the resource requirements of all tasks while maximizing the resource utilization efficiency. Taking into account the resource requirements and urgency scores of each task, it is ensured that the tasks with high priority can obtain the required resources. The generated resource matching scheme lists the priorities of each task for obtaining the resources in detail, and provides an explicit resource allocation guide for task execution of the system.

And in the scheduling policy sub-module, the task list to be executed is subjected to priority ordering and optimization of the scheduling policy through a genetic algorithm. Based on the resource matching scheme, a genetic algorithm is used for simulating the natural selection and genetic mechanism process, and the task execution sequence is optimized. The genetic algorithm continuously and iteratively searches an optimal task scheduling scheme through the steps of population initialization, selection, crossing, mutation and the like. The algorithm considers multi-dimensional factors such as the urgency of the task, the resource allocation condition, the execution time and the like, and reserves the scheduling scheme with the best effect through a natural selection mechanism. The generated task execution plan plans the execution sequence and time schedule of each task in detail, and ensures that the system realizes the most efficient task execution under limited resources.

Referring to fig. 2 and 8, the abnormal state recognition module includes a state monitoring sub-module, an abnormal analysis sub-module, and a result generation sub-module;

The state monitoring submodule is based on a task execution plan, combines the data input of multiple sensors of the robot, adopts a time sequence analysis method to monitor the working state and the external environment of the robot in real time, and generates a state monitoring result;

The abnormal analysis submodule applies an isolation forest algorithm based on the state monitoring result to detect abnormal points of the monitoring data, identifies the isolation degree of data points by constructing a plurality of isolation trees, identifies an abnormal state deviating from a normal behavior mode and generates an abnormal state analysis result;

The result generation sub-module adopts report generation technology to evaluate the abnormal state based on the abnormal state analysis result, comprises the abnormal type, degree and influence, and provides an abnormal state description and response scheme by integrating the analysis result and the context information to generate an abnormal state result.

In the state monitoring submodule, the working state and the external environment of the robot are monitored in real time by combining a time sequence analysis method with multi-sensor data input of the robot. The sensor data is sorted in time order to form time series data. Time series analysis performs in-depth analysis by observing the rules and patterns of data over time, such as periodicity, trending, etc. The method can effectively monitor the tiny change of the robot working state and the influence of the external environment, and timely capture potential problems and anomalies. The generated state monitoring result comprises a time sequence analysis report of each sensor data of the robot, and an important data basis is provided for the recognition of abnormal states.

In the abnormality analysis sub-module, abnormal point detection is carried out on the monitoring data in the state monitoring result by applying an isolation forest algorithm. The isolation forest algorithm identifies the degree of isolation of data points by building multiple isolation trees, with outliers typically being more easily isolated in the isolation tree and therefore requiring fewer splits to be identified. Is suitable for processing high-dimensional data and large-scale data sets, and efficiently identifies abnormal states deviating from normal behavior patterns. The generated abnormal state analysis result describes the detected abnormal data points and the isolation degree thereof in detail, and provides accurate basis for subsequent abnormal response and processing.

In the result generation sub-module, the analysis result of the abnormal state is comprehensively evaluated through report generation technology, wherein the analysis result comprises the type, degree and possible influence of the abnormal state. And integrating the analysis result and the related context information by using a report generation technology to form an easy-to-understand abnormal state report. The report not only provides detailed description of the abnormal state, but also comprises response schemes and suggestions aiming at different abnormal types, provides real-time and useful information for robot operators or maintenance personnel, helps the robot operators or maintenance personnel to quickly and accurately respond to the abnormal state, and ensures the stable operation and safety of the robot system.

Referring to fig. 2 and 9, the decision support module includes a risk analysis sub-module, a scheme optimization sub-module, and a policy making sub-module;

The strategy making sub-module adopts a rule engine to combine the operation capability and the environmental condition of the current robot to make an action strategy based on the optimized action scheme, and makes an execution step and an expected target by analyzing the content and the implementation condition of the scheme to generate a decision support scheme.

In the risk analysis submodule, the abnormal state result is deeply analyzed through a decision tree algorithm and a Bayesian network, and risk scores and income expectations are given to each action scheme. The decision tree algorithm systematically evaluates the potential risk and benefit of each action scheme by constructing differentiated decision paths, so that the decision process is more visual and interpretable. Each decision node represents an action selection and each path represents a series of decision results, revealing a variety of results resulting from different action schemes. Bayesian networks are used to probabilistic analyze decisions containing uncertainty factors, taking into account the effect of uncertainty of various internal and external factors on decision results. The method can provide a probability-based risk assessment for each action scheme under the condition of considering uncertainty, and the generated action scheme risk assessment result details the risk score and the income expectation of each scheme, so that a scientific basis is provided for subsequent scheme selection and optimization.

In the scheme optimization sub-module, the action scheme is optimized through a multi-objective optimization algorithm. Using the pareto front analysis method, a combination of action schemes that achieves the best balance among the multiple targets is identified. The multi-objective optimization algorithm searches for a scheme that performs best on multiple evaluation indexes by considering risk controllability and expected benefits of the scheme, namely, a scheme that can minimize risks and maximize benefits. The pareto front analysis can identify all the optimal solution sets, and provides a series of feasible optimal scheme choices for decision makers. The generated optimized post-action schemes detail the comprehensive evaluation results of each scheme, and provide a feasible scheme for realizing the optimal balance between risks and benefits.

In the policy making sub-module, action policies are formulated by the rules engine in combination with the current robot's operational capabilities and environmental conditions. The rule engine analyzes the optimized action schemes, and combines the actual operation capability of the robot and the current working environment condition to formulate specific execution steps and expected targets for each action scheme. The rule engine automatically generates the most suitable action strategy according to the content and implementation conditions of the scheme by defining a series of logic rules, so that the implementation of the strategy is ensured to be in line with the capability range of the robot and the current operation environment. The generated decision support scheme provides detailed guidance for the operation of the robot, including specific execution steps of each action, expected achievement targets and faced challenges and countermeasures, and ensures that the robot can efficiently and safely complete tasks.

Referring to fig. 2 and 10, the response execution module includes an operation implementation sub-module, a security measure sub-module, and a feedback arrangement sub-module;

the operation implementation submodule is used for analyzing the current operation state and the required conversion of the robot based on a decision support scheme, adjusting the behavior mode of the robot according to the guidance of the scheme, executing behavior adjustment and task starting operation, and generating an operation adjustment result;

The safety measure sub-module adopts a dynamic risk management strategy based on an operation adjustment result, dynamically adjusts safety measures according to action risk levels and current environmental conditions, and generates a safety adjustment scheme by monitoring the operation environment and the state of the robot in real time, sending out a safety alarm and activating a fault recovery program;

The feedback arrangement sub-module is based on a safety adjustment scheme, adopts a feedback analysis technology, collects operation implementation and safety adjustment results, evaluates and arranges the execution process and the results, and generates a decision execution scheme by analyzing the success rate of the operation and the processing efficiency index of the safety event and formulating an improvement opinion and a follow-up scheme.

In the operations implementation sub-module, the decision support scheme is implemented by applying a state machine model. The state machine model is used to analyze and track the current operating state of the robot and state transitions required according to a decision support scheme. This model allows the system to change state in an orderly and predictable manner, ensuring that adjustments in robot behavior and task initiation operations meet the requirements of a predetermined scheme. Including adjusting the behavior pattern of the robot, such as transitioning from a standby state to a state in which a particular task is performed, according to the protocol guidelines. The method ensures that the behavior of the robot is consistent with the decision support scheme, and the generated operation adjustment result records the change of the state of the robot and the detailed condition of behavior adjustment, thereby providing direct evidence and feedback for executing the decision support scheme for the system.

In the security measure sub-module, the security measures are dynamically adjusted by adopting a dynamic risk management strategy. Based on the operation adjustment results, the policy adjusts security measures in real time, such as issuing a security alarm when high risk operation is detected or activating a recovery procedure when a failure occurs, taking into account the action risk level and the current environmental conditions. The dynamic risk management strategy can rapidly respond to the changed environment and state by monitoring the operation environment and the state of the robot in real time, and timely take necessary safety measures, and the generated safety adjustment scheme lists the safety measures and adjustment basis in detail, so that the safety and reliability of the operation of the robot are enhanced.

In the feedback arrangement sub-module, the results of operation implementation and safety adjustment are comprehensively evaluated and arranged by adopting a feedback analysis technology. And collecting the results of operation implementation and the effect of safety adjustment, carrying out deep analysis on the execution process and the results, and evaluating the success rate of the operation and the processing efficiency of the safety event. The feedback analysis technology provides basis for continuous improvement of the system by analyzing the collected data and identifying successful elements and existing problems in the operation. The generated decision execution scheme summarizes the evaluation result and the improvement opinion of the execution process, provides precious experience and guidance of follow-up actions for future operation, ensures that the system can be continuously optimized, and improves the operation efficiency and the safety management level.

The present invention is not limited to the above embodiments, and any equivalent embodiments which can be changed or modified by the technical disclosure described above can be applied to other fields, but any simple modification, equivalent changes and modification made to the above embodiments according to the technical matter of the present invention will still fall within the scope of the technical disclosure.

Claims

1. Bionic robot in the aspect of speech recognition control, its characterized in that: the bionic robot in the aspect of voice recognition control comprises a voice recognition module, a user intention analysis module, a behavior mode recognition module, a safety enhancement module, a task priority adjustment module, an abnormal state recognition module, a decision support module and a response execution module;

2. The biomimetic robot in terms of speech recognition control of claim 1, wherein: the voice text information comprises transcribed content, confidence scores and time stamps, the user intention information comprises operation targets, operation objects and operation attributes, the behavior pattern analysis results comprise conventional behavior patterns, abnormal behavior indexes and behavior trend analysis, the security verification results comprise user identity verification states, behavior compliance assessment and security risk levels, the task execution plan comprises task sequences, task priorities and expected completion time, the abnormal state results comprise abnormal types, abnormal degrees and recommended response measures, the decision support schemes comprise recommended action schemes, potential risk analysis and expected effect assessment, and the decision execution schemes comprise execution states and execution result assessment.

3. The biomimetic robot in terms of speech recognition control of claim 1, wherein: the voice recognition module comprises a feature extraction sub-module, a time sequence analysis sub-module and a text conversion sub-module;

4. The biomimetic robot in terms of speech recognition control of claim 1, wherein: the user intention analysis module comprises a semantic analysis sub-module, an intention recognition sub-module and a requirement mapping sub-module;

5. The biomimetic robot in terms of speech recognition control of claim 1, wherein: the behavior pattern recognition module comprises a data analysis sub-module, a pattern establishment sub-module and an abnormality detection sub-module;

6. The biomimetic robot in terms of speech recognition control of claim 1, wherein: the security enhancement module comprises an identity verification sub-module, a behavior compliance sub-module and a risk rating sub-module;

7. The biomimetic robot in terms of speech recognition control of claim 1, wherein: the task priority adjustment module comprises an emergency analysis sub-module, a resource matching sub-module and a scheduling strategy sub-module;

8. The biomimetic robot in terms of speech recognition control of claim 1, wherein: the abnormal state identification module comprises a state monitoring sub-module, an abnormal analysis sub-module and a result generation sub-module;

9. The biomimetic robot in terms of speech recognition control of claim 1, wherein: the decision support module comprises a risk analysis sub-module, a scheme optimization sub-module and a strategy making sub-module;

10. The biomimetic robot in terms of speech recognition control of claim 1, wherein: the response execution module comprises an operation implementation sub-module, a safety measure sub-module and a feedback arrangement sub-module;