WO2019218263A1 - Extreme learning machine-based extreme ts fuzzy inference method and system - Google Patents

Abstract

An extreme learning machine-based extreme TS fuzzy inference method and system. The method comprises: clustering an original condition attribute value matrix corresponding to a training data set by means of a k-means clustering algorithm; constructing an expansion decision attribute value matrix according to the clustering result; training a single extreme learning machine by using the expansion decision attribute value matrix to obtain an output layer weight and the trained extreme learning machine; inputting a new sample into the trained extreme learning machine to obtain a premise firing strength and a qualified consequent of a fuzzy rule; and performing defuzzification according to the firing strength and the qualified consequent to obtain a predicted output of the new sample. A single extreme learning machine is trained by using an expansion decision attribute value matrix so that parameters which do not require iteration are optimized; the training process can be quickly completed; the training time is short; a softmax function-based defuzzification operation can efficiently standardize the firing strength; the predicted output data can be efficiently output.

Description

Limit TS fuzzy inference method and system based on extreme learning machine Technical field

The invention relates to the technical field of computers, and in particular to a limit TS fuzzy inference method and system based on an extreme learning machine.

Background technique

The Takagi-Sugeno (TS) fuzzy inference system was first proposed by Japanese scholars T. Takagi and M. Sugeno, and further refined by M.Sugeno and GTKang. TS fuzzy reasoning is also called Sugeno fuzzy reasoning or Takagi-Sugeno-Kang (TSK). Fuzzy reasoning. The core idea is to "fuse a simple linear system to fit a complex nonlinear system" by using a series of localizations by fuzzification of the input, inference calculus based on fuzzy rules, and defuzzification of the output. The linear model goes to approximate the purpose of an overall nonlinear model. Theoretical studies show that the TS fuzzy inference system can approximate any nonlinear model with arbitrary precision.

The key point of constructing the TS fuzzy inference system is the determination of the fuzzy rule front trigger intensity (Firing Strength: measure the amount of input and fuzzy rule matching) and the final conclusion of the Qualified Consequent (input corresponding fuzzy rule output). . The premise of determining the trigger strength and the true value of the conclusion is to determine the learning parameters of the membership function in the rule predecessor and the regression coefficients of the multiple linear regression model in the rule poster. The classical TS fuzzy inference system is based on Adaptive Neural-Fuzzy Inference System (ANFIS) (published by IEEE TSMC in 1993), which uses a five-layer topology similar to neural networks and uses back propagation. The algorithm and least squares method determine the premise parameters and Consequent Coefficients of the fuzzy rules.

However, the main drawback of ANFIS is the long training time.

Summary of the invention

The main object of the present invention is to provide a method and system for constructing a limit TS fuzzy inference system, which aims to solve the technical problem that the TS fuzzy inference system has a long training time in the prior art.

To achieve the above object, a first aspect of the present invention provides a TS fuzzy rule inference method based on an extreme learning machine, including:

Step A: K-means clustering algorithm is used to cluster the original conditional attribute value matrix corresponding to the training data set, and the clustering result is obtained, and the expansion decision is constructed according to the clustering result and the predicted output data of the training data set. Attribute value matrix;

Step B: training the single extreme learning machine by using the extended decision attribute value matrix, obtaining the output layer weight of the extreme learning machine, and the trained extreme learning machine;

Step C: input an unknown new sample into the trained limit learning machine, and obtain a trigger strength of the fuzzy rule front piece of the new sample and a conclusion true value of the fuzzy rule back piece;

Step D: Defuzzify the new sample according to the trigger strength, the true value of the conclusion, and the preset Softmax function, to obtain a predicted output of the new sample.

To achieve the above object, a second aspect of the present invention provides a limit TS fuzzy inference system based on an extreme learning machine, including:

a clustering building module is configured to cluster the original conditional attribute value matrix corresponding to the training data set by using a K-means clustering algorithm to obtain a clustering result, and according to the clustering result and the predicted output of the training data set Data construction extends the decision attribute value matrix;

a training module, configured to use the extended decision attribute value matrix to train a single extreme learning machine, obtain an output layer weight of the extreme learning machine, and a trained extreme learning machine;

An input obtaining module, configured to input an unknown new sample into the trained extreme learning machine, to obtain a triggering strength of the fuzzy rule front piece of the new sample and a conclusion true value of the fuzzy rule back piece;

The defuzzification module is configured to defuzzify the new sample according to the trigger strength, the conclusion true value, and the preset Softmax function, to obtain a predicted output of the new sample.

The invention provides a limit TS fuzzy inference method based on an extreme learning machine, the method comprising: clustering a raw condition attribute value matrix corresponding to a training data set by using a K-means clustering algorithm to obtain a clustering result, and according to the clustering The class result and the predicted output data of the training data set construct a matrix of extended decision attribute values, and the single limit learning machine is trained by using the extended decision attribute value matrix to obtain the output layer weight of the extreme learning machine, and the ultimate learning machine after training. The unknown new sample is input into the trained extreme learning machine, and the trigger strength of the fuzzy rule front piece of the new sample and the conclusion true value of the fuzzy rule back piece are obtained, according to the trigger strength, the conclusion true value and the preset softmax function, The new sample is defuzzified to obtain the predicted output of the new sample. Compared with the prior art, the single extreme learning machine is trained by using the extended decision attribute value matrix, so that the parameter optimization without iteration can be completed, the training process can be completed quickly, the training time is short, and further, the softmax function based solution blur is performed. The operation can effectively realize the normalization processing of the trigger strength and effectively realize the output of the predicted output data.

DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the description of the prior art will be briefly described below. Obviously, the drawings in the following description are only It is a certain embodiment of the present invention, and those skilled in the art can obtain other drawings according to these drawings without any creative work.

1 is a schematic flow chart of a limit TS fuzzy inference method based on an extreme learning machine according to an embodiment of the present invention;

2 is a schematic structural diagram of a limit TS fuzzy inference system based on an extreme learning machine according to an embodiment of the present invention;

3 is a schematic diagram showing the fuzzy inference performance of E-TSFIS on a Laser regression data set in an embodiment of the present invention;

4 is four fuzzy membership functions of the third rule in the fuzzy rule base constructed by E-TSFIS on the Laser regression data set according to an embodiment of the present invention;

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. The embodiments are merely a part of the embodiments of the invention, and not all of the embodiments. All other embodiments obtained by a person skilled in the art based on the embodiments of the present invention without creative efforts are within the scope of the present invention.

1 is a schematic flowchart of a limit TS fuzzy inference method based on an extreme learning machine according to an embodiment of the present invention, where the method includes:

Step 101: Using a K-means clustering algorithm to cluster the original conditional attribute value matrix corresponding to the training data set, obtain a clustering result, and construct an extended decision attribute value matrix according to the clustering result and the predicted output data of the training data set;

In an embodiment of the invention, given a training data set, the training data set is as follows:

Where D represents the training data set, N is the number of samples, D is the number of conditional attributes of the sample, R is the fuzzy rule, and t _n is the fuzzy rule of the nth sample.

Based on the above training data set, the corresponding Original Condition Attribute-Value Matrix OCM is obtained as follows:

Where X represents the original conditional attribute value matrix corresponding to the training data set, and x ₁₁ to x _ND represents the samples in the training data set. Assuming that the number of clusters is K and K>1, it can be understood that the number of fuzzy rules in the fuzzy rule base is the same as the number of clusters, and is also K.

In the embodiment of the present invention, the K-means clustering algorithm is used to cluster the original conditional attribute value matrix corresponding to the training data set, and the clustering result is obtained, and the clustering result and the predicted output data of the training data set are used to construct the expansion decision. The attribute value matrix, the matrix of the extended decision attribute value is as follows:

among them,

Where n = 1, 2, ..., N, k = 1, 2, ..., K.

表示训练数据集中的样本。 Where O represents the expanded decision attribute value matrix, Z represents the clustering result, and Y represents the predicted output data of the training data set,

Represents a sample in the training data set.

Step 102: Using a matrix of extended decision attribute values to train a single extreme learning machine, obtaining an output layer weight of the extreme learning machine, and an extreme learning machine after training;

In the embodiment of the present invention, a single extreme learning machine is trained by using the above extended decision attribute value matrix, wherein the extreme learning machine is also called a single implicit layer feedforward neural network, and the extreme learning machine includes L implicit words. A layer node, and L is greater than or equal to the number of conditional attributes D of the sample.

Among them, the input layer weight of the extreme learning machine is as follows:

表示第l(l＝1,2,3,.......,L)个隐含层节点的输入层权重。 Wherein, the input layer weight is randomly selected (for example, obeying a uniformly distributed random number in the interval [-1, 1]), and

Indicates the input layer weight of the l (l = 1, 2, 3, ..., L) hidden layer nodes.

Among them, the hidden layer offset of the extreme learning machine is as follows:

Where σ _l represents the offset of the l (l=1, 2, 3, . . . , L) hidden layer nodes, and the hidden layer nodes are obtained by the sigmoid activation function.

After the input layer weight and the implicit layer offset of the extreme learning machine have been determined, the extended decision attribute value matrix is input into the extreme learning machine, and the extreme learning machine is trained to obtain the output layer weight of the extreme learning machine, which can be understood. After obtaining the weight of the output layer of the extreme learning machine, it indicates that the training of the extreme learning machine has been completed.

The output layer weights of the Extreme Learning Machine are as follows:

It should be noted that because HB=O, the calculation formula of the output layer weight can be: B=(H ^T H) ^-1 H ^T O

among them,

among them,

Where n=1,2,...,N,l=1,2,...,L,H is the implicit layer output matrix of the extreme learning machine after the input of the original conditional attribute value matrix corresponding to the training data set.

Step 103: Input an unknown new sample into the trained extreme learning machine, and obtain a trigger strength of the fuzzy rule front piece of the new sample and a conclusion true value of the fuzzy rule back piece;

In the embodiment of the present invention, the unknown new learning sample may be processed by the trained extreme learning machine to obtain the triggering strength of the new fuzzy sample for the fuzzy rule and the conclusion true value of the fuzzy rule. Specifically, based on the implicit layer output matrix of the extreme learning machine and the output layer weight, the original conditional attribute value matrix corresponding to the new sample, according to the following formula, the triggering strength of the fuzzy rule front piece and the fuzzy value of the fuzzy rule Hou Jian:

Where Z is the trigger strength, Y is the conclusion true value, (h ₁ h ₂ ... h _L ) is the implicit layer output matrix of the extreme learning machine after the new sample input limit learning machine, B is the output layer weight, and D is the new The number of conditional attributes of the sample, x _d represents the dth data in the new sample, w _d1 , w _d2 to w _dL represent the input layer weight of the extreme learning machine, and L represents the number of hidden layer nodes of the extreme learning machine, σ ₁ , σ ₂ to σ _L represent the implicit layer offset of the extreme learning machine.

Step 104: Defuzzify the new sample according to the trigger strength, the true value of the conclusion, and the preset Softmax function, to obtain a predicted output of the new sample.

In the embodiment of the present invention, after obtaining the trigger strength of the fuzzy rule front piece and the conclusion true value of the fuzzy rule back piece, the new sample is defuzzified by using the trigger strength, the conclusion true value, and the preset softmax function. Obtain the predicted output of the new sample, which can be done as follows:

Where Q represents the predicted output of the new sample, K represents the number of clusters at the time of clustering, z _k represents the triggering strength of the fuzzy rule front of the new sample, τ represents the temperature parameter of the softmax function, and is greater than 0, y _k represents The true value of the conclusion of the fuzzy rule of the new sample.

The purpose of using the softmax function here is that, in the limit TS fuzzy reasoning, the trigger strength of the new sample for the fuzzy rule front is not guaranteed to be greater than 0. Therefore, the softmax function is used to normalize the trigger strength of the fuzzy rule's front piece. It is transformed into a K-dimensional probability vector (v ₁ v ₂ ... v _K ), where each element in the probability vector has a value between (0, 1) and the sum of all elements is 1, such as :

且，

And,

In the embodiment of the present invention, the single extreme learning machine is trained by using the extended decision attribute value matrix, so that the parameter optimization without iteration can be completed, the training process can be completed quickly, the training time period, and further, the solution based on the softmax function. The fuzzification operation can effectively realize the normalization processing of the trigger strength and effectively realize the output of the predicted output data.

In addition, the fuzzy rule base can be further obtained. Specifically, refer to step 105 to step 108 in the embodiment shown in FIG. 2, as follows:

Step 105: Input the training data set into the trained extreme learning machine, obtain the trigger strength of the fuzzy rule front piece corresponding to the actual output data and the training data, and construct the actual extended decision attribute value matrix by using the trigger strength and the actual output data;

且可得到训练数据集对应的模糊规则前件的触发强度

并利用触发强度及实际输出数据构建实际拓展决策属性值矩阵，如下： In the embodiment of the present invention, the training data set is input into the trained extreme learning machine, and the extreme learning machine outputs the actual output data of the training data set.

And the trigger strength of the fuzzy rule front piece corresponding to the training data set can be obtained.

And use the trigger strength and actual output data to construct the actual extended decision attribute value matrix, as follows:

among them,

Where B represents the output layer weight of the extreme learning machine, (h _n1 h _n2 ... h _nL ) represents the hidden layer output matrix of the extreme learning machine, N represents the number of samples in the training data set, and K represents the number of clusters.

转化成触发强度概率向量(v _n1 v _n2 … v _nK)，即触发强度概率向量为： In order to avoid the influence of negative values in the trigger strength on the construction of the fuzzy rule base, the intensity will be triggered here.

Converted into a trigger strength probability vector (v _n1 v _n2 ... v _nK ), ie the trigger strength probability vector is:

表示第n个样本对应第k个类簇的触发强度，τ表示softmax函数的温度参数，且大于0。 Where v _nk represents the trigger strength probability vector of the nth sample corresponding to the kth cluster.

Indicates the trigger strength of the nth sample corresponding to the kth cluster, and τ represents the temperature parameter of the softmax function, and is greater than zero.

Step 106: Select a uniformly distributed random number from the preset interval as a center of the fuzzy membership function in the fuzzy rule front piece;

Step 107: Obtain a trigger strength probability vector of the fuzzy rule preamble of the training data set based on the actual extended decision attribute value matrix, and use the center of the fuzzy membership function to solve the fuzzy condition under the condition that the trigger strength probability vector is equal to the product of the fuzzy membership function The radius of the membership function;

Step 108: When there is a non-positive value in the radius of the fuzzy membership function, return to step 106. When the radius of the fuzzy membership function is positive, the fuzzy rule base is obtained based on the center and radius of the fuzzy membership function.

In the embodiment of the present invention, a randomly distributed random number is selected from the preset interval, for example, selected from the interval [0, 1]. Among them, the selected random number represents the center m _kd of the fuzzy membership function in the fuzzy rule front.

Let the trigger strength probability vector be equal to the product of the membership function, ie:

Deriving the above formula, you can get:

表示触发强度概率向量，

表示第n个且具有第D个样本 属性的样本的隶属度函数，x _nd表示第n个且具有第d个样本属性的样本，c _kd表示第k个类簇且具有第d个样本属性的样本对应的模糊隶属度函数的中心，m _kd表示第k个类簇且具有第d个样本属性的样本对应的模糊隶属度函数的半径，D表示样本属性的个数。 among them,

Indicates the trigger strength probability vector,

a membership function representing the nth sample with the Dth sample attribute, x _nd represents the nth sample with the _dth sample attribute, c _kd represents the kth cluster and has the dth sample attribute The center of the fuzzy membership function corresponding to the sample, m _kd represents the radius of the fuzzy membership function corresponding to the sample of the kth cluster and having the dth sample attribute, and D represents the number of sample attributes.

Further, based on the derived formula, a matrix product operation as shown below is constructed:

among them,

Where v ₁₁ to v _NK represent the trigger strength probability vector, m ₁₁ to m _KD represent the radius of the fuzzy membership function, and x ₁₁ to x _ND represent the value in the original conditional attribute value matrix corresponding to the training data set, c ₁₁ To c _KD represents the center of the fuzzy membership function.

所以，求解模糊隶属度函数的半径可以通过求解该矩阵的伪逆获得： due to

Therefore, the radius of the fuzzy membership function can be obtained by solving the pseudo inverse of the matrix:

After the radius of the fuzzy membership function is obtained, if there is a non-positive value in the radius of the fuzzy membership function, the process returns to step 106, and the center of the fuzzy membership function is re-selected, and then solved. When the radius of the fuzzy membership function is positive, the fuzzy rule base is obtained based on the center and radius of the fuzzy membership function.

In the embodiment of the present invention, by using a plurality of "randomly selecting the radius of the fuzzy membership function---solving the radius of the fuzzy membership function", until there is no negative value in the radius of the obtained fuzzy membership function. So that the fuzzy rule base can be obtained by using the center and radius of the fuzzy membership function.

Among them, the fuzzy rule base is as follows:

Where R ₁ to R _K represent _K fuzzy rules in the fuzzy rule base, and A ₁₁ to A _KD represent membership degrees, the membership function including the center and radius of the membership function, x ₁ , x ₂ ,..., x _D represents a sample, y ₁ to y _K represent output data, K represents the number of clusters, and D represents the number of conditional attributes.

In the embodiment of the present invention, by using a single extreme learning machine to simultaneously calculate the triggering strength of the fuzzy rule front piece and the true value of the post-part conclusion, the inference speed is extremely fast, and the iterative parameter optimization is not needed, and the training process can be completed quickly, and the training is performed. The time is short, and at the same time, an efficient fuzzy rule base construction mechanism is provided, which can quickly and accurately determine the membership function in the fuzzy rule predecessor, and can provide a complete fuzzy rule base.

It should be noted that, in the above technical solution, a single implicit layer feedforward neural network trained based on the extended decision attribute value matrix for training fuzzy reasoning may also be used in the method of error back propagation, and the above fuzzy membership function There may be various types, such as Gaussian membership function, triangular membership function, trapezoidal membership function, bell membership function, etc. In practical applications, the corresponding type of membership function can be selected based on specific needs. Make a limit.

2 is a schematic structural diagram of a limit TS fuzzy inference system based on an extreme learning machine according to an embodiment of the present invention, including:

The clustering construction module 201 is configured to cluster the original conditional attribute value matrix corresponding to the training data set by using a K-means clustering algorithm to obtain a clustering result, and according to the clustering result and the prediction of the training data set Output data construction to expand the decision attribute value matrix;

The training module 202 is configured to use the extended decision attribute value matrix to train a single extreme learning machine to obtain an output layer weight of the extreme learning machine, and a trained extreme learning machine;

An input obtaining module 203, configured to input an unknown new sample into the trained extreme learning machine, to obtain a triggering strength of the fuzzy rule front piece of the new sample and a conclusion true value of the fuzzy rule back piece;

The defuzzification module 204 is configured to defuzzify the new sample according to the trigger strength, the conclusion true value, and the preset Softmax function, to obtain a predicted output of the new sample.

In the embodiment of the present invention, the system further includes:

The input construction module 205 is configured to input the training data set into the trained extreme learning machine, obtain the trigger strength of the actual output data and the fuzzy rule front piece corresponding to the training data, and utilize the trigger strength and the actual output data. Construct a matrix of actual extended decision attribute values;

The selecting module 206 is configured to select, from the preset interval, a uniformly distributed random number as a center of the fuzzy membership function in the fuzzy rule front piece;

The solving module 207 is configured to obtain, according to the actual extended decision attribute value matrix, a trigger strength probability vector of the fuzzy rule preamble of the training data set, where the trigger strength probability vector is equal to a condition of a product of the fuzzy membership function Calculating a radius of the fuzzy membership function by using a center of the fuzzy membership function;

a determining module 208, configured to return the selection module when there is a non-positive value in a radius of the fuzzy membership function, and when the radius of the fuzzy membership function is a positive value, based on the fuzzy membership degree The center and radius of the function get the fuzzy rule base.

It should be noted that the content of each module in the embodiment shown in FIG. 2 is similar to the content of each step in the embodiment shown in FIG. 1 . For details, refer to the content in the embodiment shown in FIG. 1 , and details are not described herein.

In the embodiment of the present invention, by using a single extreme learning machine to simultaneously calculate the triggering strength of the fuzzy rule front piece and the true value of the post-part conclusion, the inference speed is extremely fast, and the iterative parameter optimization is not needed, and the training process can be completed quickly, and the training is performed. The time is short, and at the same time, an efficient fuzzy rule base construction mechanism is provided, which can quickly and accurately determine the membership function in the fuzzy rule predecessor, and can provide a complete fuzzy rule base.

The experimental data of the technical solution of the embodiment of the present invention is provided below:

In the Laser Regression Data Set (4 conditional attributes, 993 samples, available at KEEL: http://www.keel.es/), the reasoning of the "limit TS fuzzy inference system E-TSFIS" in the embodiment of the present invention The performance was verified. The number of fuzzy rules in E-TSFIS is K=5, the number of hidden layer nodes in the extreme learning machine is L={100,150,...,950,1000}, and the temperature parameter τ=0.005 in Softmax function. We tested the training error on the entire data set of E-TSFIS (using the mean squared Mean Squared Error-MSE metric). For each hidden layer node, we use the mean of 10 MSEs to represent the final inference result. The experimental results are shown in Figure 3. The schematic diagram of the fuzzy inference performance of the E-TSFIS (limit fuzzy inference system) on the Laser regression data set. .

From Fig. 3, we can see that with the increase of the number L of hidden layer nodes in the extreme learning machine, the training error of E-TSFIS gradually decreases and presents a convergence trend, which indicates that we only use one extreme learning machine for fuzzy reasoning. The idea is feasible and effective. At the same time, in this experiment we reconstruct a fuzzy rule base for Laser regression dataset. The center and radius of the Gaussian membership function in the fuzzy rule antecedent are as follows (5 rules, 4 conditional attributes):

with

Based on the above C matrix and M matrix, we can construct the fuzzy membership function in the fuzzy rule base. For example, for the third rule, the four fuzzy membership functions in the antecedent are:

with

The graph of the corresponding membership function is shown in Figure 4. It is the four fuzzy membership functions of the third rule in the fuzzy rule base constructed by E-TSFIS on the Laser regression data set.

An embodiment of the present invention further provides a terminal, including a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the computer program, the implementation is as shown in the embodiment shown in FIG. Each step in the limit TS fuzzy inference method based on the extreme learning machine.

In the several embodiments provided by the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be another division manner, for example, multiple modules or components may be combined or Can be integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or module, and may be electrical, mechanical or otherwise.

The modules described as separate components may or may not be physically separated. The components displayed as modules may or may not be physical modules, that is, may be located in one place, or may be distributed to multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules.

The integrated modules, if implemented in the form of software functional modules and sold or used as separate products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention, which is essential or contributes to the prior art, or all or part of the technical solution, may be embodied in the form of a software product stored in a storage medium. A number of instructions are included to cause a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and the like. .

It should be noted that, for the foregoing method embodiments, for the sake of brevity, they are all described as a series of action combinations, but those skilled in the art should understand that the present invention is not limited by the described action sequence. Because certain steps may be performed in other sequences or concurrently in accordance with the present invention. In the following, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

In the above embodiments, the descriptions of the various embodiments are all focused, and the parts that are not detailed in a certain embodiment can be referred to the related descriptions of other embodiments.

The foregoing is a description of a limit learning machine-based limit TS fuzzy rule inference method and system provided by the present invention. For those skilled in the art, according to the idea of the embodiment of the present invention, the specific implementation manner and the application range are In view of the above, the contents of this specification are not to be construed as limiting the invention.

Claims (10)

A limit TS fuzzy inference method based on extreme learning machine, characterized in that the method comprises: Step A: K-means clustering algorithm is used to cluster the original conditional attribute value matrix corresponding to the training data set, and the clustering result is obtained, and the expansion decision is constructed according to the clustering result and the predicted output data of the training data set. Attribute value matrix; Step B: training the single extreme learning machine by using the extended decision attribute value matrix, obtaining the output layer weight of the extreme learning machine, and the trained extreme learning machine; Step C: input an unknown new sample into the trained limit learning machine, and obtain a trigger strength of the fuzzy rule front piece of the new sample and a conclusion true value of the fuzzy rule back piece; Step D: Defuzzify the new sample according to the trigger strength, the true value of the conclusion, and the preset Softmax function, to obtain a predicted output of the new sample. The method of claim 1 further comprising: Step E: input the training data set into the trained extreme learning machine, obtain the trigger strength of the fuzzy rule front piece corresponding to the actual output data and the training data, and construct the actual expansion decision by using the trigger strength and the actual output data. Attribute value matrix; Step F: selecting a uniformly distributed random number from the preset interval as a center of the fuzzy membership function in the fuzzy rule front piece; Step G: Obtain a trigger strength probability vector of a fuzzy rule preamble of the training data set based on the actual extended decision attribute value matrix, and use the trigger strength probability vector equal to a product of the fuzzy membership function Solving a radius of the fuzzy membership function by a center of the fuzzy membership function; Step H: When there is a non-positive value in the radius of the fuzzy membership function, return to the step F, when the radius of the fuzzy membership function is positive, based on the center of the fuzzy membership function And the radius gets the fuzzy rule base. The method according to claim 1, wherein the step C specifically comprises: And based on the implicit layer output matrix of the extreme learning machine and the output layer weight, and the original condition attribute value matrix corresponding to the new sample, the triggering strength of the fuzzy rule front piece and the fuzzy rule post-product are obtained according to the following formula Conclusion True value:

Wherein, Z represents the trigger strength, Y represents the conclusion true value, and (h ₁ h ₂ ... h _L ) represents the hidden layer output matrix of the extreme learning machine after the new sample is input to the extreme learning machine, and B represents the output. Layer weight, D represents the number of conditional attributes of the new sample, x _d represents the dth data in the new sample, w _d1 , w _d2 to w _dL represents the input layer weight of the extreme learning machine, and L represents the limit learning machine The number of hidden layer nodes, σ ₁ , σ ₂ to σ _L, represent the implicit layer offset of the extreme learning machine. The method according to claim 1, wherein the step D specifically comprises: The defuzzification process is performed using the following formula to obtain the predicted output of the new sample:

Where Q represents the predicted output of the new sample, K represents the number of clusters at the time of clustering, z _k represents the triggering strength of the fuzzy rule front of the new sample, and τ represents the temperature parameter of the softmax function, and is greater than 0, y _k represents the true value of the conclusion of the fuzzy rule of the new sample. The method according to claim 2, wherein in the step G, the radius of the fuzzy membership function is solved according to the following formula:

among them:

Where v ₁₁ to v _NK represent the trigger strength probability vector, m ₁₁ to m _KD represent the radius of the fuzzy membership function, and x ₁₁ to x _ND represent the value in the original conditional attribute value matrix corresponding to the training data set, c ₁₁ To c _KD represents the center of the fuzzy membership function. A limit TS learning system based on extreme learning machine, characterized in that the system comprises: a clustering building module is configured to cluster the original conditional attribute value matrix corresponding to the training data set by using a K-means clustering algorithm to obtain a clustering result, and according to the clustering result and the predicted output of the training data set Data construction extends the decision attribute value matrix; a training module, configured to use the extended decision attribute value matrix to train a single extreme learning machine, obtain an output layer weight of the extreme learning machine, and a trained extreme learning machine; An input obtaining module, configured to input an unknown new sample into the trained extreme learning machine, to obtain a triggering strength of the fuzzy rule front piece of the new sample and a conclusion true value of the fuzzy rule back piece; The defuzzification module is configured to defuzzify the new sample according to the trigger strength, the conclusion true value, and the preset Softmax function, to obtain a predicted output of the new sample. The system of claim 6 wherein the system further comprises: An input construction module, configured to input the training data set into the trained extreme learning machine, obtain the trigger strength of the actual output data and the fuzzy rule front piece corresponding to the training data, and construct the excitation intensity and the actual output data Actually expand the decision attribute value matrix; The selecting module is configured to select a uniformly distributed random number from the preset interval as a center of the fuzzy membership function in the fuzzy rule front piece; a solution module, configured to obtain, according to the actual extended decision attribute value matrix, a trigger strength probability vector of a fuzzy rule preamble of the training data set, where the trigger strength probability vector is equal to a product of the fuzzy membership function Solving a radius of the fuzzy membership function by using a center of the fuzzy membership function; a determining module, configured to return the selection module when there is a non-positive value in a radius of the fuzzy membership function, and when the radius of the fuzzy membership function is a positive value, based on the fuzzy membership function The center and radius get a fuzzy rule base. The system according to claim 6, wherein said input obtaining module is specifically configured to: And based on the implicit layer output matrix of the extreme learning machine and the output layer weight, and the original condition attribute value matrix corresponding to the new sample, the triggering strength of the fuzzy rule front piece and the fuzzy rule post-product are obtained according to the following formula Conclusion True value:

Wherein, Z represents the trigger strength, Y represents the conclusion true value, and (h ₁ h ₂ ... h _L ) represents the hidden layer output matrix of the extreme learning machine after the new sample is input to the extreme learning machine, and B represents the output. Layer weight, D represents the number of conditional attributes of the new sample, x _d represents the dth data in the new sample, w _d1 , w _d2 to w _dL represents the input layer weight of the extreme learning machine, and L represents the limit learning machine The number of hidden layer nodes, σ ₁ , σ ₂ to σ _L, represent the implicit layer offset of the extreme learning machine. The system according to claim 6, wherein the defuzzification module is specifically configured to: The defuzzification process is performed using the following formula to obtain the predicted output of the new sample:

Where Q represents the predicted output of the new sample, K represents the number of clusters at the time of clustering, z _k represents the triggering strength of the fuzzy rule front of the new sample, and τ represents the temperature parameter of the softmax function, and is greater than 0, y _k represents the true value of the conclusion of the fuzzy rule of the new sample. The system according to claim 7, wherein the solving module is specifically configured to solve the radius of the fuzzy membership function according to the following formula:

among them:

Where v ₁₁ to v _NK represent the trigger strength probability vector, m ₁₁ to m _KD represent the radius of the fuzzy membership function, and x ₁₁ to x _ND represent the value in the original conditional attribute value matrix corresponding to the training data set, c ₁₁ To c _KD represents the center of the fuzzy membership function.

Images