Unmanned equipment control behavior pattern analysis method

Abstract

The invention relates to an unmanned equipment control behavior pattern analysis method. The method comprises the following steps: acquiring to-be-analyzed control training data of an operator; the control training data comprises eye movement focus data, electroencephalogram concentration degree data, posture of two arms of a human body and voice information; inputting the to-be-analyzed control training data into the control decision analysis model to obtain a control decision result; the decision analysis model is obtained by training a training set; performing abnormal value elimination on to-be-analyzed control training data through the control decision analysis model, performing validity preprocessing on the data after abnormal value elimination, obtaining feature data, calculating the credibility of the feature data, obtaining the support rate of a corresponding decision plan according to the type of the feature data and the decision tendency degree, and calculating the decision plan support rate. And support rate updating is carried out in combination with the credibility, data fusion processing is carried out through the updated support rate and the weighted value after fuzzy quantization, and a control decision result is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of control decision analysis, and particularly to a method for analyzing the manipulation behavior mode of unmanned equipment. Background Art

[0002] In the field of remote real-time control, the manipulation of remote unmanned equipment by operators mainly relies on manipulation peripherals and software-assisted manipulation logic. The design of manipulation peripherals and the formulation of manipulation logic mainly depend on the research and understanding of the functions of remotely operated unmanned equipment and the manipulation behaviors of operators by designers. However, it is difficult for designers to form quantifiable standards based on their perceptual understanding of the above factors, and there is a lack of scientific research methods for the manipulation behavior modes of operators with different age structures, genders, and knowledge levels. Therefore, it is necessary to develop scientific research methods to guide the optimal design of manipulation processes and functions.

[0003] In the paper "Research on One Person Two Machines Manipulation Technology Based on Multi-Channel Behavior Characteristics" by Zhang Shuwei of Nanjing University, a manipulation mode was designed starting from the analysis of the characteristics of manipulation input peripherals. This type of research method is not applicable to the improved design of existing application devices and cannot obtain the manipulation effect data required for improved design. Among them, the evaluation and analysis method for manipulation effects is also not applicable to the evaluation of complex multi-step manipulation behaviors.

[0004] In the paper "Research on Driving Behavior in the Merging Area of Cold Region Night Expressway Based on Eye Tracker" by Ding Shitao of Jilin Jianzhu University, only the analysis of the images captured by the camera and the data collected by the eye tracker was carried out, and finally improvement suggestions for external road signs and markings were obtained, which are difficult to play a prompting role in the operation behaviors of the operators themselves. Summary of the Invention

[0005] Since the operation end needs to perform complex control on various actuators of the controlled unmanned equipment during the remote operation process, it is necessary to equip appropriate operation peripherals and formulate reasonable manipulation mechanisms to achieve effective control of the controlled end. Different types of operators have different requirements for manipulation mechanisms and manipulation interactions. How to design manipulation peripherals, interaction interfaces, and control mechanisms that can adapt to more operators requires obtaining more data on manipulation behavior modes as a reference. Traditional design methods can only design operation peripherals, interaction interfaces, and control mechanisms through simple perceptual understanding and past design experience. Therefore, the present invention requires a scientific research method for operation modes to digitally record, behaviorally analyze, and evaluate the current manipulation situation, use this method to quickly and efficiently collect a large amount of manipulation data of operators, and through further statistical analysis, better understand the current manipulation pain points and optimization directions.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A method for analyzing the control behavior pattern of unmanned equipment, comprising:

[0008] Collecting control training data to be analyzed of an operator; the control training data includes: eye movement focus data, EEG concentration data, human body arm postures, and voice information;

[0009] Inputting the control training data to be analyzed into a control decision analysis model to obtain a control decision result; the decision analysis model is obtained by training with a training set, and the training set includes: control training data, feedback data, and data labels containing true operation intentions;

[0010] Removing outliers from the control training data to be analyzed through the control decision analysis model, performing validity preprocessing on the data after removing outliers to obtain feature data, calculating the credibility of the feature data, obtaining the support rate of the corresponding decision plan according to the type and decision tendency degree of the feature data, and updating the support rate in combination with the credibility, and performing data fusion processing through the updated support rate and the weighted value after fuzzy quantization to obtain a control decision result.

[0011] Optionally, removing outliers from the control training data to be analyzed includes:

[0012] Obtaining the same-type data of the control training data to be analyzed, calculating the distance between the same-type data and other same-type data outside the same-type data, and constructing a distance matrix based on the distance;

[0013] Based on the distance matrix, obtaining the similarity degree between the same-type data, performing normalization processing on the similarity degree between the same-type data, and obtaining the trust function of the same-type data;

[0014] Setting an outlier threshold: δ = DSUP(m) × 2%; where δ is the outlier threshold, DSUP(m) is the median of DSUP(s i ) and DSUP(s i ) is the trust function;

[0015] When DSUP(m) - δ ≤ DSUP(s i ) ≤ DSUP(m) + δ, then save the trust function value and remove the outliers.

[0016] Optionally, obtaining the feature data includes:

[0017] Performing feature unit extraction on the data after removing outliers; the feature unit extraction is: screening out numerical values and characters with decision features from the data containing redundant information, and using key-value pairs to search for features and attributes corresponding to the features in the data after removing outliers to obtain feature units;

[0018] Perform an intersection operation on the said feature units to obtain an independent feature data set.

[0019] Optionally, during the process of searching for data features, introduce a degree of difference to evaluate the degree of conformity between the data features and the corresponding attributes:

[0020]

[0021] where n is the number of attributes, m is the number of intersections, ta i (u j ) is the statistical quantity extracted from the feature unit u j into the attribute a i and T is the statistical quantity of all feature units.

[0022] Optionally, calculating the credibility of the said feature data includes:

[0023] Obtain the weights corresponding to all data in the attribute, and calculate the similarity between each feature data and the attribute according to the weights:

[0024] Sim=(v i ,a j )=v i / ||v i ||;

[0025] where v i is the vector of the weights corresponding to all data in the attribute;

[0026] According to the said similarity, perform a matching annotation on the attribute set to obtain the said credibility:

[0027]

[0028] where SIMX a is the similarity set, sim i is the similarity corresponding to the attribute i, and |X a | is the number of attributes in the attribute set X a .

[0029] Optionally, obtaining the support rate of the said corresponding decision-making plan includes:

[0030] Divide the feature data obtained after effective preprocessing into binary type, degree type, and random variable; the binary type is: there are only two results, yes or no, the degree type is: the control decision has a hierarchical difference and is divided into multiple planning levels, and the random variable is: has a preset distribution characteristic;

[0031] Obtain the support rate of the binary type data for the corresponding decision-making plan:

[0032]

[0033] Among them, m is the number of true of the boolean type, and n is the number of false of the boolean type;

[0034] Obtaining the support rate of the corresponding decision plan for the degree type data and the random variable data includes:

[0035] When the degree type data, the random variable data and the decision plan influence degree are positively correlated, obtaining the support rate of the corresponding decision plan:

[0036]

[0037] Among them, μ is the expectation, σ is the variance, x is the data, m is the defined data acquisition interval, and i is the index of x within the m interval;

[0038] When the degree type data, the random variable data and the decision plan influence degree are negatively correlated, obtaining the support rate of the corresponding decision plan:

[0039] I′(x) = 1 - I(x);

[0040] Among them, I(x) is the support rate of the corresponding decision plan obtained when it is positively correlated.

[0041] Optionally, the support rate update includes:

[0042] Using the credibility r corresponding to the data source i and the support rate to perform the support rate update:

[0043]

[0044] Among them, n is the dimension of the decision set, Inf ij represents the degree to which data i supports the execution of decision plan j by the module, that is, the support rate, Inf ij_min is the minimum value of the support rate, Inf ij_ave is the average value of the support rate, Inf ij_max is the maximum value of the support rate, r i is the credibility.

[0045] Optionally, obtaining the updated support rate:

[0046] Inf ij ′ = λ1Inf ij_min + λ2Inf ij_ave + λ3Inf ij_max ;

[0047] Among them, λ1, λ2 and λ3 are all constant coefficients affected by the decision tendency.

[0048] Optionally, obtaining the weighted value after fuzzy quantization includes:

[0049] Quantize the fuzzy semantics according to the decision tendency degree to obtain the weighted value of the corresponding data source:

[0050] w i = s(i - n) - s((i - 1) / n);

[0051] where s(.) is fuzzy quantization, n is the dimension of the decision set, and i is the sequence of variables in the decision set.

[0052] Optionally, obtaining the decision tendency degree includes:

[0053] According to the decision planning set and the weighted vector corresponding to the decision planning set, obtain the weighted decision relationship, and measure the decision tendency degree for the weighted decision relationship.

[0054] The beneficial effects of the present invention are:

[0055] The present invention can obtain rich multi-source sensing data related to manipulation, and perform data preprocessing on the uncontrollability of sensing information, especially human biological information, and unconscious operation data. Refer to the operation video and feedback data to form key value pairs of feature data and effective attributes, and classify them into binary type, degree type, and random variable type, and then evaluate their credibility for post-processing use.

[0056] For the problem of conflicts in multi-sensor pointing results that may be introduced by the sensors and human motion characteristics of the present invention, a method for removing outliers of the same type of data based on Euclidean distance is used to solve the problem of inconsistent strategy planning corresponding to multi-source data.

[0057] From the perspective of human-machine control, the present invention applies theories such as fuzzy control logic, non-parametric statistics, and intelligent cooperation technology to study key theories and methods for modeling the process of task aggregation and collaborative response of operators, especially the realization of multi-source information fusion and human-machine collaborative efficiency in different operation tasks. Establish a microscopic collaborative simulation model of operators, which is a key technology for the control of unmanned equipment, the development, design, and improvement of human-machine interaction systems. It can greatly improve the sampling efficiency, collect more manipulation experience data of operators, and has very important theoretical value and practical significance for improving operation comfort, reducing stagnant time, reducing redundant operations, and improving control efficiency.

[0058] The present invention uses a data-based method as a means for collecting and analyzing the operation behavior of operators, which can solve the problems of insufficient multi-source data of sampling personnel, shallow data mining level, analysis staying at the perceptual level, and lack of deep data-based means in the analysis of general operation modes.

[0059] The present invention uses a software automation method to solve the problem of the explosion of manual analysis workload caused by the increase in the number of sampling personnel, can select different populations as control groups in a larger range, and can more conveniently apply statistical means for post-processing.

[0060] The present invention graphically and visually presents the manipulation stage and effects in the form of comparison graphs of the operation sequence, operation frequency, and operation duration, and uses the operation sequence list and operation mode analysis diagram as means to compare operation effects, and summarizes the operation habits and concentrated points of operation errors of different operating populations. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0062] Figure 1 It is a flowchart of a method for analyzing the operation behavior mode of an unmanned equipment according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0064] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0065] Currently, when operating unmanned equipment, there is room for improvement in the immersion, humanized design, and coherence of the operation method of human-machine interaction. It is necessary to comprehensively apply design knowledge and human factors engineering knowledge to optimize the design of human-machine interaction strategies, especially the interaction interface and interaction rules. Therefore, the electroencephalogram signals, eye signals, voice signals, and the postures of the human body and arms of the operator are collected and processed simultaneously as interaction reference data and control resources, and the analysis and output of manipulation behaviors are carried out in combination with the status information fed back by the controlled end. By effectively using the above multi-source data to accurately output manipulation instructions, a multi-source data fusion algorithm based on weighted decision-making of an expert system is proposed. And to solve the high-conflict problem that may be introduced by multi-sensors, an evaluation rule for similar data based on offline learning is designed, and outlier rejection based on Euclidean distance is adopted on the basis of the rule. Subsequently, aiming at the non-deterministic characteristics of multi-source data, credibility is introduced. Credibility is the reliability degree of the conclusion that the data supports a certain decision-making plan. A fuzzy support degree solving algorithm is designed to obtain the influence degree of various data on the decision-making of the working module; then, using the control decision tendency, the fuzzy semantics is quantified to obtain the corresponding weighted value, and all decision items are calibrated with priorities through the support degree and fused with the quantified weighted value; finally, on the basis of the fused data, the decision-making plan for the manipulation behavior is realized, further improving the fluency of human-machine interaction, improving the recognition accuracy of the control intention of the operator, and thus improving the overall control efficiency. It is achieved to improve the matching accuracy of the operation behavior and the control effect of the unmanned equipment by collecting multiple biological signals of the human body and cooperating with each other for interaction resources.

[0066] As Figure 1 shown, this embodiment discloses a method for analyzing the manipulation behavior mode of unmanned equipment, including: collecting the control training data to be analyzed of the operator; the control training data includes but is not limited to: eye movement focus data, electroencephalogram concentration data, postures of the human arms, and voice information; inputting the control training data to be analyzed into the control decision analysis model to obtain the control decision result; the decision analysis model is obtained by training with a training set, and the training set includes: control training data, feedback data, and data labels containing real operation intentions; outlier rejection is performed on the control training data to be analyzed through the control decision analysis model, and the data after outlier rejection is preprocessed for effectiveness to obtain feature data. Outlier rejection is performed on the control training data to be analyzed through the control decision analysis model, and the data after outlier rejection is preprocessed for effectiveness to obtain feature data, and the credibility of the feature data is calculated. According to the type and decision tendency degree of the feature data, the support rate of the corresponding decision-making plan is obtained, and the support rate is updated in combination with the data source credibility. Data fusion processing is performed through the updated support rate and the weighted value after fuzzy quantization to obtain the control decision result.

[0067] Specifically, data collection and offline learning:

[0068] All functional peripherals of the operating end and all actuators of the operated end are introduced in detail one by one for each single module to form a module introduction in the operation manual. Then, the interlocking usage relationships of each module and the interlocking operation conditions of each peripheral are summarized to form an introduction to the operation functions in the operation manual. Based on the operation habits and experience of professional operators, operation guides for specific subjects are formed, and specific operation learning videos are retained for the personnel participating in the test to learn from the experience. Then, the test personnel are invited to observe the actual operation process of the operation subjects of the unmanned equipment by professional technicians on-site, so that the test personnel can generalize the operation experience they have learned into their own actual operation experience.

[0069] The operator wears multi-source bio-information acquisition equipment and operates according to the standard subject process to collect videos, corresponding feedback data and control data during the operation of the unmanned equipment as basic research data. Design offline control data learning software. During the playback of typical unmanned equipment behavior control videos and feedback data, analyze the operator's control training data, including but not limited to: input data such as eye movement focus data, EEG concentration data, human arm postures, and voice information. And match the input data with the feedback data as a research reference data set. While the operator performs necessary operations in the control video, every time an operation is completed, the auxiliary personnel pause the collection once and input the real operation intention of the operator just now as a data label, thus completing the complete process of data collection and data marking for offline learning.

[0070] Furthermore, outlier rejection of the control training data to be analyzed includes: obtaining the same type of data of the control training data to be analyzed, calculating the distance between the same type of data and other data outside the same type of data, and constructing a distance matrix based on the distance; based on the distance matrix, obtaining the similarity degree between the same type of data, normalizing the similarity degree between the same type of data, and obtaining the belief function of the same type of data; setting an outlier threshold: δ = DSUP(m) × 2%; where δ is the outlier threshold, DSUP(m) is the median of DSUP(s i ) and DSUP(s i ) is the belief function; when DSUP(m) - δ ≤ DSUP(s i ) ≤ DSUP(m) + δ, then save the belief function value and reject the outlier.

[0071] Specifically, the outlier rejection process:

[0072] Regarding the conflict problem of multi-sensor pointing results that may be introduced by sensors and human motion characteristics, an outlier removal method for similar data based on Euclidean distance is used. Based on the data evaluation rules obtained from offline learning preprocessing, the outliers in the multi-sensor data are removed. The basic idea is that the smaller the distance between two similar data, the greater the similarity degree and the greater the authenticity of the data, and vice versa. By obtaining the average value of the sum of the distances between pairwise data and setting corresponding thresholds to remove outliers. Let the number of sensors be n, then the similar data s i ={s i |i=1,2,…,n}, s i and the distance from other similar data s i except s j can be expressed as: i≠j,i=1,2,…,n, and the distance between s i and other similar data s i except s j can be calculated and expressed as a distance matrix:

[0073]

[0074] The distance Dd(d ij ) reflects the similarity degree with other data. The smaller the element value in Dd(d ij ), the greater the similarity degree.

[0075] The similarity Dd(d i ) is normalized as the formula: Dd(s i )=Dd(d ij )∑ni=1∑nj=1Dd(d ij ), and ∑nj=1DG(s i )=1. The smaller the distance between similar data, the greater the similarity degree and the greater the authenticity of the data. After obtaining the normalized similarity DG(s i ) of a data, the belief function of the similar data can be obtained:

[0076] DSUP(s i )=1-DG(s i )i=1, 2,…n;

[0077] The belief function DSUP value of similar data reflects the authenticity of the data. Usually, the DSUP(s i ) values do not vary much, and the larger the DSUP(s i ), the more authentic the data. When outliers appear in the similar data, the DSUP(s i ) value will be very small, and the DSUP(s i) values will be very close; conversely, the normal values are relatively large, and the threshold δ is centrally set to δ = DSUP(m) × 2%, where DSUP(m) is the median of DSUP(s i ) When DSUP(m) - δ ≤ DSUP(s i ) ≤ DSUP(m) + δ, retain the value of DSUP(s i ) and eliminate the abnormal values to obtain the correct data of the same category.

[0078] Furthermore, obtaining the feature data includes: extracting feature units from the data after eliminating the abnormal values; the feature unit extraction is: screening out the numerical values and characters with decision-making features from the data containing redundant information, and using key-value pairs to search for the features and their corresponding attributes in the data after eliminating the abnormal values to obtain the feature units; performing an intersection operation on the feature units to obtain an independent feature data set.

[0079] Specifically, data validity preprocessing:

[0080] The data preprocessing stage is responsible for integrating the multi-source data of human biological information into data packets with the same structure. And due to a certain degree of uncontrollability of the human body, the data may be mixed with invalid information brought by human characteristics and unconscious actions of the operator. Therefore, it is necessary to extract feature units from the multi-source data, that is, to screen out the numerical values and characters with decision-making features from the data containing redundant information. To avoid the sparse characteristics of the data, here key-value pairs are used to search for the features and their corresponding attributes in the data. Thus, the final expression form of the feature unit is as follows:

[0081] C = {<u1, ta(u1)>, <u2, ta(u2)>,... <u i , ta(u i )>};

[0082] In the formula, u i represents the feature unit; ta(u i ) represents the statistical quantity of the corresponding attribute of u i . According to all the feature units, performing an intersection operation on them can obtain the separate feature data sets of each sensing information. In the process of extracting the feature data, in order to optimize the extraction performance, the degree of difference is introduced to evaluate the degree of conformity between the feature data and the effective attributes. The calculation formula is as follows:

[0083]

[0084] In the formula, n is the number of attributes, m is the number of intersections, ta i (u j ) is the statistical quantity of the feature unit u j extracted into the attribute a i , and T is the statistical quantity of all the feature units.

[0085] Assume that all feature data corresponds to a weight in the attribute, then the vector can be expressed as v = {v1, v2,... v i}, for any feature data u j and the similarity calculation between the attribute a i is as follows:

[0086] Sim=(v i , a j ) = v i / ||v i ||;

[0087] According to the similarity, perform matching annotation on the attribute set X a , and thus obtain the data credibility calculation formula as:

[0088]

[0089] In the formula, SIMX a is the similarity set, sim i is the similarity corresponding to the attribute i, |X a | is the number of attributes in X a .

[0090] Furthermore, obtaining the support rate of the corresponding decision-making plan includes:

[0091] Divide the feature data obtained after effective preprocessing into binary type, degree type, and random variable; the binary type is: there are only two results, yes or no, the degree type is: the control decision has a hierarchical difference, divided into multiple planning levels, and the random variable is: has a preset distribution characteristic;

[0092] Specifically, weighted data support processing:

[0093] Multi-source data of human body biological information will generate different types according to the differences of interaction objects. Therefore, after preprocessing, the multi-source data of human body biological information is divided into three categories: binary type, degree type, and random variable for analysis. Among them, the binary type means that when the module interacts, the transmitted data is Boolean data, and there are only two results, yes or no. For example, whether the arm posture of the operator is changing consciously. The degree type means that the module control decision has a hierarchical difference and can be divided into multiple planning levels. For example, the degree of EEG concentration, not concentrated, low concentration, medium concentration, high concentration. And the data of the random variable type has a certain distribution characteristic. For example, the coordinate area where the eye movement data is located. This type of variable also needs to be classified or graded according to the actual situation of the interface interaction software and other influencing factors during the control process. The subsequent data fusion method can adopt the same method as the degree category.

[0094] In the process of obtaining the support degree of the corresponding control strategy for multi-source data, when performing data conversion, if the binary data source is selected from the data set, assuming that the numbers of true and false in the Boolean type correspond to m and n respectively, then the support rate of this data for the future module decision-making plan is expressed as follows:

[0095]

[0096] If other types of data are selected from the data set, the description of the hierarchical data usually adopts a direct or inverse relationship. Due to the serious uncertainty of multi-source data, especially when designing the decision-making plan support for random variables, fuzzy processing is introduced. Assuming that the expectation of the random data is μ and the variance is σ, when the influence degree of this data on the decision-making plan is positively correlated, the corresponding support rate calculation method is as follows:

[0097]

[0098] When the influence degree of this data on the decision-making plan is negatively correlated, the corresponding support rate calculation method is as follows:

[0099] I′(x) = 1 - I(x);

[0100] Assume that the decision-making plan of a certain work module is represented as the set D = {d1, d2,..., d n}, and its corresponding weighted vector is represented as W = {w1, w2,..., w n}, where n represents the dimension of the decision set, then the weighted decision relationship is expressed as:

[0101]

[0102] In the formula, it satisfies According to the fuzziness of multi-source data, the calculation method for any weighted value is:

[0103] w i = s(i - n) - s((i - 1) / n);

[0104] In the formula, s(.) represents fuzzy quantization, and its processing process is described as follows:

[0105]

[0106] In the formula, α and β represent quantization factors, and their value ranges are α, β ∈ [0, 1]. For the weighted decision relationship, the calculation for measuring the decision tendency degree is:

[0107]

[0108] For multiple support degrees output by this project, each has its own function. The behavior feature values corresponding to each sensing information can be regarded as the support degrees of the sensing information for each behavior feature corresponding to the sensor, and used as the input for subsequent analyzers and controllers.

[0109] Furthermore, obtaining the degree of decision tendency includes: obtaining the weighted decision relationship according to the decision planning set and the weighted vector corresponding to the decision planning set, and measuring the degree of decision tendency for the weighted decision relationship.

[0110] Specifically, multi-source data fusion processing:

[0111] When the unmanned equipment is working, the control planning of its task module is affected by two types of data sources, namely feedback data and control data Let k represent the number of data sources, and their corresponding credibility is expressed as r = {r1, r2,..., r k}. Then the fusion processing flow of multi-source data can be described as follows:

[0112] 1) Preprocess to obtain the feature data and its credibility, and store them in the data set; the credibility is the reliability degree of the data supporting a certain decision planning, similar to the weight of the sensor reliability. Its main function is to comprehensively represent the credibility degree of the source data of each sensor, so as to update the support rate during multi-sensor data fusion, ensuring that when performing multi-source data fusion, not only the direct support information provided by each data source is considered, but also the reliability of these data sources themselves is comprehensively considered, making the final decision more robust and accurate.

[0113] 2) Solve the support rate Inf ij =(a ij , b ij , p ij ) for the corresponding decision planning of different data, and satisfy 0 ≤ a ij ≤ b ij ≤ p ij ≤ 1. Inf ij represents the degree to which data i supports the module to execute decision planning j.

[0114] 3) According to the decision tendency, adopt the corresponding fuzzy quantization processing to calculate the weighted value w i corresponding to data source i.

[0115] 4) Use the r i and Inf ij values corresponding to data source i to perform the update operation of Inf ij , which is described as follows:

[0116]

[0117] The new support degree obtained by adopting fuzzy data conversion is:

[0118] Inf ij ′ = λ1Inf ij_min + λ2Inf ij_ave + λ3Inf ij_max ;

[0119] In the formula, λ1, λ2 and λ3 are all constant coefficients affected by the decision-making tendency. If the condition p(w) < 0.5 holds, then let λ1 = 1 - 2p(w), λ2 = 2p(w), λ3 = 0; p(w) is the degree of decision-making tendency.

[0120] If the condition p(w) ≥ 0.5 holds, then λ1 = 0, λ2 = 2 - 2p(w), λ3 = 2p(w) - 1.

[0121] 5) Using the weighted value w i and the updated support rate, the data is subjected to fusion processing, thereby obtaining a weighted decision relationship, and then the control decision result is determined accordingly.

[0122] Multi-source information fusion is a theory and method for processing multiple information, which can comprehensively process data at different times and spaces, so as to obtain a more accurate and reliable description of the real environment. Its basic principle and starting point are: make full use of multiple information sources, through reasonable allocation and use of them and the information they provide, combine the redundant or complementary information of multiple information sources in space or time according to a certain criterion, in order to obtain a consistent interpretation or description of the measured object, so that this information system can obtain better performance than the system composed of subsets of its individual components. Information fusion can be divided into data-level information fusion, feature-level information fusion and decision-level information fusion according to the level relative to information representation. The advantages of multi-source information fusion are prominently reflected in aspects such as information fault tolerance, complementarity, real-time performance and low cost. Data-level information fusion refers to the fusion of matching collected data, which belongs to the lowest level of fusion. Feature-level information fusion belongs to the intermediate level, which is feature-level joint recognition. Before fusion, the fusion system first extracts features from the original information, and then conducts comprehensive analysis and processing on the feature information. Decision-level information fusion is a high-level fusion, which is the final result of the three-level fusion, and its purpose is to provide a basis for control decision-making, and the fusion result directly affects the decision-making level.

[0123] Specifically, the comparison method:

[0124] After building the operation behavior analysis model by the above method, the operation behavior habits of different testers are analyzed in real time, and a decision-making planning sequence list and the corresponding original control data pairs are output, so as to intuitively compare the operation processes of different groups of people, and save the data for further increasing statistical means to analyze the operation behavior characteristics, control interaction rules and optimization directions of different groups of people. And according to the characteristics of the operation subjects, a visual control behavior pattern analysis chart is drawn, with the operation sequence as the abscissa, the operation frequency of the same step as the ordinate, and the operation time-consuming of the same step as the marked point area, to draw the control behavior pattern analysis chart. The operation behavior pattern is displayed to researchers from multiple dimensions of data and visualization, so as to better formulate control rules and optimize the interface interaction design.

[0125] A research method for the operation behavior pattern of a multi-dimensional presence-sensing servo unmanned equipment by operators with different age structures, genders, knowledge levels, etc. under remote real-time control conditions, and the control effect is evaluated in a dataized and visualized manner to guide the optimization design of the control mode. For example, when the familiarity of the testers with the operation process tends to be stable, there are obvious differences in the mistakes of different operators in the same operation steps. The testers with the same attributes are divided according to the different types of mistakes, and are divided into the following three categories: technicians VS front-line users, older testers VS younger testers, female testers VS male testers. Then, the operation step sequence is drawn on the X-axis, and the operation frequency of the operation steps is compared on the Y-axis, and the operation time of the step is represented by the size of the graphic area, and the operation situation is displayed and compared graphically. According to the characteristics of the drawn graph, the operation differences of the three reference groups of people are specifically compared. Then, according to the operation differences corresponding to the specific operation steps, combined with the characteristics of the reference groups of the testers, the causes of the operation differences are analyzed.

[0126] This embodiment addresses the above-mentioned several problems affecting control. By referring to the data collected in the modeling stage and understanding the control habits corresponding to different groups of people, it is possible to specifically adopt methods such as upgrading the corresponding operation peripherals, improving the corresponding operation mechanisms, adding necessary marking scales, and adding necessary automated steps to solve the corresponding problems existing in the control process of the unmanned equipment.

[0127] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for analyzing the control behavior pattern of unmanned equipment, characterized in that, Including: Collecting control training data to be analyzed of an operator; The control training data includes: eye movement focus data, EEG concentration data, human body's two-arm postures, and voice information; Inputting the control training data to be analyzed into a control decision analysis model to obtain a control decision result; the decision analysis model is obtained by training with a training set, and the training set includes: control training data, feedback data, and data labels containing real operation intentions; Removing outliers from the control training data to be analyzed through the control decision analysis model, preprocessing the data after removing outliers for validity to obtain feature data, calculating the credibility of the feature data, obtaining the support rate of the corresponding decision plan according to the type and decision tendency degree of the feature data, and updating the support rate in combination with the credibility, and performing data fusion processing through the updated support rate and the weighted value after fuzzy quantization to obtain a control decision result.

2. The method for analyzing the unmanned equipment control behavior pattern according to claim 1, wherein Removing outliers from the control training data to be analyzed includes: Obtaining the same-kind data of the control training data to be analyzed, calculating the distance between the same-kind data and other same-kind data outside the same-kind data, and constructing a distance matrix based on the distance; Based on the distance matrix, obtaining the similarity degree between the same-kind data, and performing normalization processing on the similarity degree between the same-kind data to obtain the belief function of the same-kind data; Set the outlier threshold: δ = DSUP(m) × 2%; where δ is the outlier threshold, and DSUP(m) is the median of DSUP(s i ) and DSUP(s i ) is the belief function; When DSUP(m)-δ ≤ DSUP(s i ) ≤ DSUP(m)+δ, the trust function value is saved and the outliers are removed.

3. The method for analyzing the unmanned equipment control behavior pattern according to claim 1, wherein Obtaining the feature data includes: Performing feature unit extraction on the data after removing outliers; the feature unit extraction is: screening out the numerical values and characters with decision features from the data containing redundant information, and using key-value pairs to search for the features and the attributes corresponding to the features in the data after removing outliers to obtain feature units; Performing an intersection operation on the feature units to obtain an independent feature data set.

4. The method for analyzing the control behavior mode of unmanned equipment according to claim 3, characterized in that In the process of searching for data features, introducing a degree of difference to evaluate the conformity degree between the data features and the corresponding attributes: Among them, n is the number of attributes, m is the number of intersections, and ta i (u j ) is the statistical quantity of the feature unit u j extracted into the attribute a i , and T is the statistical quantity of all feature units.

5. The method for analyzing the control behavior pattern of unmanned equipment according to claim 1, wherein Calculating the credibility of the feature data includes: Obtaining the weights corresponding to all data in the attribute, and calculating the similarity between each feature data and the attribute according to the weights; Sim=(v i ,a j )=v i / ||v i ||; Among them, v i is the vector of the weights corresponding to all data in the attributes; According to the similarity, performing matching annotation on the attribute set to obtain the credibility; Among them, SIMX a is the similarity set, and sim i is the similarity corresponding to attribute i, |X a | is the number of attributes in the attribute set X a The number of attributes within.

6. The method for analyzing the control behavior pattern of unmanned equipment according to claim 1, characterized in that, Obtaining the support rate of the corresponding decision plan includes: Dividing the feature data obtained after validity preprocessing into binary type, degree type, and random variable; the binary type is: there are only two results of yes or no, the degree type is: the control decision has a hierarchical difference, divided into multiple planning levels, and the random variable is: having a preset distribution feature; Obtaining the support rate of the corresponding decision plan for the binary type data: Where m is the number of true of the boolean type, and n is the number of false of the boolean type; Obtaining the support rate of the corresponding decision plan for the degree type data and the random variable data includes: When the degree type data and the random variable data are positively correlated with the influence degree of the decision plan, obtaining the support rate of the corresponding decision plan: Where μ is the expectation, σ is the variance, x is the data, m is the defined data acquisition interval, and i is the index of x within the m interval range; When the degree type data, the random variable data and the decision-making planning influence degree are negatively correlated, obtain the support rate of the corresponding decision-making planning: I′(x) = 1 - I(x); where I(x) is the support rate of the corresponding decision-making planning obtained when it is positively correlated.

7. The method for analyzing the operation behavior mode of unmanned equipment according to claim 1, wherein Performing support rate update includes: Using the credibility r corresponding to the data source i and the support rate to perform support rate update: where n is the dimension of the decision set, Inf ij represents the degree to which data i supports the module to execute decision plan j, that is, the support rate, Inf ij_min is the minimum value of the support rate, Inf ij_ave is the average value of the support rate, Inf ij_max is the maximum value of the support rate, r i is the credibility.

8. The method for analyzing the control behavior mode of unmanned equipment according to claim 7, characterized in that Obtain the updated support rate: Inf ij ′ = λ1Inf ij_min + λ2Inf ij_ave + λ3Inf ij_max ; where λ1, λ2 and λ3 are all constant coefficients affected by the decision-making tendency.

9. The method for analyzing the unmanned equipment control behavior pattern according to claim 1, wherein Obtaining the weighted value after fuzzy quantization includes: According to the decision-making tendency degree, perform quantization processing on the fuzzy semantics to obtain the weighted value of the corresponding data source: w i = s(i - n) - s((i - 1) / n); where s(.) is fuzzy quantization, n is the dimension of the decision-making set, and i is the sequence of variables in the decision-making set.

10. The method for analyzing the unmanned equipment control behavior pattern according to claim 9, wherein Obtaining the decision-making tendency degree includes: According to the decision-making planning set and the weighted vector corresponding to the decision-making planning set, obtain the weighted decision-making relationship, and measure the decision-making tendency degree for the weighted decision-making relationship.

Images