Short-period multi-round situational awareness reliability evaluation method considering anchoring effect and confirmation deviation

Abstract

The invention provides a short-period multi-round situational awareness reliability evaluation method considering anchoring effect and confirmation deviation, which comprises the following steps of: 1, acquiring relevant knowledge of abnormal states, analyzing possible reasons of the current abnormal states, and determining a corresponding solution of each possible reason; 2, qualitative modeling: considering the dynamic nature of the short-period multi-turn SA, the anchoring effect and turn correlation caused by confirmation deviation, and modeling a short-period multi-turn SA establishment process by using a dynamic Bayesian network; step 3, quantitative modeling: on the basis of constructing the dynamic Bayesian network structure, determining node parameters in the dynamic Bayesian network; and step 4, short-period multi-turn SA reliability evaluation: evaluating the reliability of the short-period multi-turn SA according to the evidence and the determination of the conditional probability table of each node. According to the method, the problem that the round correlation caused by the anchoring effect and the confirmation deviation is not considered in the conventional short-period multi-round SA reliability evaluation is solved.

Description

Technical Field

[0001] The present invention provides a short-term multi-round situation awareness (SA) reliability assessment method considering anchoring effect and confirmation bias, analyzes and describes the SA round correlation caused by anchoring effect and confirmation bias, and their mutual promotion and adverse effects that hinder the establishment of correct SA, and models the establishment of dynamic short-term multi-round SA and evaluates its reliability, belonging to the field of SA reliability modeling and assessment. Background Art

[0002] With the development of automation technology, the role of the operator has changed from manipulator to supervisor, planner and solver of abnormal state, among which SA is crucial to decision-making. Endsley, a scholar from the Department of Industrial Engineering at Texas Tech University, describes SA as: perception of environmental elements, understanding of their significance, and prediction of their recent state. The present invention adopts this definition and defines SA under abnormal state as the perception of current abnormal related information, understanding of the cause of the abnormality, and prediction of the utility of the selected solution. Establishing SA under abnormal state requires testing and adjusting solutions in multiple rounds of interaction. SA failure is the main threat to system safety. According to Endsley's statistics, 88% of aviation accidents are related to the loss of SA under abnormal state. For example, the Boeing 737-8 (MAX) accident and the Air France 447 crash were both related to the pilot's SA failure. Therefore, effective and accurate reliability evaluation of multi-round SA is crucial to improving the safety level of the human-machine system.

[0003] In abnormal states, the establishment of SA is characterized by multi-round interactions with short time intervals. This paper mainly studies this short-cycle multi-round SA. It is different from long-cycle multi-round SA, which relies on experience and knowledge accumulation. In the establishment process of short-cycle multi-round SA, the operator's knowledge structure is unlikely to change over time. The recognition of feedback from previous actions is the key to correcting erroneous SA. Feedback can trigger more useful evidence, support the exclusion of erroneous options, and encourage operators to consider more possible system states. However, cognitive biases including anchoring effect and confirmation bias can hinder the correction process. Specifically, the anchoring effect will bias the operator's judgment of the system state towards the previous view, resulting in insufficient adjustment of the belief in the erroneous SA in the current round; confirmation bias draws attention to information that confirms the previous view, resulting in the neglect of key information in the current round. For example, in the Boeing 737-8 (MAX) accident, these biases led to the pilot's failure to correct SA in time. The anchoring effect and confirmation bias can promote each other, strengthen the erroneous SA belief, and SA is not isolated in each round. Therefore, it is important to incorporate the round correlation caused by the anchoring effect and confirmation bias in the reliability assessment of short-cycle multi-round SA.

[0004] Quantitative methods for SA reliability assessment have been widely studied. Some studies focus on quantifying the impact of static and dynamic performance-forming factors on SA reliability by extending human factors analysis methods. Other studies conduct SA reliability assessment by modeling the dynamic establishment process of SA. For example, Kim and Seong, scholars from the Department of Nuclear and Quantum Engineering at the Korea Advanced Institute of Science and Technology, proposed a SA analysis method for nuclear power plant operators based on mental models. They used this method to further evaluate the impact of attention, memory, and mental models on SA. Li et al., scholars from the School of Artificial Intelligence and Automation at Huazhong University of Science and Technology, used dynamic Bayesian networks to quantitatively evaluate SA failures to consider the dynamics and time correlation of cognitive states. In addition, Stroeve et al., scientists at the Netherlands Aerospace Center, proposed a mathematical model to describe the establishment and update process of SA. They incorporated the model into a multi-agent modeling method to evaluate the reliability of SA. However, these methods are not suitable for evaluating the reliability of short-term multi-round SAs because they do not consider feedback from actions.

[0005] To address this problem, more and more studies have turned their attention to the feedback effect to evaluate the reliability of dynamically updated short-term multi-round SA. For example, Zhao and Smidts, scholars from the Department of Mechanical and Aerospace Engineering at Ohio State University, used Bayesian reasoning to model cognitive processes. As the number of rounds increases, feedback will activate more system-related knowledge. They proposed an "imperfect" belief propagation algorithm to describe the uncertainty in knowledge retrieval and estimated the reliability of short-term multi-round SA through Monte Carlo simulation. Other studies focus on the situation where new evidence triggered by feedback is used for SA updating. Zhang et al., scholars from the School of Civil Aviation of Nanjing University of Aeronautics and Astronautics, used the visual and auditory warnings caused by the previous round of operations as new evidence to evaluate the reliability of the pilot's short-term multi-round SA based on Bayesian networks. The latest version of the "Information, Decision and Action in the Crew Environment" model has developed a new semantic network-based reasoning module that can simulate the operator's knowledge-based reasoning to update the SA after action feedback.

[0006] Although the above methods have successfully evaluated the reliability of short-term multi-round SA, they ignore the round correlation caused by anchoring effect and confirmation bias. Empirical and mechanism studies provide theoretical guidance for overcoming this problem. They qualitatively illustrate the impact of correlation caused by confirmation bias on information perception and the impact of correlation caused by anchoring effect on belief updating. However, there is still a gap in the methods used to quantify and model these round correlation relationships in quantitative short-term multi-round SA reliability assessment. Summary of the invention

[0007] (1) Purpose:

[0008] The present invention provides a short-term multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias. It focuses on the round correlation caused by anchoring effect and confirmation bias based on the ideas and characteristics of dynamic Bayesian networks, and proposes a SA reliability modeling method and modeling process based on dynamic Bayesian networks for the short-term multi-round SA establishment process. This modeling method plays an important role in assessing the reliability of SA in risk scenarios and finding weak links in the design of human-machine systems.

[0009] (2) Technical solution:

[0010] The present invention provides a short-period multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias, including four steps of obtaining relevant knowledge of abnormal states, qualitative modeling, quantitative modeling and short-period multi-round SA reliability assessment. The method uses a dynamic Bayesian network to model the dynamic establishment of short-period multi-round SA. The dynamic Bayesian network uses a network within a single time slice to describe cognitive activities within a single interaction round. The round correlation caused by the anchoring effect and confirmation bias is explicitly modeled as a connection between adjacent time slices. Then, the short-period multi-round SA reliability is evaluated based on the dynamic Bayesian network. This method plays an important role in improving the reliability of personnel SA and enhancing the safety of human-machine systems.

[0011] The present invention provides a short-term multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias, and the process is as follows: Figure 1 As shown, the specific steps are as follows:

[0012] Step 1: Obtain relevant knowledge about the abnormal state, analyze the possible causes of the current abnormal state, and determine the corresponding solutions for each possible cause;

[0013] Step 2: Qualitative modeling, considering the dynamics of short-term multi-round SA and the round correlation caused by anchoring effect and confirmation bias, and using dynamic Bayesian network to model the short-term multi-round SA establishment process;

[0014] Step 3: Quantitative modeling: on the basis of constructing the dynamic Bayesian network structure, determine the node parameters in the dynamic Bayesian network;

[0015] Step 4: Short-term multi-round SA reliability assessment: Based on the evidence and the conditional probability table of each node, the reliability of short-term multi-round SA is assessed;

[0016] Through the above steps, the dynamic Bayesian network is used to model the dynamic establishment of short-term multi-round SA, and the network within a single time slice is used to describe the cognitive activities within a single interaction round; the round correlation caused by the anchoring effect and confirmation bias is explicitly modeled as the connection between adjacent time slices; the determination of the conditional probability table is guided by the quantitative method of round correlation, and finally the reliability of short-term multi-round SA is evaluated based on the dynamic Bayesian network, which solves the problem that the round correlation caused by the anchoring effect and confirmation bias was not considered in the previous reliability evaluation of short-term multi-round SA.

[0017] Among them, the "obtaining relevant knowledge of abnormal states" described in step 1 is the basis for the construction of the entire dynamic Bayesian network. The main purpose is to obtain system fault knowledge and human-computer interaction task knowledge for the research object, and identify the possible causes of the current abnormal state and the subsystem / component failures that cause these causes based on the system fault knowledge and human-computer interaction task knowledge, as well as the corresponding solutions for each possible cause, including the following:

[0018] The construction of dynamic Bayesian networks depends on the knowledge related to abnormal states. This knowledge can be divided into system fault knowledge and human-computer interaction task knowledge. System fault knowledge includes the possible causes of the current abnormal state and the subsystem / component failures that cause these causes. Possible fault causes have similar fault symptoms, which may confuse the operator with the actual cause and other causes. Operators often infer the actual cause based on their cognition of the fault states of these subsystems / components. Human-computer interaction task knowledge contains the information needed to perceive the actual cause and the corresponding solution for each possible cause. This information is related to the cause of the abnormality or can indicate the fault state of the subsystem / component.

[0019] The “qualitative modeling” described in step 2 considers the dynamics and round correlation of short-term multi-round SA, and uses a dynamic Bayesian network to establish a short-term multi-round SA model under abnormal conditions, including the following steps:

[0020] Step 1): Identify key nodes in the dynamic Bayesian network

[0021] like Figure 2 As shown, the method proposed in the present invention uses a set of nodes to model the perception of each key information, which include "evidence" nodes, "attention" nodes and "perception" nodes. Different SAs correspond to different "action" nodes, and their consequences are represented by "utility" nodes. In addition, "anchor" nodes are identified in the dynamic Bayesian network to reflect the round correlation caused by the anchoring effect and confirmation bias. These nodes will be expanded and given specific meanings when applied to actual situations. However, the scope of node identification must be limited to prevent the inclusion of information and knowledge that is not related to abnormal states;

[0022] Step 2): Determine the status of all nodes

[0023] This sub-step is to identify the state of the node according to its possible modes. Usually, the state is defined according to the possible operation state of the node related to the system function, such as failure or success.

[0024] Step 3): Determine correlation within a single round

[0025] The correlations between nodes are established on a single time slice of the dynamic Bayesian network to describe the cognitive activities in a single interaction round, including the establishment of SA and utility evaluation. The establishment of correlation arcs in the SA establishment process starts with observing the relevant nodes, then the "system state" nodes, and finally the corresponding "action" nodes. In utility evaluation, the development of correlation arcs is similar to the SA establishment process. The above nodes and arcs describe the process of operators understanding and handling abnormal states in a single round of interaction;

[0026] Step 4): Establish correlation between adjacent rounds

[0027] This sub-step aims to establish connections between different DNB slices, which reflects round correlations. The most significant round correlations are caused by anchoring effects and confirmation bias.

[0028] The connection between the "anchor" node and the "SA" node is used to reflect the correlation caused by the anchoring effect. The correlation caused by the confirmation bias can be described by the connection between the "anchor" node of the previous round and the "attention" node of the current round. In addition, the perception of discrete information (such as alarms) is related to their perception state in the previous round. Moreover, there is a correlation between the SA of adjacent rounds. Figure 2 In Fig. 2, these dependencies are represented by arrows from the “sensing” node and the “SA” node pointing to themselves. The above nodes and connections are combined with a Bayesian network monolith to describe the establishment of short-term multi-round SA.

[0029] The “quantitative modeling” described in step 3 is to determine the parameters in the dynamic Bayesian network on the basis of constructing the dynamic Bayesian network structure, including the parameter quantification of nodes with round correlation influence and nodes without round correlation influence, including the following steps:

[0030] Step 1): Determine the status of the "evidence" node in each round

[0031] The "evidence" node represents the information of the updated SA perceived by the operator in each interaction round. The identification of its state depends on the trend of observable variables. These observable variables include discrete information (such as alarms) and continuous information (such as system status such as aircraft altitude). The use of continuous information in dynamic Bayesian networks must be discretized. The discretization results also reflect the operator's subjective estimates of these continuous variables.

[0032] Step 2): Determine the conditional probability table of nodes without round correlation

[0033] The goal of this sub-step is to determine conditional probability tables for nodes related to system states without considering correlations caused by anchoring effects and confirmation bias. Conditional probability tables represent the operator's knowledge and mental model of abnormal states and can be determined based on their previous training and experience. If the operator is well trained and has sufficient knowledge, these conditional probability tables can be obtained from the system fault logic or standard operating manual. In addition, questionnaires and expert judgment of operators are also effective methods.

[0034] Step 3): Quantify correlations caused by anchoring and confirmation bias

[0035] This sub-step aims to determine the conditional probability table of nodes related to anchoring effect and confirmation bias in the dynamic Bayesian network. To parameterize the nodes related to the anchoring effect (i.e., the "SA" nodes), it is first necessary to clarify the initial evaluation of the evidence set X. It represents the causal strength of the evidence set supporting the anchor point without considering the influence of the anchoring effect. Therefore, the evaluation can be obtained by referring to the determination of the conditional probability table of the node without cognitive bias. On this basis, the evidence is classified according to the reference point R. Then use equations (1) and (2) to determine the conditional probability table of the "SA" node.

[0036]

[0037] In the formula, and They represent the conditional probability that the SA in round r is consistent with the initial anchor point when the operator believes or does not believe the anchor point in round r-1, respectively; represents the set of perceived evidence; X is the set of evidence; For anchor point; represents the initial causal strength of the evidence set X supporting the anchor point; R is the reference point, whose value can be set to 0.5; α and β (0<α, β<1) are constants reflecting the operator's sensitivity to negative and positive evidence, respectively.

[0038] In addition, it is important to note that the process of establishing short-term multi-round SA is a sequential decision-making process. When the correct SA is established within a certain round, the abnormal state can be successfully resolved. Short-term multi-round SA is only wrong when the accurate SA is not achieved in all the rounds that have occurred.

[0039] In the anchor-based scenario evaluation process, the correlation caused by confirmation bias affects information perception. This influence determines the probability of information being perceived and is represented by the "attention" node in the dynamic Bayesian network model. Therefore, quantifying the correlation caused by confirmation bias is to calculate the conditional probability table of the "attention" node. First, use formula (3) to calculate the salience of information i in the rth round (i.e. Then, the value of information under different anchor states is determined according to equations (4) and (5) (i.e. ). Finally, and Substituting into equation (6) and equation (7), the conditional perception probability of information i can be calculated.

[0040]

[0041] In the formula, represents the salience of each information in round r; and They represent the prominence of the information i based on its rate of change, whether the information i is an alarm signal, and the human-machine interface design; γ c , γ a and γ d are the weights of the three salience parameters.

[0042]

[0043] In the formula, is the value of information i in round r; is the support of information i for the anchor point in the r-1th round.

[0044]

[0045] In the formula, and Indicates whether the operator believes or does not believe the Anchor r-1 The value of information i in round r under the condition of .

[0046]

[0047] In the formula, A ri is the activation of information i in round r; γ and λ are the weights of salience and value.

[0048]

[0049] In the formula, represents the bout correlation caused by confirmation bias; τ and s are parameters that affect the activation threshold and noise on perception, respectively.

[0050] Among them, the "short-term multi-round SA reliability evaluation" described in step 4 is to calculate the joint probability distribution of the dynamic Bayesian network based on the evidence and the conditional probability table of each node, and finally obtain the reliability of the short-term multi-round SA, including the following contents:

[0051] According to the evidence and the determination of the conditional probability table of each node, the dynamic Bayesian network realizes the quantitative modeling of short-term multi-round SA, and its joint probability distribution can be calculated as follows:

[0052]

[0053] In the formula, P(Z 1:Rn ) is the joint distribution; is the node i in the interaction round r; yes The parent node set of may include nodes in round r or round r-1; n is the number of nodes in a single round slice; Rn is the total number of interaction rounds. Therefore, the reliability of short-term multi-round SA can be expressed as:

[0054]

[0055] (3) Efficacy and advantages

[0056] The present invention provides a short-period multi-round SA reliability assessment method based on a dynamic Bayesian network, which can obtain abnormal state-related knowledge of the research object, and identify the possible causes of the current abnormal state and the corresponding subsystem / component failures and corresponding solutions; can consider the round correlation and use the dynamic Bayesian network to model the short-period multi-round SA; can quantify the round correlation caused by the anchoring effect and confirmation bias; can effectively and accurately assess the reliability of the short-period multi-round SA, and improve the safety level of the human-machine system. The short-period multi-round SA reliability assessment method described in the present invention is scientific, highly applicable, and has broad promotion and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is the overall framework of the proposed method.

[0058] Figure 2 Schematic diagram of short-period multi-round SA based on dynamic Bayesian network.

[0059] Figure 3To provide the working principle of maneuverability enhancement system and related systems.

[0060] Figure 4 Dynamic Bayesian network structure for Boeing 737-8 (MAX) accidents.

[0061] Figure 5 It is the reliability of short-term multi-round SA in 20 interaction rounds.

[0062] Figure 6 It is the short-term multi-round SA reliability under different anchoring levels.

[0063] The serial numbers, symbols and codes in the figure are explained as follows:

[0064] Figure 2 and Figure 4 In the figure, the white thin solid oval box is the "evidence" node, the white dotted oval box is the "attention" node, the white thick solid oval box is the "perception" node, the black oval box is the "system state and SA" node, the gray thin solid oval box is the "action" node, the gray dotted oval box is the "utility" node, and the gray thick solid oval box is the "anchor point" node; the thick dotted line represents the correlation caused by the anchoring effect, the thin dotted line represents the correlation caused by the confirmation bias, the thick solid line represents the connection within a single round, and the thin solid line represents the correlation in other rounds.

[0065] Figure 3 The white rectangular box in the middle represents the structure and function of the system; the gray rectangular box represents the visual alarm; and the black rectangular box represents the auditory alarm.

[0066] Figure 5 The middle square icon line and the triangle icon line represent the short-period multi-round SA reliability with and without considering round correlation, respectively; Figure 6 The medium circular icon line, square icon line and triangle icon line represent the short-term multi-round SA reliability at high, medium and low anchoring levels, respectively. DETAILED DESCRIPTION

[0067] The present invention provides a short-term multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias; the method is carried out in the following four steps in sequence: Figure 1 As shown; the present invention takes the Boeing 737-8 (MAX) accident as the background, and aims at the multi-round interaction process between the pilot and the aircraft after the abnormal operation of the maneuvering characteristics enhancement system, adopts the above reliability evaluation method, and models the dynamic establishment of short-period multi-round SA; the model takes into account the round correlation caused by the anchoring effect and confirmation bias, and guides the determination of the conditional probability table through the quantification method of the round correlation to verify the feasibility and accuracy of the present invention.

[0068] The specific implementation plan is described in detail as follows:

[0069] Step 1: Acquire knowledge about abnormalities

[0070] Based on the accident reports, we extracted the system failure knowledge and human-computer interaction task knowledge of the Boeing 737-8 (MAX) accidents. Figure 3 The operating principles of the MCAS and related systems are shown, which represent the knowledge of system failures.

[0071] The knowledge of human-computer interaction tasks includes the information that needs to be perceived and the solutions to the current anomalies. When the transmitted parameters are abnormal, the alarm system will trigger various visual alarms ( Figure 3 Medium grey rectangle) and audible alarm ( Figure 3 These warning and indication parameters can help the pilot determine the actual cause of the current abnormality. For abnormal operation of the maneuvering characteristics augmentation system, the solution is to cut off the automatic trim of the horizontal stabilizer, while for failure of the speed trim system, the solution is to manually operate the angle of attack trim.

[0072] Step 2: Qualitative Modeling

[0073] Step 1): Identify key nodes in the dynamic Bayesian network

[0074] The dynamic Bayesian network identifies the cognitive activities caused by the abnormal operation of the maneuvering characteristics augmentation system and the knowledge of abnormal states as nodes. For information perception, each warning and indication parameter (such as speed and altitude) is represented by a set of observation-related nodes in the dynamic Bayesian network. In addition, the operational knowledge and fault knowledge of the maneuvering characteristics augmentation system related systems are identified as "system-related" nodes and "action" nodes. In each round, the "SA" node representing the pilot's judgment on the cause of the emergency is selected as the target node to reflect the reliability of short-cycle multi-round SA. In addition, inherent nodes (such as "utility" nodes and "anchor" nodes) are set to reflect the SA belief adjustment caused by action feedback.

[0075] Step 2): Determine the status of all nodes

[0076] For this case, the cause of the accident was the abnormal operation of the maneuvering characteristics augmentation system, which may be misjudged by the pilot as a failure of the speed trim system. Therefore, they are identified as two states of the "SA" node, representing correct SA and incorrect SA. Each SA state corresponds to an "action" node. For correct SA and incorrect SA, they are cutting off the automatic trim of the horizontal stabilizer and manually operating the angle of attack trim.

[0077] Table 1 lists the other nodes and their status.

[0078] Table 1 Description of nodes in dynamic Bayesian network

[0079]

[0080]

[0081] Step 3): Determine correlation within a single round

[0082] In a single turn, the correlation between the perception information and systems / subsystems such as the "Air Data Module" can be represented by a single segment of the dynamic Bayesian network, which reflects the pilot's knowledge of the maneuver characteristics augmentation system related systems. In order to study the reliability of short-term multi-turn SA, we focus on the case where the operator anchors the wrong SA, and only consider the utility evaluation of the wrong action that can produce the next turn anchoring.

[0083] Step 4): Establish correlation between adjacent rounds

[0084] The round correlation of short-term multi-round SA caused by anchoring effect and confirmation bias can be expressed by the connection between the “anchor” node and the “SA” node ( Figure 4 The thick dashed line in the figure) and the connection between the “anchor” node and the “attention” node ( Figure 4 The thin dotted line in the figure indicates this. Figure 4 As shown in Figure 2, by combining the connections within and between time slices, the qualitative dynamic Bayesian network model in the case study can be constructed. Figure 4In the figure, the states of the “attention” nodes (A1-A7) and the corresponding “evidence” nodes (terrain warning, “feeling pressure difference” warning, descent rate warning, speed warning, speed inconsistency, altitude warning, altitude inconsistency) jointly affect the states of the “perception” nodes (terrain warning Per, “feeling pressure difference” warning Per, descent rate warning Per, speed warning Per, speed inconsistency Per, altitude warning Per, altitude inconsistency Per). Only when the state of the “evidence” node is occurred and the state of the “attention” node is noticed, the state of the “perception” node is occurred; the “perception” node affects the “system state” node and ultimately affects the state of the “SA” node, and generates the corresponding “action” node according to the state of the “SA” node; the “attention” nodes (A8-A10) and the “evidence” nodes The state of (angle of attack, altitude, speed) determines the state of the "perception" node (angle of attack Per, altitude Per, speed Per), and ultimately affects the state of the "utility" node. The states of the "action" node and the "utility" node determine the state of the "anchor" node. If the "SA" node correctly judges the cause of the system failure, the round ends. If the "SA" node makes an incorrect judgment, it enters the next round. The round correlation caused by the anchoring effect makes the "anchor" node affect the "SA" node of the next round. The round correlation caused by the confirmation bias makes the "anchor" node affect the state of the "attention" node (A1~A7). Other round correlations affect some "perception" nodes and the "SA" node of the next round. If the "SA" node always makes incorrect judgments, it will continue to the next round until an accident occurs.

[0085] Step 3: Quantitative Modeling

[0086] Step 1): Determine the status of the "evidence" node in each round

[0087] The state of the “evidence” node in each interaction round is determined by the measurement data. The visual and auditory alarms are shown in Table 2.

[0088] Table 2 Alarms in each interaction round

[0089]

[0090] Step 2): Determine the conditional probability table of nodes without round correlation

[0091] According to the system's fault logic and expert judgment, the conditional probability table of the "system / subsystem related" node is determined. The conditional probability table with probability logic is selected to encode the pilot's knowledge. Take the conditional probability table of the "SA" node in the first round as an example, as shown in Table 3.

[0092] Table 3 Conditional probability table of "SA" node (r=1)

[0093]

[0094] S: success, F: failure

[0095] Step 3): Quantify correlations caused by anchoring and confirmation bias

[0096] The pilot's reasoning process for establishing SA is supported by new evidence and guided by anchor points. The pilot makes subjective adjustments to the initial assessment of the strength of evidence due to the anchoring effect. Considering the anchoring effect and the influence of the anchor point in the previous SA, the conditional probability table of the "SA" node can be adjusted, as shown in Table 4.

[0097] Table 4 only gives the conditional probability when SA fails in the r-1th round, that is, SA[r-1] = FOS. When SA is correct in the r-1th round, the conditional probability is

[0098] Table 4 Conditional probability table of the “SA” node (r>1) with correlation caused by anchoring effect

[0099]

[0100] S: Success, F: Failure; Initial anchor point = speed trim system failure

[0101] In each round of interaction, the pilot in the cockpit of the aircraft will receive different evidence. However, due to the existence of confirmation bias, the pilot may not be able to perceive all the evidence. The parameters for calculating the conditional probability of each "attention" node related to confirmation bias are shown in Tables 5 and 6.

[0102] Table 5 Parameters quantifying correlations caused by confirmation bias

[0103]

[0104]

[0105] Table 6 Salience and value of evidence related to confirmation bias

[0106]

[0107] Step 4: Short-term multi-round SA reliability assessment

[0108] After determining the parameters of all nodes, the reliability of short-period multi-round SA in the case is calculated using equations (8) and (9).

[0109] In this case, the reliability of short-term multi-round SA with and without considering the correlation of SA rounds is as follows: Figure 5As shown in the figure, it can be found that in the initial rounds, the probability of establishing a correct SA is very small. This is mainly due to insufficient training on the knowledge of the maneuvering characteristics augmentation system, unclear description of the abnormal checklist, and missing key information. Moreover, with the appearance of the key information of the "feeling pressure difference" alarm in the 7th round, the reliability of SA has undergone a mutation, regardless of whether the anchoring effect and confirmation bias are considered. Obviously, without considering the correlation of the rounds, the degree of mutation and the reliability of SA are much higher.

[0110] In addition, the reliability of short-term multi-round SA under other anchoring levels, including high anchoring effect and low anchoring effect, was evaluated and compared. Figure 6 The conclusions at other anchoring levels are similar to those at the medium anchoring level. For example, the addition of the "feeling pressure difference" alarm caused a mutation, which reduced the impact of the anchoring effect and confirmation bias. The difference is that as the anchoring level decreases, the degree of mutation and the reliability of short-term multi-round SA both increase significantly.

Claims

1. A short-term multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias, characterized by: Here are the steps: Step 1: Obtain relevant knowledge about the abnormal state, analyze the possible causes of the current abnormal state, and determine the corresponding solutions for each possible cause; Step 2: Qualitative modeling, considering the dynamics of short-term multi-round SA and the round correlation caused by anchoring effect and confirmation bias, and using dynamic Bayesian network to model the short-term multi-round SA establishment process; Step 3: Quantitative modeling: on the basis of constructing the dynamic Bayesian network structure, determine the node parameters in the dynamic Bayesian network; Step 4: Short-term multi-round SA reliability evaluation, based on the evidence and the conditional probability table of each node, the reliability of short-term multi-round SA is evaluated.

2. According to claim 1, a short-term multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias is characterized by: In step 1, the construction of the dynamic Bayesian network depends on the knowledge about the abnormal states; It is divided into system fault knowledge and human-computer interaction task knowledge; system fault knowledge includes the possible causes of the current abnormal state and the subsystem / component failures that cause these causes; human-computer interaction task knowledge contains the information needed to perceive the actual cause and the corresponding solution for each possible cause; this information is related to the cause of the abnormality, or indicates the fault state of the subsystem / component.

3. The short-term multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias according to claim 1, characterized in that: In step 2, a dynamic Bayesian network is used to model the short-term multi-round SA establishment process, including the following steps: Step 2.1): Identify key nodes in the dynamic Bayesian network The perception of each key information is modeled using a set of nodes, including "evidence" nodes, "attention" nodes, and "perception" nodes; different SAs correspond to different "action" nodes, and their consequences are represented by "utility" nodes; in addition, "anchor" nodes are identified in the dynamic Bayesian network to reflect the round correlation caused by the anchoring effect and confirmation bias; Step 2.2): Determine the status of all nodes Identify the state of a node based on its possible modes, and define its state based on the possible operating states of the node related to the system function, including failure or success; Step 2.3): Determine correlation within a single round Establish the correlation between nodes on a single time slice of the dynamic Bayesian network to describe the cognitive activities in a single interaction round, including the establishment of SA and utility evaluation; the establishment of the correlation arc in the SA establishment process starts with observing the relevant node, then the "system state" node, and finally the corresponding "action" node; Step 2.4): Establish correlation between adjacent rounds Establishing connections between different DNB slices reflects round correlations; the most significant round correlations are caused by anchoring effects and confirmation bias.

4. The short-term multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias according to claim 3 is characterized by: The connection between the "anchor" node and the "SA" node is used to reflect the correlation caused by the anchoring effect; the correlation caused by the confirmation bias is described by the connection between the "anchor" node of the previous round and the "attention" node of the current round; in addition, the perception of discrete information is related to their perceptual state in the previous round; and there is a correlation between the SAs of adjacent rounds.

5. The short-term multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias according to claim 1, characterized in that: In step 3, the node parameters in the dynamic Bayesian network are determined, including the parameter quantification of nodes with and without round correlation effects. The steps are as follows: Step 3.1): Determine the status of the "evidence" node in each round The "evidence" node represents the information perceived by the operator to update the SA in each interaction round; The identification of its state depends on the trend of observable variables; These observable variables include discrete information and continuous information; using continuous information in a dynamic Bayesian network requires discretization; Step 3.2): Determine the conditional probability table of nodes without round correlation Determine conditional probability tables for nodes related to system states without considering correlations caused by anchoring effects and confirmation bias; conditional probability tables represent the operator's knowledge and mental model of abnormal states, determined based on their previous training and experience; if the operator is well trained and has sufficient knowledge, these conditional probability tables are obtained from the system fault logic or standard operating manual; Step 3.3): Quantify correlations caused by anchoring and confirmation bias Determine the conditional probability table of nodes related to anchoring effect and confirmation bias in the dynamic Bayesian network; to parameterize the anchoring effect related nodes, namely the "SA" nodes, it is necessary to clarify the initial evaluation of the evidence set X; it represents the causal strength of the evidence set supporting the anchor point without considering the influence of the anchoring effect; refer to the determination of the conditional probability table of the node without cognitive bias to obtain the evaluation.

6. A short-period multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias according to claim 5, characterized in that: Classify the evidence according to the reference point R; then use equations (1) and (2) to determine the conditional probability table of the "SA" node; In the formula, and They represent the conditional probability that the SA in round r is consistent with the initial anchor point when the operator believes or does not believe the anchor point in round r-1, respectively; represents the set of perceived evidence; X is the set of evidence; For anchor point; represents the initial causal strength of the evidence set X in supporting the anchor point; R is the reference point, whose value is set to 0.5; α and β, 0<α, β<1, are constants reflecting the operator's sensitivity to negative and positive evidence, respectively.

7. A short-period multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias according to claim 5 or 6, characterized in that: The process of establishing short-cycle multi-round SA is a sequential decision-making process; when the correct SA is established within a certain round, the abnormal state is successfully resolved; the short-cycle multi-round SA is wrong only when the accurate SA is not achieved in all the rounds that have occurred.

8. The short-term multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias according to claim 7, characterized in that: In the anchor-based scenario evaluation process, the correlation caused by confirmation bias affects information perception; this influence determines the probability of information being perceived and is represented by the "attention" node in the dynamic Bayesian network model; therefore, quantifying the correlation caused by confirmation bias is to calculate the conditional probability table of the "attention" node; first, use formula (3) to calculate the salience of information i in the rth round, that is, Then, the value of information under different anchor point states is determined according to equations (4) and (5), namely: Finally, and Substituting into equation (6) and equation (7) we can calculate the conditional perception probability of information i; In the formula, represents the salience of each information in round r; and They represent the prominence of the information i based on its rate of change, whether the information i is an alarm signal, and the human-machine interface design; γ c , γ a and γ d are the weights of the three salience parameters; In the formula, is the value of information i in round r; is the support of information i for the anchor point in round r-1; In the formula, and Indicates whether the operator believes or does not believe the Anchor r-1 The value of information i in round r under the condition of In the formula, A ri is the activation of information i in round r; γ and λ are the weights of salience and value; In the formula, represents the bout correlation caused by confirmation bias; τ and s are parameters that affect the activation threshold and noise on perception, respectively.

9. The short-term multi-round situational awareness reliability assessment method considering anchoring effect and confirmation bias according to claim 1, characterized in that: In step 4, based on the evidence and the conditional probability table of each node, the reliability of the short-period multi-round SA is evaluated, including the following: Based on the evidence and the conditional probability table of each node, the dynamic Bayesian network realizes the quantitative modeling of short-term multi-round SA, and its joint probability distribution is calculated as follows: In the formula, P(Z 1:Rn ) is the joint distribution; is the node i in the interaction round r; yes The parent node set of , which may contain nodes in the rth round or the r-1th round; n is the number of nodes in a single round slice; Rn is the total number of interaction rounds; therefore, the reliability of short-term multi-round SA is expressed as:

Images