A method for processing abnormal data applied to robots
Through deep learning models, the severity of abnormal data of robots is evaluated and processed, and the problem that traditional methods cannot identify and process abnormal data is solved, which improves the stability and security of robot operations, and enhances the adaptability and long-term effectiveness of the model. .
Patent Information
- Application Number
- CN202410912986.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-09
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-07-09
AI Technical Summary
Traditional data processing methods cannot effectively identify and process robot abnormal data, resulting in operational instability and security issues, lack of a standardized abnormal data severity assessment system, and cannot give priority to the processing of the most serious abnormal data.
Using deep learning architecture to build models, establish classification rules through real-time sensor data preprocessing and annotation, evaluate the severity of abnormal data, and use multiple influencing parameters (functional impact, security, economic impact, system stability, timeliness, Predictability) calculates the final decision value, prioritizes the processing of serious abnormal data, and improves generalization ability through dynamic update of the model.
It realizes standardized evaluation and intelligent processing of the severity of abnormal data, ensuring that the robot can promptly prioritize the handling of serious abnormalities when encountering abnormal situations, improving the stability and safety of the robot's operation, enhancing the adaptability and long-term effectiveness of the model sex.
Smart Images

Figure CN118467930B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to an abnormal data processing method applied to a robot. Background Art
[0002] With the widespread application of robotics in industrial automation, service industry and other fields, the stability and reliability of robots in performing tasks have become particularly important. However, robots may encounter various abnormal situations in actual work, such as sensor misreading, mechanical failure, environmental changes, etc. If these abnormalities are not handled in time, they may affect the performance of the robot and even lead to mission failure or safety accidents. Therefore, effective processing of abnormal data of robots is the key to ensuring their normal operation;
[0003] Traditional data processing methods may not be able to effectively identify and process these abnormal data, resulting in instability or safety issues in robot operation. There is a lack of a standardized evaluation system for the severity of abnormal data, which makes it impossible to prioritize the most serious abnormal data. At the same time, the exception handling strategy is not intelligent and adaptive enough, and cannot take the most appropriate response measures according to the specific circumstances of the exception. Therefore, an abnormal data processing method for robots is designed.
[0004] In order to solve the above defects, a technical solution is now provided. Summary of the invention
[0005] The purpose of the present invention is to solve the problem that traditional data processing methods lack a standardized evaluation system for the severity of abnormal data, resulting in the inability to prioritize the most serious abnormal data, and to propose an abnormal data processing method applied to robots.
[0006] The purpose of the present invention can be achieved through the following technical solutions:
[0007] An abnormal data processing method applied to a robot comprises the following steps:
[0008] S1: First, the collected real-time sensor data of the robot is preprocessed, and labels or annotations are added to the data to form a labeled data set and store it in the database;
[0009] S2: Then, we label the dataset, determine the deep learning architecture, build and initialize the model parameters, select the loss function and optimization algorithm, perform the model training cycle, and monitor the training process;
[0010] S3: The real-time collected data is passed to the trained deep learning model, which extracts and transforms features, quickly calculates and infers, determines data anomalies, and compares them with the data in the database.
[0011] S4: Extract key features based on the abnormal data monitored by the model, establish classification rules, analyze the type and severity of abnormal data, and evaluate the impact of abnormal data on different dimensions of the robot. The analysis steps for the severity of abnormal data are as follows:
[0012] Obtain the influencing parameters of the severity of the impact of abnormal data on the robot. The influencing parameters include the degree of impact on the robot function, safety-related factors, economic factors, system stability, timeliness factors and predictability. The power loss value, safety assessment value, meridian shadow value, system shadow value, delay value and prediction value are obtained through analysis of the influencing parameters. After normalization, the power loss value, safety assessment value, meridian shadow value and system shadow value are used as the basic severity assessment standards for abnormal data, and are substituted into the following formula: To get the basic evaluation value YPJ, where They are power loss value, safety assessment value, shadow value and system shadow value. They are the preset weight coefficients of power loss value, safety assessment value, shadow value and system shadow value;
[0013] Then substitute the hysteresis value and the prediction value into the following formula as the correction factors of the basic evaluation value YPJ: To get the final judgment value ZPJ, where YC and YP are the hysteresis value and the prediction value respectively. are the preset weight coefficients of the hysteresis value and the prediction value respectively; and the final judgment value ZPJ is used as the final standard for measuring the severity of abnormal data;
[0014] Then, the final judgment values calculated from different abnormal data are sorted by size, and the abnormal data with large final judgment values are processed first, and the final judgment value of the abnormal data is compared with the three preset final judgment value intervals. The three final judgment value intervals correspond to the severity levels of the abnormal data, namely mild abnormality, moderate abnormality and severe abnormality;
[0015] S5: Adopt different processing strategies according to the severity of abnormal data;
[0016] S6: Regularly collect new data to update deep learning models, introduce new anomaly types and scenarios to improve model generalization capabilities, continuously train models and perform performance verification.
[0017] Furthermore, the analysis process of the power loss value, safety assessment value, longitude and latitude value in S4 is as follows:
[0018] Impact on robot functions: The percentage of task completion rate reduction caused by the abnormal data, the number of failures of key functions, and the time required for function recovery are collected. After normalization, the percentage of task completion rate reduction and the number of failures of key functions are multiplied and divided by the time required for function recovery to obtain the power loss value. This power loss value is used as a standard to measure the impact of the abnormal data on the robot function.
[0019] Safety-related factors: The risk level of abnormal data to personnel safety is analyzed and a risk score of 1-100 is assigned; the potential degree of equipment damage is converted into a damage percentage; the degree of deviation from safety specifications is quantified by statistically analyzing the similarity with safety specifications. The higher the similarity, the lower the degree of deviation from safety specifications.
[0020] The obtained risk score, damage percentage and similarity are normalized and substituted into the following formulas: The safety assessment value APZ is obtained, and this safety assessment value is used to measure the impact of abnormal data on safety-related factors; where fp, ss and xs are risk score, damage percentage and similarity respectively;
[0021] Economic impact factors: By analyzing the production delays caused by abnormal data, using the stop time as a measurement standard, and then collecting the amount of property losses caused by abnormal data and the negative impact on cost-effectiveness, the impact is assigned an impact score of 1-10. The larger the impact score, the greater the negative impact on cost-effectiveness;
[0022] After normalizing the obtained stop time and impact score, the cylinder model is established with the stop time as the bottom circle area and the impact score as the height. The volume of the cylinder model is calculated and recorded as the shadow value, which is used as the standard for measuring the economic impact of abnormal data.
[0023] System stability: By counting the frequency of system crashes or errors caused by the abnormal data and the degree of impact on the overall system performance, and quantifying the degree of impact through the percentage of performance degradation, the frequency of system crashes or errors and the percentage of performance degradation are normalized, and a base circle is established with the system crash or error frequency multiplied by the frequency weight coefficient as the radius, and a cone model is established with the performance degradation percentage multiplied by the performance weight coefficient as the height. The surface area of the cone model is calculated and recorded as the shadow value, and this shadow value is used as the standard for measuring the impact of abnormal data on system stability.
[0024] Furthermore, the analysis process of the hysteresis value and the prediction value in S4 is as follows:
[0025] Timeliness factor: Count the length of delay or lag caused by the abnormal data and the degree of impact on real-time response, and quantify the degree of impact on real-time response through impact time. Then, normalize and sum the delay or lag time and impact time, and use the preset angle to establish two sides of a triangle. Supplement the other side of the triangle to form a complete triangle. Calculate the area of the triangle and record it as the hysteresis value. This hysteresis value is used as a standard to measure the impact of abnormal data on timeliness.
[0026] Predictability: Analyze the predictability difficulty of the abnormal data and assign a difficulty score of 1-10. The higher the difficulty score, the greater the predictability difficulty. Then, count the frequency of occurrence of the abnormal data from historical data. Normalize the difficulty score and the frequency of occurrence of abnormal data, multiply them, and add a correction factor to obtain the predicted value, which is used as the standard for measuring the predictability of abnormal data.
[0027] Furthermore, the specific operation steps of S5 are as follows:
[0028] When the abnormal data is judged to be a mild abnormality, the abnormal data is recorded and automatic repair is attempted, including adjusting the robot's behavior or re-collecting data;
[0029] When the abnormal data is judged to be moderately abnormal, the functions related to the abnormal data are restricted;
[0030] When the abnormal data is judged to be a serious abnormality, the system is restarted or shut down urgently, and an early warning signal is issued in time to request manual intervention.
[0031] Furthermore, the specific operation steps of S1 are as follows:
[0032] First, data from different sensors equipped in the robot are collected to collect multi-dimensional data;
[0033] Then, the collected real-time sensor data is pre-processed and labeled;
[0034] Preprocessing includes filtering the data, removing possible noise, and formatting the data;
[0035] Labeling adds labels or annotations to the data to facilitate the learning and understanding of the model. The annotation content includes the classification of normal data and abnormal data.
[0036] After completing the preliminary preprocessing and labeling of the data, a labeled data set is formed and stored in the database.
[0037] Furthermore, the specific operation steps of S2 are as follows:
[0038] Use a large-scale labeled data set in the database, including normal data and various types of abnormal data, determine the deep learning architecture according to the characteristics of the problem, and use the selected deep learning architecture to build a deep learning model;
[0039] After the model is determined, the model parameters are initialized, and the loss function that matches the task is determined to measure the model prediction error, and then the optimization algorithm is determined, including stochastic gradient descent or its variants;
[0040] Then the model is trained in a loop. The specific process is as follows:
[0041] Batch loading of training data;
[0042] Calculate the output of the model;
[0043] Calculate the error based on the loss function;
[0044] Update model parameters through optimization algorithms;
[0045] Use matching metrics to monitor the training process, including accuracy and loss values;
[0046] Adjust model parameters and training strategies based on monitoring results to optimize model performance;
[0047] During training, model performance is regularly evaluated on the validation set.
[0048] Based on the performance of the validation set, determine whether to stop training early to prevent overfitting;
[0049] Compare the performance of different models or different parameter settings and select the optimal model;
[0050] Save the final trained model.
[0051] Furthermore, the specific operation steps of S3 are as follows:
[0052] The real-time collected data is passed into the trained deep learning model;
[0053] The model extracts and transforms the features of the input data, and uses the model’s algorithms and parameters to perform rapid calculations and inferences;
[0054] Judge the anomalies of the data and determine whether there are anomalies by comparing it with the normal data pattern and according to the judgment criteria of the model;
[0055] The results of abnormal judgment are output, and the real-time acquired data is compared with the database. When there is new training data, the model is automatically updated according to the new training data to improve the accuracy of model detection.
[0056] Furthermore, the specific operation steps of S6 are as follows:
[0057] First, determine the data source, set the collection frequency, and regularly collect new data to update the deep learning model;
[0058] Introducing new exception types and scenarios;
[0059] Continue to train the updated model using new data and new anomaly types, and validate the model’s performance and generalization capabilities;
[0060] Then, according to the actual operation of the robot, the parameters and weights of the model are dynamically adjusted;
[0061] The updated model is then deployed to the robot system, the robot’s performance in actual operation is observed, feedback information is collected, and the model and update process are improved based on the feedback.
[0062] Compared with the prior art, the present invention has the following beneficial effects:
[0063] The present invention, by introducing influencing parameters of multiple dimensions, can standardize the evaluation of the severity of abnormal data, not only considering the impact of the abnormality on the robot function, but also including multiple aspects such as safety, economy, system stability, timeliness and predictability, and compare the severity of multiple abnormal data, giving priority to abnormal data with greater severity, making the abnormality processing more comprehensive and systematic;
[0064] The present invention also includes a mechanism for dynamically updating the model. By regularly collecting new data and introducing new types of anomalies, the deep learning model is continuously trained and optimized, so that the model can adapt to the ever-changing operating environment and new abnormal situations, thereby improving the generalization ability and long-term effectiveness of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings;
[0066] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION
[0067] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0068] It should be understood that the terms "include" and "comprising" used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.
[0069] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and claims, the singular forms of "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.
[0070] like Figure 1 As shown, a method for processing abnormal data applied to a robot comprises the following steps:
[0071] Multimodal data acquisition:
[0072] Data from different sensors equipped in the robot are collected. The sensors equipped in the robot include: visual sensors, sound sensors, temperature sensors, etc., to collect multi-dimensional data; then the collected real-time sensor data is preliminarily preprocessed and labeled; preprocessing includes filtering the data, removing possible noise, and formatting the data; labeling adds labels or annotations to the data to facilitate model learning and understanding, and the labeled content includes the classification of normal data and abnormal data; after completing the preliminary preprocessing and labeling of the data, a labeled data set is formed and stored in the database for subsequent use and analysis.
[0073] Deep learning model training:
[0074] Use a large-scale labeled data set in the database, including normal data and various types of abnormal data, determine the deep learning architecture according to the characteristics of the problem, including convolutional neural network (CNN) or recurrent neural network (RNN), and use the selected deep learning architecture to build a deep learning model;
[0075] After the model is determined, the model parameters are initialized, and the loss function that matches the task is determined to measure the model prediction error, and then the optimization algorithm is determined, including stochastic gradient descent (SGD) or its variants;
[0076] The model is then trained in a loop. The specific process is as follows: load training data in batches; calculate the output of the model; calculate the error based on the loss function; update the model parameters through the optimization algorithm; and then use matching indicators to monitor the training process, including accuracy and loss value;
[0077] According to the monitoring results, adjust the model parameters and training strategies to optimize the model performance; during the training process, regularly evaluate the model performance on the validation set; based on the performance of the validation set, decide whether to stop training early to prevent overfitting; compare the performance of different models or different parameter settings and select the optimal model; save the last trained model for subsequent use.
[0078] Real-time anomaly detection:
[0079] The real-time collected data is passed into the trained deep learning model, and the model extracts and transforms the features of the input data. The model uses the algorithm and parameters to perform rapid calculations and inferences, judges the anomalies of the data, and determines whether there are anomalies by comparing with normal data patterns and based on the judgment criteria of the model; the results of the anomaly judgment are output, and the real-time acquired data is compared with the database. When there is new training data, the model is automatically updated according to the new training data to improve the accuracy of model detection.
[0080] Intelligent classification and evaluation:
[0081] Extract key features from the abnormal data monitored by the model to facilitate the classification of abnormal data, and then establish classification rules based on common abnormal types, including hardware failures and environmental interference; analyze different abnormal data to find correlations and differences, and then match the extracted features with the established classification rules to determine the type of abnormal data; then evaluate the severity of the classified abnormal data, and obtain the impact parameters of the severity of the impact of the abnormal data on the robot. The impact parameters include:
[0082] Impact on robot functions: The percentage of task completion rate reduction caused by the abnormal data, the number of failures of key functions, and the time required for function recovery are collected. After normalization, the percentage of task completion rate reduction and the number of failures of key functions are multiplied and divided by the time required for function recovery to obtain the power loss value. This power loss value is used as a standard to measure the impact of the abnormal data on the robot function.
[0083] Safety-related factors: By analyzing the risk level of abnormal data to personnel safety and assigning a risk score of 1-100; the potential degree of equipment damage is converted into a damage percentage; the degree of deviation from safety specifications is quantified by statistically analyzing the similarity with safety specifications. The higher the similarity, the lower the degree of deviation from safety specifications; the obtained risk score, damage percentage and similarity are normalized and substituted into the following formulas: The safety assessment value APZ is obtained, and this safety assessment value is used to measure the impact of abnormal data on safety-related factors; where fp, ss and xs are risk score, damage percentage and similarity respectively;
[0084] Economic impact factors: By analyzing the production delays caused by abnormal data and using the stop time as a measurement standard, the amount of property losses caused by the abnormal data and the negative impact on cost-effectiveness are collected, and the impact is assigned an impact score of 1-10. The larger the impact score, the greater the negative impact on cost-effectiveness. After normalizing the obtained stop time and impact score, the stop time is used as the bottom circle area and the impact score is used as the height to establish a cylindrical model. The volume of the cylindrical model is calculated and recorded as the shadow value. This shadow value is used as the standard for measuring the economic impact of abnormal data;
[0085] System stability: By counting the frequency of system crashes or errors caused by abnormal data and the degree of impact on the overall system performance, and quantifying the degree of impact through the percentage of performance degradation, the frequency of system crashes or errors and the percentage of performance degradation are normalized, and a bottom circle is established with the system crash or error frequency multiplied by the frequency weight coefficient as the radius, and a cone model is established with the performance degradation percentage multiplied by the performance weight coefficient as the height. The surface area of the cone model is calculated and recorded as the shadow value, and this shadow value is used as the standard for measuring the impact of abnormal data on system stability, where the frequency weight coefficient and the performance weight coefficient are set between 20-60 and 6-10 respectively;
[0086] Timeliness factor: Count the length of delay or lag caused by the abnormal data and the degree of impact on real-time response, and quantify the degree of impact on real-time response through impact time. Then, normalize and sum the obtained delay or lag time and impact time, and establish two sides of a triangle at a preset angle. Supplement the length of the other side of the triangle to form a complete triangle. Calculate the area of the triangle and record it as the hysteresis value. This hysteresis value is used as a standard to measure the impact of abnormal data on timeliness.
[0087] Predictability: Analyze the predictability difficulty of the abnormal data and assign a difficulty score of 1-10. The higher the difficulty score, the greater the predictability difficulty. Then count the frequency of the abnormal data from historical data. Normalize the difficulty score and the frequency of abnormal data, multiply them, and add a correction factor to get the predicted value, which is used as the standard for measuring the predictability of abnormal data. The correction factor is set between 1.5 and 5.2.
[0088] After normalizing the power loss value, safety assessment value, meridian shadow value, system shadow value, delay value and prediction value obtained by analysis, the power loss value, safety assessment value, meridian shadow value and system shadow value are used as the basic severity assessment standard of abnormal data and substituted into the following formula: To get the basic evaluation value YPJ, where They are power loss value, safety assessment value, shadow value and system shadow value. are the preset weight coefficients of power loss value, safety assessment value, shadow value and system shadow value, and their values are 1.25, 0.98, 1.11 and 0.86 respectively;
[0089] Then substitute the hysteresis value and the prediction value into the following formula as the correction factors of the basic evaluation value YPJ: To get the final judgment value ZPJ, where YC and YP are the hysteresis value and the prediction value respectively. are the preset weight coefficients of the hysteresis value and the prediction value, and are 1.24 and 0.97 respectively; and the final judgment value ZPJ is used as the final standard for measuring the severity of abnormal data;
[0090] The final judgment values ZPJ calculated from different abnormal data are then sorted by size, and the abnormal data with a large final judgment value ZPJ is processed first. The final judgment value ZPJ of the abnormal data is compared with the three preset final judgment value intervals. The three final judgment value intervals correspond to the severity levels of the abnormal data, namely mild abnormality, moderate abnormality and severe abnormality.
[0091] Adaptive processing strategy:
[0092] For minor anomalies, the abnormal data is recorded and automatic repairs are attempted, including adjusting the robot's behavior or recollecting data. For moderate anomalies, functions related to the abnormal data are restricted to avoid the expansion of problems or losses. For serious anomalies, the system is restarted or shut down urgently, and early warning signals are issued in a timely manner, requesting manual intervention as appropriate.
[0093] Model dynamic update:
[0094] Determine the data source, set the collection frequency, and regularly collect new data to update the deep learning model; introduce new anomaly types and scenarios to improve the generalization ability of the model; use new data and new anomaly types to continuously train the updated model, and verify the performance and generalization ability of the model; then dynamically adjust the model's parameters and weights based on the actual operation of the robot; then deploy the updated model to the robot system, observe the robot's performance in actual operation, collect feedback information, and continuously improve the model and update process based on the feedback.
[0095] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.
Claims
1. A method for processing abnormal data applied to robots, It is characterized in that The following steps are involved: S1: First, the collected real-time sensor data of the robot is preprocessed, and labels or annotations are added to the data to form a labeled data set and store it in the database; S2: Then, we label the dataset, determine the deep learning architecture, build and initialize the model parameters, select the loss function and optimization algorithm, perform the model training cycle, and monitor the training process; S3: The real-time collected data is passed to the trained deep learning model, which extracts and transforms features, quickly calculates and infers, determines data anomalies, and compares them with the data in the database. S4: Extract key features based on the abnormal data monitored by the model, establish classification rules, analyze the type and severity of abnormal data, and evaluate the impact of abnormal data on different dimensions of the robot. The analysis steps for the severity of abnormal data are as follows: Obtain the influencing parameters of the severity of the impact of abnormal data on the robot. The influencing parameters include the degree of impact on the robot function, safety-related factors, economic factors, system stability, timeliness factors and predictability. The power loss value, safety assessment value, meridian shadow value, system shadow value, delay value and prediction value are obtained through analysis of the influencing parameters. After normalization, the power loss value, safety assessment value, meridian shadow value and system shadow value are used as the basic severity assessment standard for abnormal data, and are substituted into the following formula: To get the basic evaluation value YPJ, where They are power loss value, safety assessment value, shadow value and system shadow value. They are the preset weight coefficients for power loss value, safety assessment value, shadow value and system shadow value; Then substitute the hysteresis value and the prediction value into the following formula as the correction factors of the basic evaluation value YPJ: To get the final judgment value ZPJ, where YC and YP are the hysteresis value and the prediction value respectively. are the preset weight coefficients of the hysteresis value and the prediction value respectively; and the final judgment value ZPJ is used as the final standard for measuring the severity of abnormal data; Then, the final judgment values calculated from different abnormal data are sorted by size, and the abnormal data with large final judgment values are processed first, and the final judgment value of the abnormal data is compared with the three preset final judgment value intervals. The three final judgment value intervals correspond to the severity levels of the abnormal data, namely mild abnormality, moderate abnormality and severe abnormality; The analysis process of power loss value, safety assessment value, shadow value and system shadow value is as follows: Impact on robot functions: The percentage of task completion rate reduction caused by the abnormal data, the number of failures of key functions, and the time required for function recovery are collected. After normalization, the percentage of task completion rate reduction and the number of failures of key functions are multiplied and divided by the time required for function recovery to obtain the power loss value. This power loss value is used as a standard to measure the impact of the abnormal data on the robot function. Safety-related factors: The risk level of abnormal data to personnel safety is analyzed and a risk score of 1-100 is assigned; the potential degree of equipment damage is converted into a damage percentage; the degree of deviation from safety specifications is quantified by statistically analyzing the similarity with safety specifications. The higher the similarity, the lower the degree of deviation from safety specifications. The obtained risk score, damage percentage and similarity are normalized and substituted into the following formulas: The safety assessment value APZ is obtained, and this safety assessment value is used to measure the impact of abnormal data on safety-related factors; where fp, ss and xs are risk score, damage percentage and similarity respectively; Economic impact factors: By analyzing the production delays caused by abnormal data, using the stop time as a measurement standard, and then collecting the amount of property losses caused by abnormal data and the negative impact on cost-effectiveness, the impact is assigned an impact score of 1-10. The larger the impact score, the greater the negative impact on cost-effectiveness; After normalizing the obtained stop time and impact score, the cylinder model is established with the stop time as the bottom circle area and the impact score as the height. The volume of the cylinder model is calculated and recorded as the shadow value, which is used as the standard for measuring the economic impact of abnormal data. System stability: By counting the frequency of system crashes or errors caused by abnormal data and the degree of impact on the overall system performance, and quantifying the degree of impact through the percentage of performance degradation, the frequency of system crashes or errors and the percentage of performance degradation are normalized, and a base circle is established with the system crash or error frequency multiplied by the frequency weight coefficient as the radius, and a cone model is established with the performance degradation percentage multiplied by the performance weight coefficient as the height. The surface area of the cone model is calculated and recorded as the shadow value, and this shadow value is used as the standard for measuring the impact of abnormal data on system stability; S5: Adopt different processing strategies according to the severity of abnormal data; S6: Regularly collect new data to update deep learning models, introduce new anomaly types and scenarios to improve model generalization capabilities, continuously train models and perform performance verification.
2. According to claim 1, a method for processing abnormal data applied to a robot, It is characterized in that The analysis process of the hysteresis value and the prediction value in S4 is as follows: Timeliness factor: Count the length of delay or lag caused by the abnormal data and the degree of impact on real-time response, and quantify the degree of impact on real-time response through impact time. Then, normalize and sum the delay or lag time and impact time, and use the preset angle to establish two sides of a triangle. Supplement the other side of the triangle to form a complete triangle. Calculate the area of the triangle and record it as the hysteresis value. This hysteresis value is used as a standard to measure the impact of abnormal data on timeliness. Predictability: Analyze the predictability difficulty of the abnormal data and assign a difficulty score of 1-10. The higher the difficulty score, the greater the predictability difficulty. Then, count the frequency of occurrence of the abnormal data from historical data. Normalize the difficulty score and the frequency of occurrence of abnormal data, multiply them, and add a correction factor to obtain the predicted value, which is used as the standard for measuring the predictability of abnormal data.
3. According to claim 1, a method for processing abnormal data applied to a robot, It is characterized in that The specific operation steps of S5 are as follows: When the abnormal data is judged to be a mild abnormality, the abnormal data is recorded and automatic repair is attempted, including adjusting the robot's behavior or re-collecting data; When the abnormal data is judged to be moderately abnormal, the functions related to the abnormal data are restricted; When the abnormal data is judged to be a serious abnormality, the system is restarted or shut down urgently, and an early warning signal is issued in time to request manual intervention.
4. The abnormal data processing method applied to a robot according to claim 1, It is characterized in that The specific operation steps of S1 are as follows: First, data from different sensors equipped in the robot are collected to collect multi-dimensional data; Then, the collected real-time sensor data is pre-processed and labeled; Preprocessing includes filtering the data, removing possible noise, and formatting the data; Labeling adds labels or annotations to the data to facilitate the learning and understanding of the model. The annotation content includes the classification of normal data and abnormal data. After completing the preliminary preprocessing and labeling of the data, a labeled data set is formed and stored in the database.
5. The abnormal data processing method applied to a robot according to claim 1, It is characterized in that The specific operation steps of S2 are as follows: Use a large-scale labeled data set in the database, including normal data and various types of abnormal data, determine the deep learning architecture according to the characteristics of the problem, and use the selected deep learning architecture to build a deep learning model; After the model is determined, the model parameters are initialized, and the loss function that matches the task is determined to measure the model prediction error, and then the optimization algorithm is determined, including stochastic gradient descent or its variants; Then the model is trained in a loop. The specific process is as follows: Batch loading of training data; Calculate the output of the model; Calculate the error based on the loss function; Update model parameters through optimization algorithms; Use matching metrics to monitor the training process, including accuracy and loss values; Adjust model parameters and training strategies based on monitoring results to optimize model performance; During training, model performance is regularly evaluated on the validation set. Based on the performance of the validation set, determine whether to stop training early to prevent overfitting; Compare the performance of different models or different parameter settings and select the optimal model; Save the final trained model.
6. The abnormal data processing method applied to a robot according to claim 1, It is characterized in that The specific operation steps of S3 are as follows: The real-time collected data is passed into the trained deep learning model; The model extracts and transforms the features of the input data, and uses the model’s algorithms and parameters to perform rapid calculations and inferences; Judge the anomalies of the data and determine whether there are anomalies by comparing it with the normal data pattern and according to the judgment criteria of the model; The results of abnormal judgment are output, and the real-time acquired data is compared with the database. When there is new training data, the model is automatically updated according to the new training data to improve the accuracy of model detection.
7. The abnormal data processing method applied to a robot according to claim 1, It is characterized in that The specific operation steps of S6 are as follows: First, determine the data source, set the collection frequency, and regularly collect new data to update the deep learning model; Introducing new exception types and scenarios; Continue to train the updated model using new data and new anomaly types, and validate the model’s performance and generalization capabilities; Then, according to the actual operation of the robot, the parameters and weights of the model are dynamically adjusted; The updated model is then deployed to the robot system, the robot’s performance in actual operation is observed, feedback information is collected, and the model and update process are improved based on the feedback.
Citation Information
Patent Citations
Computer security management system and method based on artificial intelligence
CN117574361A
Method for predicting completion ability of man-machine interaction user
CN117593037A
Digital authentication system for power equipment
CN117811828A
Customer service risk assessment method and system based on bidirectional feedback mechanism
CN118114985A
Temperature monitoring and early warning method for computer mainboard
CN118277212A
