Multi-expert united belief rule base complex system health state assessment method

Through the multi-expert joint confidence rule base and multiple whale optimization algorithm, the problem of inconsistent expert cognitive abilities and local optimal optimization algorithm in the health status assessment of complex systems is solved, and the accurate assessment of health status and the stability of the model are achieved.

CN118916653APending Publication Date: 2024-11-08HARBIN NORMAL UNIVERSITY
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410560342.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-05-08
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In the health status assessment of complex systems, there are inconsistent expert cognitive abilities, inconsistent information and expert silos, and existing optimization algorithms are prone to falling into local optimization, which affects the effectiveness and accuracy of the model.

Method used

The multi-expert joint confidence rule base method is adopted to establish a complex system health status evaluation model through the fusion of multi-expert knowledge and multiple whale group optimization algorithms, integrate the knowledge of multiple experts and optimize model parameters to ensure the accuracy and stability of the evaluation results.

Benefits of technology

It realizes accurate assessment of the health status of complex systems, solves the problems of incomplete knowledge of single experts and expert silos, improves the effectiveness and accuracy of the model, reduces the impact of subjectivity, and ensures that the evaluation results are consistent with reality.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118916653A_ABST
    Figure CN118916653A_ABST
Patent Text Reader

Abstract

The invention relates to the field of evaluation, in particular to a multi-expert united belief rule base complex system health status evaluation method, which comprises the following steps of: constructing a multi-expert knowledge fusion method system; establishing a complex system health state evaluation model based on a multi-expert joint confidence rule base; establishing a multi-population whale optimization algorithm suitable for a multi-expert joint confidence rule base; according to the complex system health state assessment method based on the multi-expert joint confidence rule base, a complex system equipment health state assessment model is established by utilizing a multi-expert joint execution state model; a multi-population whale optimization algorithm suitable for a multi-expert belief rule base is adopted to optimize model parameters, the effectiveness of the method is verified through lithium ion battery examples, multi-expert knowledge can be fully utilized to evaluate the health state of a complex system, the operation condition of large equipment is analyzed, and the reliability of the system is improved. And the method has good engineering application value.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of assessment, and in particular to a method for assessing the health status of a complex system using a multi-expert joint confidence rule base. Background Art

[0002] Complex systems are widely used in aerospace, rail transportation, power systems and other fields. As the system's cumulative working time increases, the system may fail, which in turn affects the safety and reliability of the system. Since these systems have multiple dimensions, multiple units, and complex coupling between processes, which makes them different from simple systems in terms of mechanism, function and behavior, it is more complicated to evaluate their health status. Therefore, how to accurately and reliably evaluate the health status of complex systems is a crucial issue.

[0003] In the current research on health status assessment methods, representative ones include model-driven health assessment methods, data-driven health assessment methods, knowledge-based health assessment methods, and semi-quantitative health assessment methods. Based on the model-driven health assessment method, the physical model is designed by experts in related fields through in-depth understanding of the nature of equipment operation and degradation. However, due to the complex operating mechanisms involved in the production process, it is challenging to develop a suitable physical model. First of all, many assumptions are often made when building such a model. These assumptions simplify the model but sacrifice the fidelity of the actual model. These modeling methods have high economic costs, low applicability, and high modeling complexity. Data-driven methods are divided into machine learning and statistical models. The system health status defined by this type of method evaluates the health status of the system through the fault threshold, but their modeling process is not transparent and requires a large number of samples, so the rationality of the data-driven model results cannot be completely convincing. Knowledge-based methods evaluate the health status of complex systems based on the knowledge provided by experts through mechanism analysis, but because complete and accurate complex system knowledge is difficult to extract and is limited by the uncertainty of expert knowledge, the accuracy of the health status assessment results will eventually be affected. In contrast, semi-quantitative methods, which use knowledge and data to model complex systems, are more intuitive and effective in assessing the health status of complex systems. They have been widely used in health status assessment, fault detection, risk assessment, and other aspects.

[0004] As a semi-quantitative method, the confidence rule base (BRB) uses the knowledge of experts to construct initial confidence rules, and then uses observational data to adjust the parameters of the confidence rules. The confidence rule base (BRB) has attracted widespread attention due to its nonlinear modeling capabilities and strong advantages in handling uncertainty. It is developed based on the evidential reasoning (ER) algorithm, fuzzy theory and IF-THEN rules. BRB has good uncertainty information processing capabilities and can accurately model various uncertainty information. It is an interpretable modeling method that can achieve high accuracy even with small sample data. These characteristics make BRB have good prospects in the health status assessment of complex systems. Its expert knowledge base can effectively contribute to the modeling of complex systems and can preliminarily solve the problem that complex systems cannot meet the actual system requirements by relying solely on subjective judgment.

[0005] However, in general BRB models, it is assumed that the expert's cognitive ability of the system mechanism meets the model accuracy requirements, and the number of experts in this field also meets the model accuracy requirements. However, in actual engineering practice, due to the different cognitive levels of experts, the number of experts in related fields does not necessarily meet the requirements of the model. Therefore, in the complex system health status assessment method, two issues need to be considered. First, due to the inconsistency of expert cognitive ability, there are disagreements in the expert information that is not unified, and there is also the phenomenon of expert islands in related fields. How to effectively integrate expert knowledge. Secondly, for ordinary whale optimization algorithms, it is easy to fall into local optimality and fail to achieve ideal optimization effects. Therefore, it is necessary to select a suitable optimization algorithm to ensure that the optimization results will not fall into local optimal solutions and how to ensure the effectiveness of the multi-expert model during the optimization process so that the number of experts will not degenerate or even become an ordinary single-expert BRB model. Based on this, the present invention proposes a multi-expert joint trust rule base complex system health status assessment method. Summary of the invention

[0006] The purpose of the present invention is to provide a complex system health status assessment method based on a multi-expert joint confidence rule base to solve the problems raised in the above background technology.

[0007] To achieve the above purpose, a complex system health status assessment method based on a multi-expert joint confidence rule base includes the following steps:

[0008] Step S1, constructing a multi-expert knowledge fusion method system;

[0009] Step S2, establishing a complex system health status assessment model based on a multi-expert joint confidence rule base;

[0010] Step S3, establishing a multi-group whale optimization algorithm suitable for a multi-expert joint confidence rule base;

[0011] Step S4, complex system health status assessment based on multi-expert joint confidence rule base.

[0012] Preferably, in step S1, in the multi-expert joint confidence rule base model, the reference value, attribute weight, rule weight, output confidence and expert constraint are given by each expert.

[0013] As a preferred embodiment: the fusion of the reference value can be expressed as formula (1):

[0014]

[0015] The number of reference values ​​is also determined by experts, where υ r represents the expert weight of the rth expert, which is determined by the expert's field and cognitive ability. In addition, it can also be determined by observed data training. and represents the n-1th, nth, and n+1th reference values ​​of the kth rule, and R is the number of experts in the modeling process;

[0016] The confidence of the fusion BRB-ME model of the output confidence can be expressed as formula (2)

[0017]

[0018] It should be noted that, affected by the incompleteness of expert knowledge, the sum of the output confidences is less than or equal to 1;

[0019] The attribute weights reflect the ability and relative importance of each attribute in expressing the system state. The attribute weights determined by experts can be further combined into formula (3):

[0020]

[0021] in is the weight of the mth attribute given by the rth expert;

[0022] The confidence rule weight represents its relative importance in the rule base. The confidence rule weight set by multiple experts can be calculated by formula (4):

[0023]

[0024] in is the weight of the kth attribute, given by the rth expert.

[0025] The expert constraints reflect the experts’ understanding of the output level. The upper and lower bounds of the constraints restrict the optimized parameters. The upper bound of the multi-expert joint constraints can be expressed as (5), and the lower bound can be expressed as (6)

[0026]

[0027]

[0028] in and They are the upper and lower bounds of the constraints on the nth output of the kth rule given by the rth expert.

[0029] Preferably, in the BRB-ME model of step S2, the knowledge of multiple experts is applied, and the relationship between the model input and output is represented by a confidence rule. The kth IF-THEN confidence rule given by the Pth expert is represented as follows:

[0030]

[0031] Where B is the output of the BRB-ME model, represents the reference value, ω k is the weight of the kth rule, ξ i (i=1,2,...,M) represents the weight of the i-th attribute, where M represents the number of attributes. i (i=1,2,…,N) represents the predicted result, γ i,k (i=1,2,…,N) represents the corresponding confidence level, where N represents the number of output results. is the parameter given by R experts.

[0032] As a preference, the input attributes in the BRB-ME model should be converted to a unified format using the following formula:

[0033]

[0034] in is the degree of match between the i-th input feature and the j-th confidence rule, and are the reference points in the kth and k+1th confidence rules, respectively, and y i represents the input attribute defined in (7), L represents the number of confidence rules in the BRB-ME model, and the matching degree of the input attribute can be transformed into a unified framework by comparing it with the attribute reference point.

[0035] As a preferred embodiment: in the BRB-ME model, different belief rules play different roles under different attribute inputs, and the activation weight represents the belief rule under information input, which can be obtained by the following formula:

[0036]

[0037]

[0038]

[0039] where ψ k is the activation weight of the kth confidence rule, ε k Indicates the matching degree of all input features with the kth confidence rule. Represents the relative weight of the i-th input attribute.

[0040] The final confidence is obtained by rule reasoning using the ER analysis algorithm. The calculation formula is as follows:

[0041]

[0042]

[0043] where γ n represents the confidence of the result, μ represents the intermediate parameter, and the confidence distribution of the evaluation result is obtained.

[0044] S(x)={(α n ,γ n ); n=1,2,...,N} (14)

[0045] Where x represents the input variable, and finally the health status assessment result of the model is obtained.

[0046]

[0047] where μ(α n ) represents the health status α n The utility value of .

[0048] As a preferred embodiment: the multi-population whale optimization algorithm (MEWOA) population in step S3 is divided into three categories: development sub-population, exploration sub-population, and moderate sub-population.

[0049] 1) Development subpopulation, used to speed up convergence and quickly find the global optimal solution. When an individual is close to the current optimal solution, it searches the surrounding area. This behavior is called local search, which can improve convergence speed and solution accuracy. Since the fitness of the development subpopulation is close to the current optimal solution, the development subpopulation focuses on performing local development to improve convergence speed and solution accuracy. The migration strategy of the development subpopulation is detailed in equation (34). The individuals in the development subpopulation evolve their positions by performing deep local search around the current optimal solution, as detailed in equation (19).

[0050] r2=2·rand[0,1] (16)

[0051] X(t+1)=X *(t)·r2 (17)

[0052] Where r2 is a random number whose value is between [0,2], and X * (t) is the position vector of the current optimal solution. The position of the individual in the development subpopulation is replaced by the increase or decrease of the position vector of the current optimal solution, thereby realizing the search in the area around the current optimal solution. This evolutionary strategy can make full use of the position information of the current optimal solution to accelerate the convergence speed and improve the solution accuracy.

[0053] 2) Individuals with poor fitness constitute the exploration subpopulation. Individuals in the exploration subpopulation are far away from the current optimal solution due to poor fitness, so they are allowed to conduct global exploration to find new solution areas. The migration strategy of the exploration subpopulation adopts the exploration stage method, as shown in formula (36). Individuals in the exploration subpopulation broaden the search range by increasing the absolute value of the position vector to enhance the global exploration ability. Formula (21) represents the evolution process of the exploration subpopulation.

[0054] r1=rand[0,1]+1 (18)

[0055] X(t+1)=X(t)·r1 (19)

[0056] Where r1 is a random number between [1,2]. If the current position is the local optimum, then formula (19) can help the individual escape from the local optimum with a certain probability. This is because under the action of formula (19), the absolute value of the current position vector will increase, allowing the individual to explore other areas away from the current position. After that, MEWOA can search other areas, making the global exploration capability more powerful.

[0057] 3) The remaining individuals form a moderate subpopulation whose individual fitness is between good and bad, which can be used to balance exploration and development. The individuals in the moderate subpopulation either explore to find new solutions or locally develop to increase the convergence speed to achieve a random balance between exploration and development. Its migration strategy is shown in formula (22).

[0058]

[0059] Individuals in the fitness sub-population evolve their positions according to adversarial learning (OBL). From equation (24), we can see that

[0060]

[0061]

[0062] Where lb is the lower bound of the problem space, ub is the upper bound, and fit() is the objective function. is a new position based on adversarial learning,

[0063] When the positions of all individuals are updated, the next iteration immediately performs population evolution. It is stipulated that MEWOA performs position update when t is an odd number, and performs population evolution when t is an even number; t represents the current number of iterations, so the probability of MEWOA performing position update and population evolution is 50% each. After each iteration, all individuals are re-ranked according to their current fitness. Therefore, the individuals in each subpopulation are dynamically updated with each iteration.

[0064] As a preferred embodiment: the step S3 multi-group whale optimization algorithm includes the following steps:

[0065] Step 1: Construct the objective function. The accuracy of the ordinary lithium-ion battery health status assessment model is limited. In order to obtain a more accurate model, the objective function of the BRB-ME model optimized by MEWOA is described as follows:

[0066] min{MSE} (23)

[0067]

[0068] Where m represents the number of training data, c and They represent the predicted value and true value of the health status of lithium-ion batteries of the BRB-ME model respectively;

[0069] Step 2: The whales update their positions through equation (30) to reach the best position for whales to catch food.

[0070] D=|C·X * (t)-X(t)| (25)

[0071] A=2a·r a -a (26)

[0072] C=2·r b (27)

[0073] X(t+1)=X * (t)-A·D (28)

[0074] Among them, t is the current iteration number, · is the element-by-element multiplication, and X * (t) is the position vector of the best whale, X(t) is the whale whose position vector needs to be updated, A and C are coefficient vectors, where r a and r b are two random vectors between [0,1], and vector a decreases linearly from 2 to 0 in proportion to the number of iterations;

[0075] Step 3: Bubble Net Attack Method

[0076] WOA's bubble net attack method is completed using formula (32),

[0077] D'=|X * (t)-X(t)| (29)

[0078] X(t+1)=D'·e bl ·cos(2πl)+X * (t) (30)

[0079] Where D' represents the distance between the current whale and the best whale, b is a constant, usually set to 1, and l is a random number between [-1,1];

[0080] Step 4: Multi-expert constraints

[0081] In the optimization stage, the expert weights are restricted.

[0082] υ r ≥k,r=(1,2,…,R) (31)

[0083] where υ r represents the expert weight belonging to the rth expert, and k represents the lower limit of the multi-expert constraint;

[0084] Step 5: Development Phase

[0085] The whale swims along a spiral path in a shrinking circle around its prey. The whale has a 50% chance of surrounding its prey and a 50% chance of attacking it with a bubble net. This process is achieved by equation (34).

[0086]

[0087] Where p is a random number between [0,1];

[0088] Step 6: Exploration Phase (Searching for Prey)

[0089] When |A|>1, the whale chooses random movement for global exploration, and its mathematical model is shown in formula (36).

[0090] D=|C·X rand (t)-X(t)| (33)

[0091] X(t+1)=X rand (t)-A·D (34)

[0092] Where X rand (t) is the whale randomly selected from the current population at the tth iteration, and |A| is the absolute value of A.

[0093] Compared with the prior art, the present invention has the following beneficial effects:

[0094] The present invention realizes the health status assessment of complex systems under multi-expert knowledge modeling, solves the problems of incomplete knowledge of single experts and expert islands, firstly, establishes a multi-expert knowledge fusion method to fuse the knowledge of multiple experts; secondly, adopts a multi-expert joint confidence rule base to establish a complex system health status assessment model to model the health status of complex systems; thirdly, adopts a multi-group whale optimization algorithm suitable for a multi-expert joint confidence rule base to optimize the model parameters to ensure the effectiveness and accuracy of the multi-expert model. The present invention can integrate the expert knowledge of multiple experts, realize the health status assessment of complex systems with a multi-expert joint confidence rule base, and ensure that the assessment results are consistent with the actual situation. Unlike the collaborative multi-expert system, the collaborative expert system reasonably divides the determined total task into several subtasks according to the domain knowledge, and each subtask corresponds to a subtask. Expert system, compared with the multi-expert joint confidence rule base, although the traditional collaborative expert system also uses multiple experts to participate in the construction of the system, each expert is responsible for a subtask and cannot participate in the entire construction. This leads to the incompleteness of expert knowledge and the inability to unify the subjectivity of the experts, which will affect the final result. The joint multi-expert confidence rule base model based on the confidence rule base integrates each expert's complete knowledge of the current system, and gives corresponding complete expert knowledge parameters from input to output, instead of being responsible for only part of the subsystem like the traditional collaborative multi-expert system. Therefore, it can ensure the completeness and stability of expert knowledge, reduce subjectivity, and have a more complete understanding of the modeling of expert systems and the health status of complex systems. It will not have a big impact on the entire system due to a certain expert, ensuring the stability and effectiveness of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 is a flow chart of the method of the present invention;

[0096] Figure 2 It is a complex system health status assessment model based on multi-expert joint confidence rule base;

[0097] Figure 3 Optimize the population partitioning diagram for multiple whale populations;

[0098] Figure 4 This is the health status assessment result of lithium-ion batteries. DETAILED DESCRIPTION

[0099] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the implementation regulations described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0100] Example

[0101] See also Figure 1-Figure 4 , FIG. 1 is a preferred embodiment of the present invention, a method for evaluating the health status of a complex system using a multi-expert joint confidence rule base, comprising the following steps:

[0102] Step S1, constructing a multi-expert knowledge fusion method system;

[0103] Step S2, establishing a complex system health status assessment model based on a multi-expert joint confidence rule base;

[0104] Step S3, establishing a multi-group whale optimization algorithm suitable for a multi-expert joint confidence rule base;

[0105] Step S4, complex system health status assessment based on multi-expert joint confidence rule base.

[0106] In this embodiment, in the multi-expert joint confidence rule base model, the reference value, attribute weight, rule weight, output confidence and expert constraint are given by each expert. The fusion methods of these four different parameters are different, as shown below:

[0107] Fusion of reference values: For the reference values ​​of the BRB-ME model, it should be ensured that the upper and lower bounds of the reference values ​​can completely include the data, and the distance between the two values ​​is reasonable. Since each expert has different cognition, the reference values ​​given are also different, and the obtained model accuracy is also different. Therefore, the reference value fusion method under the joint efforts of multiple experts can be expressed as formula (1)

[0108]

[0109] The number of reference values ​​is also determined by experts, where υ r represents the expert weight of the rth expert, which is determined by the expert's field and cognitive ability. In addition, it can also be determined by observed data training. and represents the n-1th, nth, and n+1th reference values ​​of the kth rule, and R is the number of experts in the modeling process.

[0110] Fusion of output confidence: Output confidence indicates the confidence of the output level, which directly reflects the expert's cognitive judgment of the complex system state. Due to the incompleteness of expert knowledge, the completeness of the output confidence given by each expert is different. Therefore, the confidence of the BRB-ME model can be expressed as formula (2):

[0111]

[0112] It should be noted that due to the incompleteness of expert knowledge, the sum of the output confidences is less than or equal to 1.

[0113] The attribute weight reflects the ability and relative importance of each attribute in expressing the system state. Different experts have different cognitive abilities of the system mechanism and different accuracy of attribute weight setting. Therefore, the attribute weights determined by experts can be further combined into formula (3):

[0114]

[0115] in is the weight of the mth attribute given by the rth expert.

[0116] The confidence rule weight indicates its relative importance in the rule base. It is determined by the frequency of use of the reference point combination of different attributes. The confidence rule weight needs to be set by experts in combination with the prediction of the future working state of the system. The confidence rule weight set by multiple experts can be calculated by formula (4):

[0117]

[0118] in is the weight of the kth attribute, given by the rth expert.

[0119] Expert constraints reflect the experts’ understanding of the output level. The upper and lower bounds of the constraints limit the optimized parameters to ensure that they do not violate the expert knowledge and obtain a complex system evaluation model that conforms to reality. Different experts give different upper and lower bounds of constraints. Therefore, the upper bound of the multi-expert joint constraint can be expressed as (5), and the lower bound can be expressed as (6)

[0120]

[0121]

[0122] in and They are the upper and lower bounds of the constraints on the nth output of the kth rule given by the rth expert.

[0123] In this embodiment, in the BRB-ME model, the knowledge of multiple experts is applied, and the relationship between the model input and output is represented by confidence rules. The kth IF-THEN confidence rule given by the Pth expert is represented as:

[0124]

[0125] Where B is the output of the BRB-ME model, represents the reference value, ω k is the weight of the kth rule, ξ i (i=1,2,...,M) represents the weight of the i-th attribute, where M represents the number of attributes, α i (i=1,2,...,N) represents the predicted result, γ i,k (i=1,2,...,N) represents the corresponding confidence level, where N represents the number of output results. are the parameters given by R experts, but the research fields of these R experts may be the same or different. In the BRB-ME model, the input is considered to be completely reliable and noise-free. The structure of the complex system health status assessment model is as follows: Figure 2 .

[0126] Furthermore, the attribute formats in the BRB-ME model are different and cannot be used directly, so the input attributes should be converted to a unified format using the following formula:

[0127]

[0128] in is the degree of match between the i-th input feature and the j-th confidence rule, and are the reference points in the kth and k+1th confidence rules, respectively, and y i represents the input attribute defined in (7), L represents the number of confidence rules in the BRB-ME model, and the matching degree of the input attribute can be transformed into a unified framework by comparing it with the attribute reference point.

[0129] In addition, in the BRB-ME model, different belief rules play different roles under different attribute inputs. The activation weight represents the belief rule under information input, which can be obtained by the following formula:

[0130]

[0131]

[0132]

[0133] where ψ kis the activation weight of the kth confidence rule, ε k Indicates the matching degree of all input features with the kth confidence rule, Represents the relative weight of the i-th input attribute.

[0134] The final confidence is obtained by rule reasoning using the ER analysis algorithm. The calculation formula is as follows:

[0135]

[0136]

[0137] where γ n represents the confidence of the result, and μ represents the intermediate parameter.

[0138] Get the confidence distribution of the evaluation results.

[0139] S(x)={(α n ,γ n ); n=1,2,…,N} (14)

[0140] Where x represents the input variable.

[0141] Finally, the health status assessment result of the model is obtained.

[0142]

[0143] where μ(α n ) represents the health status α n The utility value of .

[0144] In this embodiment, in step S3, the multi-population whale optimization algorithm (MEWOA) population is divided into three categories: development sub-population, exploration sub-population, and moderate sub-population.

[0145] 1) Development subpopulation, used to speed up convergence and quickly find the global optimal solution. When an individual is close to the current optimal solution, it searches the surrounding area. This behavior is called local search, which can improve convergence speed and solution accuracy. Since the fitness of the development subpopulation is close to the current optimal solution, the development subpopulation focuses on performing local development to improve convergence speed and solution accuracy. The migration strategy of the development subpopulation is detailed in Equation (34).

[0146] Individuals in the development subgroup evolve their positions by performing deep local search around the current optimal solution, as shown in Equation (19).

[0147] r2=2·rand[0,1] (16)

[0148] X(t+1)=X * (t)·r2 (17)

[0149] Where r2 is a random number whose value is between [0,2], and X * (t) is the position vector of the current optimal solution. The position of the individual in the development subpopulation is replaced by the increase or decrease of the position vector of the current optimal solution, thereby realizing the search in the area around the current optimal solution. This evolutionary strategy can make full use of the position information of the current optimal solution to accelerate the convergence speed and improve the solution accuracy.

[0150] 2) Individuals with poor fitness constitute the exploration subpopulation. Individuals in the exploration subpopulation are far away from the current optimal solution due to poor fitness, so they are allowed to conduct global exploration to find new solution areas. The migration strategy of the exploration subpopulation adopts the exploration stage method, as shown in formula (36).

[0151] Individuals in the exploration subgroup broaden the search range by increasing the absolute value of the position vector to enhance the global exploration capability. Equation (21) represents the evolution process of the exploration subgroup.

[0152] r1=rand[0,1]+1 (18)

[0153] X(t+1)=X(t)·r1 (19)

[0154] Where r1 is a random number between [1,2]. If the current position is a local optimum, equation (19) can help the individual escape from the local optimum with a certain probability. This is because under the action of equation (19), the absolute value of the current position vector will increase, allowing the individual to explore other areas away from the current position. After that, MEWOA can search other areas, making the global exploration capability more powerful.

[0155] 3) The remaining individuals constitute a moderate subpopulation whose individual fitness is between good and bad, which can be used to balance exploration and development. The individuals in the moderate subpopulation either explore to find new solutions or locally develop to increase the convergence speed to achieve a random balance between exploration and development. Its migration strategy is shown in formula (22).

[0156]

[0157] Individuals in the fitness sub-population evolve their positions according to opponent learning (OBL), as can be seen from equation (24).

[0158]

[0159]

[0160] Where lb is the lower bound of the problem space, ub is the upper bound, and fit() is the objective function. It is a new position based on adversarial learning.

[0161] When the positions of all individuals are updated, population evolution is performed immediately in the next iteration, which helps MEWOA improve population diversity and convergence speed and eliminate local optimality. It is stipulated that MEWOA updates its position when t is an odd number and performs population evolution when t is an even number; t represents the current number of iterations. Therefore, the probability of MEWOA performing position update and population evolution is 50% each.

[0162] After each iteration, all individuals are re-ranked according to their current fitness. Therefore, the individuals in each sub-population are dynamically updated with each iteration. The multi-population division based on fitness is as follows: Figure 3 .

[0163] Furthermore, step S3 of the multi-group whale optimization algorithm includes the following steps:

[0164] Step 1: Construct the objective function. The accuracy of the ordinary lithium-ion battery health status assessment model is limited. In order to obtain a more accurate model, the present invention optimizes the parameters of the BRB-ME model through the MEWOA algorithm and uses the mean square error as the evaluation index. Therefore, the objective function of the BRB-ME model is described as follows:

[0165] min{MSE} (23)

[0166]

[0167] Where m represents the number of training data, c and They represent the predicted value and true value of the health status of lithium-ion batteries of the BRB-ME model respectively.

[0168] Step 2: The whales update their positions through equation (30) to reach the best position for whales to catch food.

[0169] D=|C·X * (t)-X(t)| (25)

[0170] A=2a·r a -a (26)

[0171] C=2·r b (27)

[0172] X(t+1)=X * (t)-A·D (28)

[0173] Among them, t is the current iteration number, · is the element-by-element multiplication, and X * (t) is the position vector of the best whale, X(t) is the whale whose position vector needs to be updated, A and C are coefficient vectors, where r a and r bare two random vectors between [0,1], and vector a decreases linearly from 2 to 0 in proportion to the number of iterations.

[0174] Step 3: Bubble Net Attack Method

[0175] WOA's bubble net attack method is completed using formula (32).

[0176] D'=|X * (t)-X(t)| (29)

[0177] X(t+1)=D'·e bl ·cos(2πl)+X * (t) (30)

[0178] Where D' represents the distance between the current whale and the best whale, b is a constant, usually set to 1, and l is a random number between [-1, 1].

[0179] Step 4: Multi-expert constraints

[0180] During the optimization phase, it is necessary to ensure that the optimization results of the multi-expert parameters will not be too low or even reach 0. This can effectively prevent the multi-expert joint model from degenerating into a common expert model due to the number of experts. Therefore, the expert weights should be limited.

[0181] υ r ≥k,r=(1,2,…,R) (31)

[0182] where υ r represents the expert weight belonging to the rth expert. k represents the lower limit of the multi-expert constraint.

[0183] Step 5: Development Phase

[0184] Since the whale swims along a spiral path in a shrinking circle around its prey, there is a 50% chance that the whale will choose to surround its prey and a 50% chance that it will attack with a bubble net, which is achieved by equation (34).

[0185]

[0186] Where p is a random number between [0,1].

[0187] Step 6: Exploration Phase (Searching for Prey)

[0188] When |A|>1, the whale chooses random movement for global exploration. Its mathematical model is shown in formula (36):

[0189] D=|C·X rand (t)-X(t)| (33)

[0190] X(t+1)=X rand (t)-A·D (34)

[0191] where Xrand(t) is the whale randomly selected from the current population at the tth iteration, and |A| is the absolute value of A.

[0192] WOA does not have the ability to explore globally in the later stages of iteration, which can cause WOA to easily fall into a local optimal solution. Therefore, using the MEWOA algorithm can greatly improve this situation.

[0193] Lithium-ion batteries are widely used in electronic products, automobiles, energy storage systems, medical, industrial, military, and aerospace fields. Therefore, it is very important to accurately evaluate the health status of lithium-ion batteries. The present invention uses the No. 18650 lithium-ion battery of NASA's Center of Excellence in Prediction as the research object to evaluate the health status of lithium-ion batteries. The test conditions are charging mode and discharging mode. At a temperature of 24°C, the battery is subjected to a cyclic charge and discharge experiment until the rated capacity decreases by 30%. The experiment is stopped. The present invention uses B0006 battery to extract observation index data. In the charging mode, it is charged at a constant current (CC) of 1.5A until the battery voltage reaches 4.2V, and then continues to charge at a constant voltage (CV) until the charging current drops to 20mA. Therefore, a voltage of 3.8V is used as the initial voltage, a voltage of 4.2V is used as the end voltage, a current value of 1.5A is used as the initial current, and a current value of 0.5A is used as the end current. The number of charging cycles is 167. In order to study the health status of lithium-ion batteries, the above-mentioned performance evaluation model based on a multi-expert joint confidence rule base is used to evaluate the health status of lithium-ion batteries.

[0194] In the current study, the capacity of lithium-ion batteries is regarded as a direct indicator of the health status of lithium-ion batteries. The model defines two attributes of lithium battery capacity and sets four reference points for each attribute: long (VL), long (L), normal (N), and short (S). When lithium batteries are used in engineering, there are four selected evaluation levels to indicate their health status, completely safe (CS), safe (S), slightly poor (LB), and very poor (VB). 3.8V voltage is selected as the starting voltage and 4.2V cut-off voltage is selected as the termination voltage. Therefore, the charging time required from 3.8V voltage to 4.2V voltage is used as a health factor to characterize the health status of the battery and is recorded as the partial charging time (PC Time-CC) of the CC stage. In the actual application of the battery, there is a situation where the charging is incomplete and the cut-off current value cannot be reached. The partial CV stage is selected, that is, the current value of 1.5A is selected as the starting current and the current of 0.5A is selected as the ending current.

[0195] The reference values ​​of relevant attributes given by multiple experts are shown in Table 1. The health status of the B0006 battery pack is detailed in Table 2. Based on the confidence rule construction method, the lithium battery health status assessment model has a total of 16 confidence rules, see Table 3 and Table 4. Due to the incompleteness of each expert's expert knowledge, the sum of the confidence of each confidence rule given is not necessarily 1, but less than or equal to 1. The initial parameters of the BRB-ME lithium battery assessment status model are given by two experts with the same expert weights, and the weights of multiple initial experts are set to 1.

[0196] Table 1 Initial reference values ​​given by multiple experts

[0197]

[0198] Table 2 Initial evaluation levels given by multiple experts

[0199]

[0200] Table 3 Confidence table given by expert 1

[0201]

[0202] Table 4 Confidence table given by expert 2

[0203]

[0204]

[0205] The data set has a total of 167 groups of samples. The present invention selects 112 samples for model training and the remaining 55 samples for model test fitting. The number of optimization iterations of MEWOA is set to 300, and the initial population number is set to 180. There are a total of 190 optimization parameters, namely output confidence, attribute weight, expert weight, rule weight, reference value and final output utility. The optimized fused multi-expert attribute reference value and the corresponding attribute weight are shown in Table 5. It can be seen that the attribute PC Time-CV has a larger weight and has a greater impact on the health status assessment results. The optimized fused multi-expert evaluation level is shown in Table 6, and the optimized fused confidence table and the corresponding rule weight are shown in Table 7.

[0206] Table 5 Optimized fusion multi-expert attribute reference values ​​and corresponding attribute weights

[0207]

[0208] Table 6 Optimized fusion multi-expert evaluation level

[0209]

[0210] Table 7 Optimized fusion confidence table and corresponding rule weights

[0211]

[0212]

[0213] In this embodiment, if Figure 4 As shown in the scatter plot, the actual values ​​are distributed along the predicted values, showing a high degree of linear correlation. Then, there are two bands around the actual values ​​of the linear fit: the 95% confidence interval and the 95% prediction interval. The confidence interval indicates the certainty of the regression line, and the prediction interval indicates the credibility of future observations falling within a specific range. As shown in Table 8, the statistical indicators further quantify the performance of the model. Its determination coefficient (R 2 ) is 0.98712, indicating that the model can explain most of the data variability, and the Pearson correlation coefficient is 0.99354, emphasizing the positive correlation between the predicted value and the actual value. These high correlation and fit indicators show that the current multi-expert joint confidence rule base model has statistically significant predictive ability. As shown in Table 8, the mean square error (MSE) is 0.00071, the mean absolute error (MAE) is 0.02003, and the root mean square error (RMSE) is 0.02675, indicating that the error level is very low and the model has a high degree of accuracy and reliability. These indicators together show that the model has excellent fitting ability for the data and the prediction error is relatively small, indicating the effectiveness and accuracy of the model.

[0214] Table 8 Indicators of lithium-ion battery model

[0215]

[0216] The above content is a further detailed description of the present invention in combination with specific implementation methods. It cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, some simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as belonging to the scope of protection determined by the claims submitted for the present invention.

Claims

1. A multi-expert joint confidence rule base complex system health status assessment method, It is characterized in that The steps include: Step S1, constructing a multi-expert knowledge fusion method system; Step S2, establishing a complex system health status assessment model based on a multi-expert joint confidence rule base; Step S3, establishing a multi-group whale optimization algorithm suitable for a multi-expert joint confidence rule base; Step S4, complex system health status assessment based on multi-expert joint confidence rule base.

2. According to claim 1, a multi-expert joint confidence rule base complex system health status assessment method, Features: In step S1, in the multi-expert joint confidence rule base model, the reference value, attribute weight, rule weight, output confidence and expert constraint are given by each expert.

3. A method for evaluating the health status of complex systems using a multi-expert joint confidence rule base according to claim 2, Features: The fusion of the reference values ​​can be expressed as formula (1): The number of reference values ​​is also determined by experts, where υ r represents the expert weight of the rth expert, which is determined by the expert's field and cognitive ability. In addition, it can also be determined by observed data training. and represents the n-1th, nth, and n+1th reference values ​​of the kth rule, and R is the number of experts in the modeling process; The confidence of the fusion BRB-ME model of the output confidence can be expressed as formula (2) It should be noted that, affected by the incompleteness of expert knowledge, the sum of the output confidences is less than or equal to 1; The attribute weights reflect the ability and relative importance of each attribute in expressing the system state. The attribute weights determined by experts can be further combined into formula (3): in is the weight of the mth attribute given by the rth expert; The confidence rule weight represents its relative importance in the rule base. The confidence rule weight set by multiple experts can be calculated by formula (4): in is the weight of the kth attribute, given by the rth expert. The expert constraints reflect the experts’ understanding of the output level. The upper and lower bounds of the constraints restrict the optimized parameters. The upper bound of the multi-expert joint constraints can be expressed as (5), and the lower bound can be expressed as (6) in and They are the upper and lower bounds of the constraints on the nth output of the kth rule given by the rth expert.

4. According to claim 1, a multi-expert joint confidence rule base complex system health status assessment method, Features: In the BRB-ME model of step S2, the knowledge of multiple experts is applied, and the relationship between the model input and output is represented by confidence rules. The kth IF-THEN confidence rule given by the Pth expert is represented as: Where B is the output of the BRB-ME model, represents the reference value, ω k is the weight of the kth rule, ξ i (i=1,2,...,M) represents the weight of the i-th attribute, where M represents the number of attributes. i (i=1,2,...,N) represents the predicted result, γ i,k (i=1,2,...,N) represents the corresponding confidence level, where N represents the number of output results. is the parameter given by R experts.

5. A method for evaluating the health status of a complex system using a multi-expert joint confidence rule base according to claim 4, Features: The input attributes in the BRB-ME model should be converted into a unified format using the following formula: in is the degree of match between the i-th input feature and the j-th confidence rule, and are the reference points in the kth and k+1th confidence rules, respectively, and y i represents the input attribute defined in (7), L represents the number of confidence rules in the BRB-ME model, and the matching degree of the input attribute can be transformed into a unified framework by comparing it with the attribute reference point.

6. A method for evaluating the health status of a complex system using a multi-expert joint confidence rule base according to claim 5, Features: In the BRB-ME model, different belief rules play different roles under different attribute inputs. The activation weight represents the belief rule under information input, which can be obtained by the following formula: where ψ k is the activation weight of the kth confidence rule, ε k Indicates the matching degree of all input features with the kth confidence rule. Represents the relative weight of the i-th input attribute. The final confidence is obtained by rule reasoning using the ER analysis algorithm. The calculation formula is as follows: where γ n represents the confidence of the result, μ represents the intermediate parameter, and the confidence distribution of the evaluation result is obtained. S(x)={(a n ,c n );n=1,2,...,N} (14) Where x represents the input variable, and finally the health status assessment result of the model is obtained. where μ(α n ) represents the health status α n The utility value of .

7. A method for evaluating the health status of a complex system using a multi-expert joint confidence rule base according to claim 1, Features: In step S3, the multi-population whale optimization algorithm (MEWOA) population is divided into three categories: development sub-population, exploration sub-population, and moderate sub-population; 1) Development subpopulation, used to speed up convergence and quickly find the global optimal solution. When an individual is close to the current optimal solution, it searches the surrounding area. This behavior is called local search, which can improve convergence speed and solution accuracy. Since the fitness of the development subpopulation is close to the current optimal solution, the development subpopulation focuses on performing local development to improve convergence speed and solution accuracy. The migration strategy of the development subpopulation is detailed in equation (34). The individuals in the development subpopulation evolve their positions by performing deep local search around the current optimal solution, as detailed in equation (19). r 2 =2·rand[0,1] (16) X(t+1)=X * (t)·r 2 (17) where r 2 is a random number whose value is between [0,2], X * (t) is the position vector of the current optimal solution. The position of the individual in the development subpopulation is replaced by the increase or decrease of the position vector of the current optimal solution, thereby realizing the search in the area around the current optimal solution. This evolutionary strategy can make full use of the position information of the current optimal solution to accelerate the convergence speed and improve the solution accuracy. 2) Individuals with poor fitness constitute the exploration subpopulation. Individuals in the exploration subpopulation are far away from the current optimal solution due to poor fitness, so they are allowed to conduct global exploration to find new solution areas. The migration strategy of the exploration subpopulation adopts the exploration stage method, as shown in formula (36). Individuals in the exploration subpopulation broaden the search range by increasing the absolute value of the position vector to enhance the global exploration ability. Formula (21) represents the evolution process of the exploration subpopulation. r 1 =rand[0,1]+1 (18) X(t+1)=X(t)·r 1 (19) where r 1 is a random number between [1,2]. If the current position is the local optimum, then formula (19) can help the individual escape from the local optimum with a certain probability. This is because under the action of formula (19), the absolute value of the current position vector will increase, allowing the individual to explore other areas away from the current position. After that, MEWOA can search other areas, making the global exploration capability more powerful. 3) The remaining individuals form a moderate subpopulation whose individual fitness is between good and bad, which can be used to balance exploration and development. The individuals in the moderate subpopulation either explore to find new solutions or locally develop to increase the convergence speed to achieve a random balance between exploration and development. Its migration strategy is shown in formula (22). Individuals in the fitness sub-population evolve their positions according to adversarial learning (OBL). From equation (24), we can see that Where lb is the lower bound of the problem space, ub is the upper bound, and fit() is the objective function. is a new position based on adversarial learning, When the positions of all individuals are updated, the next iteration immediately performs population evolution. It is stipulated that MEWOA performs position update when t is an odd number, and performs population evolution when t is an even number; t represents the current number of iterations, so the probability of MEWOA performing position update and population evolution is 50% each. After each iteration, all individuals are re-ranked according to their current fitness. Therefore, the individuals in each subpopulation are dynamically updated with each iteration.

8. A method for evaluating the health status of a complex system using a multi-expert joint confidence rule base according to claim 7, Features: The step S3 multi-group whale optimization algorithm includes the following steps: Step 1: Construct the objective function. The accuracy of the ordinary lithium-ion battery health status assessment model is limited. The objective function of the BRB-ME model optimized by MEWOA is described as follows: min{MSE} (23) Where m represents the number of training data, c and They represent the predicted value and true value of the health status of lithium-ion batteries of the BRB-ME model respectively; Step 2: The whales update their positions through equation (30) to reach the best position for whales to catch food. D=|C·X * (t)-X(t)| (25) A=2a·r a -a (26) C=2·r b (27) X(t+1)=X * (t)-A·D (28) Among them, t is the current iteration number, · is the element-by-element multiplication, and X * (t) is the position vector of the best whale, X(t) is the whale whose position vector needs to be updated, A and C are coefficient vectors, where r a and r b are two random vectors between [0,1], and vector a decreases linearly from 2 to 0 in proportion to the number of iterations; Step 3: Bubble Net Attack Method WOA's bubble net attack method is completed using formula (32), D'=|X * (t)-X(t)| (29) X(t+1)=D'·e bl ·cos(2πl)+X * (t) (30) Where D' represents the distance between the current whale and the best whale, b is a constant, usually set to 1, and l is a random number between [-1,1]; Step 4: Multi-expert constraints In the optimization stage, the expert weights are restricted. u r ≥k,r=(1,2,...,R) (31) where υ r represents the expert weight belonging to the rth expert, and k represents the lower limit of the multi-expert constraint; Step 5: Development Phase The whale swims along a spiral path in a shrinking circle around its prey. The whale has a 50% chance of surrounding its prey and a 50% chance of attacking it with a bubble net. This process is achieved by equation (34). Where p is a random number between [0,1]; Step 6: Exploration Phase (Searching for Prey) When |A|>1, the whale chooses random movement for global exploration, and its mathematical model is shown in formula (36). D=|C·X rand (t)-X(t)| (33) X(t+1)=X rand (t)-A·D (34) Where X rand (t) is the whale randomly selected from the current population at the tth iteration, and |A| is the absolute value of A.

Citation Information

Cited By

  • Fresh agricultural product distribution path optimization method and system based on real-time road condition detection

    CN118917515A

  • Fresh agricultural product distribution route optimization method and system based on real-time road condition detection

    CN118917515B