CN117610610A - A network evaluation method for neuro-fuzzy systems with learning ability - Google Patents

Abstract

The invention provides a neural fuzzy system network evaluation method with learning ability, which comprises the following steps: s1, establishing a neural fuzzy system model, wherein the neural fuzzy system model adopts fuzzy logic rules under a framework of a neural network; s2, training the neural fuzzy system model by adopting an improved particle swarm intelligent optimization algorithm to obtain optimal parameters and a trained neural fuzzy system model; and S3, defining a system input value according to the actual problem, and inputting the system input value into the trained neural fuzzy system model to obtain a capacity boundary evaluation result of the system. The invention combines the reasoning capacity of the fuzzy logic and the infinite approximation function capacity of the neural network to establish a substitution model of the real system, and then adopts an intelligent optimization algorithm to enable the substitution model to be maximally close to the real model. The invention provides an optimization scheme for learning and training a complex system with more unknown parameters, and the method has the unique advantage and has certain reference significance for improving the system performance.

Description

A network evaluation method for neuro-fuzzy systems with learning ability

Technical field

The present invention relates to the field of artificial intelligence technology, specifically, to a neural fuzzy system network evaluation method with learning ability.

Background technique

Online capability boundary evaluation models based on real-time characteristic parameters have developed rapidly in intelligent technology in recent years. The research and application fields of this model involve many fields, such as industrial production, energy, environmental monitoring, medical and health, etc. Currently, a variety of online competency boundary assessment models for specific fields have been proposed and applied. With the development of Internet of Things technology and sensor technology, real-time data collection technology continues to improve, and the speed and accuracy of data collection continue to improve. This provides reliable data support for the application of online capability boundary evaluation models based on real-time characteristic parameters. With the continuous development and improvement of artificial intelligence and machine learning technology, the algorithm of the online capability boundary assessment model based on real-time characteristic parameters has also been continuously optimized, which can more accurately predict and control the boundaries of the target system. Practical application gradually unfolds: The practical application of this model continues to unfold in many fields, such as equipment fault detection and early warning in industrial automation production, fault diagnosis and control in power systems, health monitoring and diagnosis in the medical and health fields, etc. , have been widely used and verified.

The fuzzy comprehensive evaluation method is an evaluation method based on fuzzy mathematics, which can quantitatively describe fuzzy information through membership functions to solve fuzzy problems. However, there is currently no uniformly recognized specification for the determination of membership functions in the fuzzy comprehensive evaluation method, so it is subject to greater subjective influence. In recent years, type-2 fuzzy theory has gradually been successfully applied in fields such as fuzzy control, which has also promoted the development of type-2 fuzzy clustering research. However, limited by the computational complexity of type-2 fuzzy theory, the application of type-2 fuzzy clustering algorithm in evaluation models needs further research. Methods based on artificial neural networks are mainly suitable for state estimation research. However, in general, this method has poor interpretability and requires a large amount of state monitoring data and health state representation parameter data support, making it difficult to apply online.

To sum up, a fuzzy comprehensive evaluation method needs to be invented to solve the problem of ensuring the reliability and accuracy of data in a changing environment, improving the real-time and accuracy of the algorithm, and reducing the reliance on experts' subjective judgment and experience. Over-reliance, less subjectivity.

Contents of the invention

In view of this, the purpose of the present invention is to propose a neuro-fuzzy system network evaluation method with learning capabilities to solve the technical problem that existing evaluation methods are not suitable for complex systems containing multiple unknown parameters.

The technical means adopted in the present invention are as follows:

A learning-capable neuro-fuzzy system network evaluation method, which is characterized by including the following steps:

S1. Establish a neuro-fuzzy system model that adopts fuzzy logic rules under the framework of a neural network;

S2. Use the improved particle swarm intelligent optimization algorithm to train the neuro-fuzzy system model, and obtain the optimal parameters and the trained neuro-fuzzy system model;

S3. Based on the actual problem, define the system input value and input it into the trained neuro-fuzzy system model to obtain the system's capability boundary evaluation results.

Further, in S1, the neuro-fuzzy system model is as follows:

NFS＝{P,T,I,O,M,W,f}

Among them, P = {p ₁ , p ₂ ,..., p _n } is an initial library containing finite elements representing avionics system equipment, subsystems and system damage events, p _i ∈ [0,1]; T ={t ₁ ,t ₂ ,...,t _m } is a terminal library with finite elements, used to quantify the influence of elements in the initial library on the subsystem, _ti ∈[0,1]; I (O) is an input or output library, reflecting the mapping of mutations to the system; M is a mapping, in which each library node p _i has a tag value M(pi ₎ , reflecting the proposition represented by the library node The true degree of represents the uncertainty of equipment and subsystems; W = {w ₁ , w ₂ ,..., w _r } is the weight set of rules, reflecting the degree of relationship between the elements of the initial library and the terminal library ;f is a nonlinear function used to map unknown and complex relationships in the system.

Further, S2 specifically includes the following steps:

Initialization generates a group of random particles and iterates;

In each iteration, the particle continuously updates itself by tracking two extreme values;

Update particle speed and position;

Iterate over random particles to find the optimal solution.

Furthermore, among the two extreme values, one is the optimal solution found by the particle itself, which is called the individual extreme value p _best , and the other is the optimal solution found by the entire population so far, which is called the global extreme value g _best .

Furthermore, the velocity and position update formulas of the particle swarm are:

in, and/> represent the velocity and position of the d-dimensional component of particle i in the k-th generation respectively; v _max represents the maximum velocity of the particle; x _min and x _max represent the minimum and maximum positions of the particle respectively; r ₁ and r ₂ represent (0, 1 ); usually c ₁ =c ₂ =0.5 is the learning factor; the inertia weight is set to m.

Furthermore, in the speed and position update formula of the particle swarm, the update formula of the inertia weight is:

m ^k =m _min +(m _max -m _min )(k _max -k)/k _max

Among them, m ^k , m _min and m _max are the inertia weight, minimum weight and maximum weight at the current moment respectively, are the maximum number of iterations, and k is the current number of iterations.

Further, in S3, the system input includes deterministic input and uncertain input.

The present invention also provides a storage medium that includes a stored program, wherein when the program is run, any one of the above neural fuzzy system network evaluation methods with learning capabilities is executed.

The present invention also provides an electronic device, including a memory, a processor and a computer program stored on the memory and executable on the processor. The processor executes any of the above by running the computer program. A network evaluation method for neuro-fuzzy systems with learning capabilities.

Compared with the prior art, the present invention has the following advantages:

It combines the reasoning ability of fuzzy logic and the infinite approximation function ability of neural network to establish an alternative model of the real system, and then uses intelligent optimization algorithms to make the alternative model as close to the real model as possible. The neuro-fuzzy system establishes the system's capability boundary evaluation model and introduces fuzzy rules into the framework of the neural network. This method combines the infinite approximation function ability of the neural network and the learning ability of the fuzzy system, and is universal. The experimental results and analysis demonstrate the comprehensiveness and rationality of the assessment model, which can be applied to complex system capability assessment research. Especially for complex systems containing more unknown parameters, this method has unique advantages and has certain reference significance for improving system performance.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that need to be used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without exerting any creative effort.

Figure 1 is a flow chart of the method of the present invention.

Figure 2 is a diagram of the algorithm convergence process in three cases of the present invention.

Figure 3 is a comparison diagram between the best weight and the optimized weight of the present invention.

Figure 4 is a comparison chart of predicted values and actual values according to the present invention.

Figure 5 is a diagram showing the system capability evaluation results of one-dimensional variables of the present invention.

Figure 6 is a diagram showing the system capability evaluation results of two-dimensional variables of the present invention.

Detailed ways

In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only These are some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts should fall within the scope of protection of the present invention.

It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It is to be understood that the data so used are interchangeable under appropriate circumstances so that the embodiments of the invention described herein are capable of being practiced in sequences other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus that encompasses a series of steps or units and need not be limited to those explicitly listed. Those steps or elements may instead include other steps or elements not expressly listed or inherent to the process, method, product or apparatus.

As shown in Figure 1, the present invention provides a learning-capable neuro-fuzzy system network evaluation method, which includes the following steps:

S1. Structured neuro-fuzzy system modeling

Neuro-fuzzy systems combine the characteristics of neural networks and fuzzy systems, and have the infinite approximation capabilities of neural networks and the fuzzy reasoning capabilities of fuzzy logic systems. The neuro-fuzzy system (NFS) model is established and defined as follows. NFS contains seven elements, defined as

NFS＝{P,T,I,O,M,W,f}

Among them, P = {p ₁ , p ₂ ,..., p _n } is an initial library containing finite elements representing avionics system equipment, subsystems and system damage events, p _i ∈ [0,1]; T ={t ₁ ,t ₂ ,...,t _m } is a terminal library with finite elements, used to quantify the influence of elements in the initial library on the subsystem, _ti ∈[0,1]. I(O) is an input (output) library that reflects the mapping of mutations to the system. M is a map in which each library node p _i has a tag value M(pi ₎ , which reflects the true degree of the proposition represented by the library node and represents the uncertainty of the device and subsystem. W={w ₁ , w ₂ ,..., w _r } is a weight set of rules, which reflects the degree of relationship between the elements of the initial library and the terminal library. f is a nonlinear function used to map unknown and complex relationships in the system.

The NFS model uses fuzzy logic rules under the framework of neural networks to make the reasoning of system logical relationships clear and smooth. The introduction of parameters W, M and nonlinear function f clearly describes the complex weight relationship of the system. This method avoids the complexity of modeling and reduces modeling costs, which are advantages that traditional methods do not have.

The termination library T is the output set of all subsystems and the overall system. Elements of a terminating library can serve as initial library elements for other subsystems. The input library I is a subset of the initial library and represents the initial library elements transformed from all non-terminal library elements. The output library O is a subset of the terminal library. For a large complex system consisting of multiple subsystems, the outputs of some subsystems can be measured, while the outputs of some subsystems cannot be obtained due to limitations or excessive cost. The output library contains the output of all measurable subsystems. Due to the existence of uncertain factors, such as measurement errors, external interference and signal transmission losses, each element in the initial library is assigned a parameter M( _pi ) to represent the true degree of the value. This operation can improve the robustness of the model and have more experimental value. The nonlinear function f is introduced based on the fuzzy logic system.

For the output continuous function y=f(x)∈R, the input state is defined as

x＝[x ₁ ,x ₂ ,...,x _n ] ^T ∈R ⁿ

Neuro-fuzzy systems usually include fuzzy reasoning units and neural network layers, which introduce fuzzy reasoning into the neural network structure, making it more flexible for modeling complex systems. It is built on a set of "if-then" rules. If-then rules, produce relationships connecting inputs and outputs. If-Then rules are reflected in the following aspects.

Define A _j1 , A _j2 ,..., A _jn and B _j as fuzzy sets. Rule j: If x ₁ is A _{j 1} , x ₂ is A _{j 2} and ... and x _n is A _{j n} , then y is B _j . Then, the estimation error ε of the system is described as follows

y＝W ^T Φ(x)+ε

where the basic function vector is

as well as Use Gaussian function. For any continuous function f(x), there exists an FLS that satisfies:

sup|f(x)-W ^T Φ(x)|≤ε ₀

Among them, W is the adjustable weight matrix and ε ₀ is the maximum estimation error.

S2. Data-driven training and learning

The improved particle swarm intelligent optimization algorithm is used to learn the unknown weight parameters in the fuzzy neural system established in the first step. This algorithm is a global optimization method based on swarm intelligence and can implement effective search in complex spaces. This paper uses an improved particle swarm optimization algorithm to tune a large number of weight parameters in the above-mentioned aviation security system vulnerability assessment model. This method is simple, has few adjustment parameters, and can effectively solve a large number of nonlinear, nondifferentiable and multi-extreme complex problems. The algorithm first initializes to generate a group of random particles; then iterates to find the optimal solution, which is the least square norm of the error between the predicted value and the true value. In each iteration, the particle continuously updates itself by tracking two extreme values: one is the optimal solution found by the particle itself, called the individual extreme value p _best , and the other is the optimal solution found so far by the entire population. , called the global extreme value g _best , and then update the speed and position.

in, and/> represent the velocity and position of the d-dimensional component of particle i in the k-th generation respectively; v _max represents the maximum velocity of the particle; x _min and x _max represent the minimum and maximum positions of the particle respectively; r ₁ and r ₂ represent (0, 1 ); usually c ₁ =c ₂ =0.5 is the learning factor. The inertia weight is defined as m. When m is large, the particle has the ability to increase the size of the search space, and its global search ability is quite powerful. When m is small, the local search ability is strong. The traditional inertia weight is a constant. The PSO algorithm used in this article has made the following improvements. Made the following improvements

m ^k =m _min +(m _max -m _min )(k _max -k)/k _max

Among them, m ^k , m _min and m _max are the inertia weight, minimum weight and maximum weight at the current moment respectively, which are the maximum number of iterations, and k is the current number of iterations. Therefore, the inertia weight has adaptive ability to ensure the convergence of the search process. sex.

S3. Online system capability boundary assessment

Based on actual problems, define system inputs, including deterministic inputs and uncertain inputs; define some hyperparameters in the model. Input the trained NFS model to get the system's capability boundary evaluation results.

Example

The improved particle swarm optimization algorithm is integrated into the neuro-fuzzy system network model, which can predict reference weight values by learning from sample data and evaluate capability boundaries online. The steps are as follows:

1) Prepare sample data set. The data set contains 250 samples, of which 200 samples are used for training and 50 samples are used as a test set to test the accuracy of the training, that is, Ω = {(I1, O1), (I2, O2),..., (Im, OM)}, m=250. Due to the lack of experimental data, the samples here are obtained through simulations to test the method proposed in this paper.

2) Determine the number of weights and search space. The system is shown in Figure 3. There are 4 fuzzy logic systems in total, and it is assumed that each fuzzy logic system has 5 weights, so a total of 20 weight parameters are needed for optimization. All weights are searched in the interval [0, 1].

3) Initialize the parameters in the optimization algorithm. Use the weight parameter to be learned as the search space particle and determine the maximum velocity of the particle. Initialize the position and velocity of each particle, where the particle swarm is randomly generated. The number of particles is set to 100, the maximum number of iterations is 50, c1=0.5, c2=0.5.

4) Update the speed and position of particles in each dimension;

5) Decode each particle according to the sample data and find the value of the terminal library tag of the NFS model for system capability evaluation. Take the difference between the predicted value and the true value of the value marked by each terminal. Then, the individual optimal and group optimal are updated according to the fitness.

6) Perform step 4 to iterate until the termination rule is reached and the parameters are learned;

7) Obtain the optimized weight and the program ends. Based on the NFS model reasoning algorithm, an improved particle swarm optimization algorithm is introduced to enable it to have parameter learning capabilities and get rid of the dependence on experts and experience for many unknown parameters in the system.

The simulation is carried out in three cases and the fitness values of all subsystems are represented respectively. It can be seen that convergence can be achieved after using the improved particle swarm optimization algorithm for 20 iterations in the three cases. From comparing the curves in Figure 2 in the three cases, it can be concluded that increasing the knowledge information of the system can improve the optimization accuracy. Optimal weights and optimization are compared in Figure 3 with weights in Case 3. Moreover, the test results are shown in Figure 4, which shows the excellent performance of the proposed NFS model. In this way, the trained fuzzy neural system network model can be used to evaluate the capabilities of the system. When the input library changes over time, the evaluation can obtain results online in real time. Figure 5 shows the results of a real-time evaluation of the system capabilities when a variable is present in the input library. Furthermore, we can see that variable x1 has a small impact on the system capabilities, while x10 has a greater impact. Therefore, through the fuzzy neural system network model, the sensitivity of the elements in the input library can also be used to obtain system capability assessment results. When there are two variables in the system input library, the system capability evaluation result is a surface, Figure 6. According to the surface, the capability boundary of the system can be called an interval. For example, when the range of variable x1 is [0.3, 0.6] and the range of variable x9 is [0.05, 0.35], then the system's capability range is [0.1101, 0.1766], that is, the upper and lower capability boundaries are 0.1101 and 0.1766 respectively. When the variable range of variable x9 is [0.05, 0.35] and the variable range of x10 is [0.3, 0.6], then the system capability range is [0.0111, 0.3029], that is, the upper and lower capability boundaries are 0.0111 and 0.3029 respectively. Similarly, when there are multiple system input libraries in the variables, the system evaluation results can easily obtain capabilities through the NFS model. The application of fuzzy neural system network will greatly simplify the description of system functional capabilities. System capabilities establish an evaluation model with learning capabilities. On a system with four subsystems as the research object, calculation examples support its learning efficiency and accuracy. The experimental results and analysis demonstrate the comprehensiveness and rationality of the assessment model, which can be applied to complex system capability assessment research. Especially for complex systems containing more unknown parameters, this method has unique advantages and has certain reference significance for improving system performance.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, but not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: The technical solutions described in the foregoing embodiments can still be modified, or some or all of the technical features can be equivalently replaced; and these modifications or substitutions do not deviate from the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention. scope.

Claims (9)

1. A neuro-fuzzy system network evaluation method with learning ability, which is characterized by including the following steps: S1. Establish a neuro-fuzzy system model that adopts fuzzy logic rules under the framework of a neural network; S2. Use the improved particle swarm intelligent optimization algorithm to train the neuro-fuzzy system model, and obtain the optimal parameters and the trained neuro-fuzzy system model; S3. Based on the actual problem, define the system input value and input it into the trained neuro-fuzzy system model to obtain the system's capability boundary evaluation results. 2. The neuro-fuzzy system network evaluation method with learning ability according to claim 1, characterized in that, in S1, the neuro-fuzzy system model is as follows: NFS＝{P,T,I,O,M,W,f} Among them, P = {p ₁ , p ₂ ,..., p _n } is an initial library containing finite elements representing avionics system equipment, subsystems and system damage events, p _i ∈ [0,1]; T ={t ₁ ,t ₂ ,...,t _m } is a terminal library with finite elements, used to quantify the influence of elements in the initial library on the subsystem, _ti ∈[0,1]; I (O) is an input or output library, reflecting the mapping of mutations to the system; M is a mapping, in which each library node p _i has a tag value M(pi ₎ , reflecting the proposition represented by the library node The true degree of represents the uncertainty of equipment and subsystems; W = {w ₁ , w ₂ ,..., w _r } is the weight set of rules, reflecting the degree of relationship between the elements of the initial library and the terminal library ;f is a nonlinear function used to map unknown and complex relationships in the system. 3. The neuro-fuzzy system network evaluation method with learning ability according to claim 1, characterized in that S2 specifically includes the following steps: Initialization generates a group of random particles and iterates; In each iteration, the particle continuously updates itself by tracking two extreme values; Update particle speed and position; Iterate over random particles to find the optimal solution. 4. The neural fuzzy system network evaluation method with learning ability according to claim 3, characterized in that, among the two extreme values, one is the optimal solution found by the particle itself, which is called the individual extreme value p _best , The other is the optimal solution found so far by the entire population, which is called the global extreme value g _best . 5. The neural fuzzy system network evaluation method with learning ability according to claim 4, characterized in that the speed and position update formula of the particle swarm is: in, and/> represent the velocity and position of the d-dimensional component of particle i in the k-th generation respectively; v _max represents the maximum velocity of the particle; x _min and x _max represent the minimum and maximum positions of the particle respectively; r ₁ and r ₂ represent (0, 1 ); usually c ₁ =c ₂ =0.5 is the learning factor; the inertia weight is set to m. 6. The neural fuzzy system network evaluation method with learning ability according to claim 5, characterized in that, in the speed and position update formula of the particle swarm, the update formula of the inertia weight is: m ^k =m _min +(m _max -m _min )(k _max -k)/k _max Among them, m ^k , m _min and m _max are the inertia weight, minimum weight and maximum weight at the current moment respectively, are the maximum number of iterations, and k is the current number of iterations. 7. The network evaluation method of a neuro-fuzzy system with learning ability according to claim 1, characterized in that, in S3, the system input includes deterministic input and uncertainty input. 8. A storage medium, characterized in that the storage medium includes a stored program, wherein when the program is run, the neurofuzzy method with learning ability described in any one of claims 1 to 7 is executed. System network assessment methods. 9. An electronic device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the right by running the computer program The network evaluation method of a neuro-fuzzy system with learning ability according to any one of claims 1 to 7.