Search

CN-122022596-A - Team cognitive efficiency decoupling evaluation method, system and storage medium based on hybrid expert network

CN122022596ACN 122022596 ACN122022596 ACN 122022596ACN-122022596-A

Abstract

The invention discloses a team cognitive effectiveness decoupling evaluation method, a system and a storage medium based on a hybrid expert network, and belongs to the technical field of artificial intelligence and team management. The method comprises the steps of obtaining deep state characterization of an individual and dynamic cooperation network characteristics, decoupling a team cognitive state into three independent element characteristic vectors of the individual, the group and the task, constructing a MoE model comprising a gating network and three expert sub-networks, wherein the expert networks correspond to the three elements respectively, dynamically calculating each expert weight by the gating network according to the real-time state, outputting efficiency scores by each expert and generating total efficiency through weighting and fusion, and attributing efficiency loss to corresponding element dimensions and outputting diagnosis results when the expert weight exceeds a threshold value. The invention realizes the span from general division to accurate diagnosis, can clearly define the dominant factors of efficiency fluctuation, solves the problems of high index coupling degree and difficult positioning of the efficiency short plate in the traditional evaluation method, and has high interpretability and situation self-adaptive capacity.

Inventors

  • FENG WEIJIE
  • WANG XINYU
  • CHENG ZHIYONG

Assignees

  • 合肥工业大学

Dates

Publication Date
20260512
Application Date
20260313

Claims (10)

  1. 1. The team cognitive effectiveness decoupling evaluation method based on the mixed expert network is characterized by comprising the following steps of: s1, acquiring evaluation input data of a target team, wherein the evaluation input data comprises deep state characterization vectors and dynamic collaborative network feature vectors of individual intelligent agents; s2, decoupling evaluation input data into individual element feature vectors, group element feature vectors and task element feature vectors, wherein the individual element feature vectors comprise features of psychological dimension and behavioral dimension, the group element feature vectors comprise features of structural dimension and interaction dimension, and the task element feature vectors comprise features of task targets and task difficulties; S3, constructing a hybrid expert network model, wherein the model comprises a gate control network and three expert sub-networks, and the three expert sub-networks are respectively configured into an individual efficiency expert network, a group efficiency expert network and a task efficiency expert network and are respectively used for processing individual element feature vectors, group element feature vectors and task element feature vectors; s4, inputting the real-time team state vector into a gating network, wherein the gating network passes through Function dynamic calculation of three expert subnetworks at the present moment Individual performance weighting values of (2) Group performance weight And task performance weight value ; S5, each expert sub-network outputs corresponding efficiency scores according to the input feature vectors 、 And And generating a team total cognitive effectiveness value according to the weighted fusion of the weight values ; S6, when the weight value of any expert sub-network exceeds a preset threshold value And when the performance loss at the current moment is attributed to the element dimension corresponding to the expert sub-network, and the attributed diagnosis result is output.
  2. 2. The hybrid expert network-based team cognitive performance decoupling assessment method of claim 1, wherein step S1 specifically comprises: S11, acquiring an internal cognition state vector of an individual intelligent agent at each time step, wherein the internal cognition state vector at least comprises fatigue and anxiety values; S12, collecting original multi-source interaction data, wherein the original multi-source interaction data at least comprise a communication text log, a behavior operation log and task parameters; S13, performing interpolation alignment based on time stamps on the internal cognitive state vector, the edge attribute feature vector and the original multisource interaction data, performing up-sampling or high-frequency behavior log aggregation coding on the low-frequency state data, uniformly mapping the low-frequency state data or the high-frequency behavior log aggregation coding on the same time axis, and constructing an evaluation input tensor 。
  3. 3. The hybrid expert network-based team cognitive performance decoupling assessment method of claim 1, wherein in step S2: Individual element feature vector By mapping functions Generation of wherein A vector is characterized for the deep state of the node, As a result of the physiological characteristic data, And A learnable weight matrix and bias terms mapped for individual elements, Representing vector stitching; group element feature vector Pooling aggregation operations on graph structures of dynamic collaboration networks Generation of wherein In order for the edge feature to be co-operative, And Respectively a pooling function and an attention mechanism function; task element feature vector By encoding prior task parameters and real-time task logs Generation of wherein As a function of the environmental parameters, For the task a priori data, Is an environmental encoder.
  4. 4. The hybrid expert network-based team cognitive performance decoupling assessment method of claim 1, wherein constructing a hybrid expert network model in step S3 further comprises: Introducing a cognitive context learning mechanism into the model, and taking the portrait attribute and the historical behavior sequence of the individual as a cognitive memory input model; By passing through And the layer carries out association matching on the current real-time state characteristics and the cognitive memory to form a cognitive chain, so that the model is combined with the history experience to judge the rationality of the current behavior.
  5. 5. The hybrid expert network-based team cognitive performance decoupling evaluation method of claim 1, wherein the specific formula for the dynamic calculation of the weight value by the gating network in step S4 is: ; Wherein the method comprises the steps of As a real-time overall state vector, And For corresponding private sub-networks in a gating network Is provided with a learnable weight matrix and bias terms, Respectively represent individuals Group of people And tasks Three expert subnetworks.
  6. 6. The method for estimating the decoupling of the cognitive performance of the team based on the hybrid expert network according to claim 1, wherein the specific formula for generating the total cognitive performance value of the team by weighting and fusing in the step S5 is as follows: ; Wherein, the Is the first The nonlinear transformation function of the individual expert subnetworks, Inputting for the corresponding decoupling characteristics; assigning the team total cognitive efficacy value Conversion of each sub-performance value into a task execution performance index by a nonlinear mapping function Index of collaborative performance And a decision efficacy index 。
  7. 7. The hybrid expert network-based team cognitive performance decoupling assessment method of claim 1, further comprising, after step S6: S7, carrying out prior evaluation based on prior input data before the task starts, and outputting efficacy prediction scores, carrying out dynamic evaluation based on data updated in real time in the task execution process, introducing a dynamic baseline correction mechanism, and calculating deviation between a real-time state and a preset dynamic baseline ; S8, correcting the efficacy score according to the deviation value to generate a real-time efficacy score Wherein In order to penalize the coefficients, To tolerate a threshold.
  8. 8. The hybrid expert network-based team cognitive performance decoupling assessment method of claim 7, wherein said dynamic baseline deviation The calculation formula of (2) is as follows: ; Wherein the method comprises the steps of For the real-time state vector that is actually acquired, For a real-time dynamic baseline calculated from the current task context, Is a covariance matrix.
  9. 9. A hybrid expert network-based team cognitive efficacy decoupling assessment system for implementing the method of any of the preceding claims 1-8, comprising: The data acquisition and alignment module is configured to acquire evaluation input data of a target team, wherein the evaluation input data comprises deep state characterization vectors and dynamic collaboration network feature vectors of individual intelligent agents, and performs space-time alignment processing on multi-source data; The element decoupling module is connected with the data acquisition and alignment module and is configured to decouple the evaluation input data into individual element feature vectors, group element feature vectors and task element feature vectors; The mixed expert network evaluation module is connected with the element decoupling module and comprises a gating network and three expert sub-networks, wherein the three expert sub-networks are respectively configured into an individual performance expert network, a group performance expert network and a task performance expert network, and are respectively connected with the element decoupling module to receive corresponding feature vectors; A dynamic weight distribution module connected to the gating network and configured to receive the real-time team status vector by Dynamically calculating weight values of the three expert sub-networks at the current moment by using a function; The performance fusion and attribution module is connected with the three expert sub-networks and the dynamic weight distribution module, and is configured to generate a team total cognitive performance value by weighting and fusing the performance scores output by the expert sub-networks according to the weight values, and attributing the performance loss at the current moment to the element dimension corresponding to the expert sub-network when the weight value of any expert sub-network exceeds a preset threshold value, and outputting attribution diagnosis results.
  10. 10. A computer readable storage medium having stored thereon a computer program which when executed by a processor implements the steps of the hybrid expert network based team cognitive efficacy decoupling assessment method of any of claims 1 to 8.

Description

Team cognitive efficiency decoupling evaluation method, system and storage medium based on hybrid expert network Technical Field The invention relates to the technical field of artificial intelligence and team management, in particular to a team cognitive effectiveness decoupling evaluation method, a system and a computer readable storage medium based on a hybrid expert network. Background In complex task scenarios with high risk and high dynamics, such as disaster emergency rescue, emergency surgery, special joint actions, etc., a team is the smallest core unit to perform tasks. The cognition synchronicity among team members, the fluency of interaction cooperation and the psychological toughness for coping with sudden conditions directly determine success and failure and efficiency of tasks. Unlike standardized industrial equipment, human teams are a typical complex adaptive system whose cognitive states have highly non-linear, emerging and dynamic evolving characteristics, and minor psychological fluctuations or communication misunderstandings can be amplified in cascade in team networks, with serious consequences. Therefore, scientific, objective and real-time evaluation is carried out on the cognitive performance of the team, and the method has important significance for improving the success rate of the task. However, the analysis and management of team awareness now faces significant technical bottlenecks. The traditional analysis method mainly depends on expert observation, post interview or subjective questionnaire, and the method is not only high in subjectivity, but also often lags behind a task process, objective multi-source data such as interaction logs, behavior tracks and the like cannot be fused in real time, and therefore real psychological and behavioral responses of members in a stress state are difficult to accurately capture. In addition, most of the existing modeling methods are static snapshot analysis or simulation based on simple rules, and it is difficult to reproduce the dynamic recombination processes of cognitive infection, emotion resonance and collaboration relations generated by team members under task pressure. More serious, the existing evaluation system often mixes various factors such as individual state, group cooperation, task constraint and the like together, and lacks an effective decoupling means, so that when the efficiency is reduced, the problem that the accurate positioning is 'people', the problem of 'cooperation' or the problem of 'task difficulty' is difficult to accurately position, and attribution support cannot be provided for decision making. For example, chinese patent application CN104933530a discloses a system for evaluating the real-time control performance of an empty pipe, which uses an index evaluation module and a quantitative evaluation model to calculate a comprehensive evaluation score, but the scheme adopts a fixed weight weighting summation mode, which cannot adapt to the dynamic change of the weights of each element in the task process. In addition, chinese patent application CN112700158a discloses an algorithm performance evaluation method based on a multidimensional model, which sets weight coefficients of various indexes and then performs weighted summation, and also cannot dynamically adjust an evaluation focus according to a real-time situation. Although the multi-source data interaction analysis technology is mature, the system is blank in how to efficiently convert massive heterogeneous data into an interpretable hierarchical performance index system, and an adaptive evaluation architecture capable of adapting to different task situations is lacking. Therefore, there is an urgent need in the industry for a team performance evaluation solution that can implement complex cognitive influence factor structural decoupling, support dynamic adaptive evaluation, and have interpretable attribution capability. Disclosure of Invention The method aims to solve the technical problems that in the traditional team performance evaluation method, index coupling degree is high, a performance short board is difficult to position, a static weighting model cannot adapt to a dynamic situation, and explanatory attribution capability is lacking. In order to achieve the above objective, in a first aspect, the present invention provides a team cognitive effectiveness decoupling evaluation method based on a hybrid expert network, comprising the following steps: s1, acquiring evaluation input data of a target team, wherein the evaluation input data comprises deep state characterization vectors and dynamic collaborative network feature vectors of individual intelligent agents; s2, decoupling evaluation input data into individual element feature vectors, group element feature vectors and task element feature vectors, wherein the individual element feature vectors comprise features of psychological dimension and behavioral dimension, the group element feature vector