CN107392315B - Breast cancer data classification method for optimizing brain emotion learning model - Google Patents
Breast cancer data classification method for optimizing brain emotion learning model Download PDFInfo
- Publication number
- CN107392315B CN107392315B CN201710550139.7A CN201710550139A CN107392315B CN 107392315 B CN107392315 B CN 107392315B CN 201710550139 A CN201710550139 A CN 201710550139A CN 107392315 B CN107392315 B CN 107392315B
- Authority
- CN
- China
- Prior art keywords
- emotion learning
- brain emotion
- learning model
- orbital
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Abstract
The invention discloses a method for optimizing a brain emotion learning model. The brain emotion learning model simulates an emotion learning mechanism between an almond body and an orbital-frontal cortex in the brain, and has the characteristics of low calculation complexity and high calculation speed. The invention introduces a self-adaptive genetic algorithm into the improved brain emotion learning model, evaluates the model output by using a fitness function, and optimizes and improves the weight and the threshold of the almond body and the orbital-frontal cortex unit in the brain emotion learning model. The method utilizes the global search capability of the adaptive genetic algorithm and the rapid local search capability of the brain emotion learning algorithm to guide the genetic algorithm to approach the optimal solution rapidly, thereby improving the rapidity and the accuracy of the brain emotion learning model in data analysis.
Description
Technical Field
The invention relates to the field of machine learning and intelligent calculation, in particular to a breast cancer data classification method for optimizing a brain emotion learning model by adopting a self-adaptive genetic algorithm.
Background
The classification or prediction of the data may provide guidance for further decisions by the system. Therefore, data classification or prediction is widely applied in machine learning, data mining, pattern recognition, and the like. The artificial neural network is widely applied to data analysis due to good self-learning and self-adaptive performance of the artificial neural network. However, with the increase of data complexity, the traditional BP learning algorithm has a long training time and a slow convergence rate, which makes it difficult to satisfy the requirement of fast and accurate analysis of data.
In 2000, Morren proposed a Brain Emotional Learning (BEL) -based computational model based on neuro-anatomical research results, which was built based on neurobiological principles of interaction between the amygdala and the orbitofrontal cortex in the Brain, and simulated the mechanism by which Emotional stimuli induce rapid Emotional responses in the short reflex pathways of the Brain. The BEL model has low calculation complexity and high calculation speed, can overcome the defects of the traditional neural network, and shows certain advantages in data analysis in recent years.
In the BEL model, the weight regulation method between the almond body and the orbital and frontal cortex plays a decisive role in the emotional learning effect of the brain. Morren proposes a reinforcement learning method based on reward signals to adjust the weight, but the method causes the generality of the model to be poor. Later, researchers introduced Genetic Algorithms (GA) widely used for parameter optimization into the BEL model, achieved optimal adjustment of the amygdala and orbitofrontal cortex weights, and used the optimized BEL model for data analysis, although this method enhanced the versatility of the BEL model. However, the basic genetic algorithm has certain randomness and blindness in the optimization process, the population evolution speed is reduced or even does not evolve at the later stage of search, and finally, the optimal solution may not be found. The basic genetic algorithm has fixed parameter setting, especially fixed cross probability and variation probability, so that the genetic algorithm has low flexibility and low search speed in solving the problem.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a breast cancer data classification method for optimizing a brain emotion learning model by adopting an Adaptive Genetic Algorithm (Adaptive Genetic Algorithm) with variable cross probability and variation probability. In addition, the structure of the brain emotion learning model is further improved.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows: a breast cancer data classification method for optimizing a brain emotion learning model comprises the following steps:
1) establishing an improved neural network based on a brain emotion learning model according to the characteristics of the training samples;
2) initializing weights and thresholds of an almond body unit and an orbital-frontal cortex unit in the improved neural network;
3) selecting, self-adaptive crossing and self-adaptive variation operations are carried out by adopting a roulette method and an optimal individual storage strategy in a genetic algorithm, and the weight and the threshold of the almond body and the orbital frontal cortex in the neural network are optimized and improved;
4) updating the optimized weight and threshold combination to an improved neural network to obtain an optimal network structure;
5) and respectively calculating the output of the almond body, the output of the orbital-frontal cortex and the optimal network overall output under the combined action of the almond body and the orbital-frontal cortex by adopting a brain emotion learning algorithm to obtain a data classification or prediction result.
The specific implementation process of the step 1) comprises the following steps:
1) establishing a three-layer neural network based on a brain emotion learning model, wherein the node number m of an input layer is determined by the characteristic number of a training sample, the node number n of an output layer is determined by the classifiable number of the input sample, and the node number n of an initial hidden layer is determined1From empirical formulas
To obtain, wherein a is [1,10 ]]A constant;
2) removing the reward signals in the basic brain emotion learning model, adopting adaptive genetic algorithm learning to replace reinforcement learning based on the reward signals, and realizing optimization and adjustment of weight values and threshold values of an almond body and an orbital-frontal cortex in a three-layer neural network of the brain emotion learning model: forming chromosomes by weights and thresholds of an input layer, a hidden layer and an output layer of the three layers of neural networks, training the network through a self-adaptive genetic algorithm, and selecting the optimal chromosome, wherein values on the chromosome gene positions correspond to optimized neural network parameters;
3) on the basis of a brain emotion learning model, threshold values are added to the amygdala and the orbital and frontal cortex neurons of a hidden layer of the neural network according to an interaction mechanism between the amygdala and the orbital and frontal cortex in the brain, and the improved neural network based on the brain emotion learning model is obtained.
The threshold value range of the almond body and the orbital and frontal cortex neurons is [ -1,1].
The specific implementation process of the step 3) comprises the following steps:
1) distributing the weight value and the threshold value sequence of the improved neural network on the chromosome gene sequence; the chromosome codes for the form: ch ═ v1,...,vm+1,w1,...,wm,bo,ba]Wherein v is1,...,vm+1Represents the weight between the nodes of the almond body, baIs the amygdaloid neuron threshold, w1,...,wmRepresents the weight between the orbital and frontal cortex nodes, boThe number of genes contained in each chromosome is 2m +3 by the threshold value of the orbital-frontal cortex neurons;
2) initializing weight and threshold values in the chromosome, and calculating improved brain emotion learning network output E by adopting a brain emotion learning algorithm, wherein the calculation formula is as follows:
E＝Ea-Eo
wherein E isaIndicating the output of the almond body, Eo indicating the output of the orbital and frontal cortex, SiTraining samples representing input, AthSignals transmitted to the amygdala body through the thalamus, Ath＝max(S1,S2,...,Sm)；
3) The chromosome is evaluated using a fitness function defined as:
wherein, ChkRepresenting the distribution of weight values and threshold values in chromosomes under the kth input mode; ekThe actual output of the brain emotion learning network in the kth input mode; t iskN represents the number of training samples for the desired output corresponding to the kth input pattern;
4) actual output E using modified neural networkkAnd a target value TkCalculating a chromosome fitness value F, if the error between the current fitness value F and the last iteration fitness value F is greater than an allowable error threshold value, and the requirement of fitness is not met, changing the chromosome by adopting selection, self-adaptive intersection and variation operations in a genetic algorithm, and realizing the updating of the weight and the threshold value on the locus;
calculating and improving the output E of the neural network according to a formula in the step 2), and evaluating the fitness value of the chromosome by using a fitness function in the step 3) until the error between the current iteration result and the last iteration result is smaller than an error threshold value, so that the best chromosome, namely the combination of the optimized weight and the threshold value is obtained;
the error threshold value is 0.0001-0.001.
Compared with the prior art, the invention has the beneficial effects that: 1. the brain emotion learning model is a novel neural network model, the characteristics of low computational complexity and high training speed of the brain emotion learning model are fully utilized, the brain emotion learning network is further optimized by combining the search function of the adaptive genetic algorithm, and the recognition capability of the network is enhanced; 2. compared with a reinforcement learning method based on reward signals, the learning method based on the genetic algorithm can flexibly design a network structure according to training samples, so that sample characteristics are more effectively learned, and the universality of a model is improved. 3. By improving the genetic operator and adopting a self-adaptive intersection and variation mode, the local convergence of the standard genetic algorithm, namely the phenomenon of early maturity, can be avoided to a great extent, so that the global and local search capability of the algorithm is enhanced. The method is used in the fields of machine learning, pattern recognition, data mining and the like, and can greatly improve the rapidity and the accuracy of breast cancer data analysis.
Drawings
FIG. 1 is a block diagram of the basic structure of a brain emotion learning model;
FIG. 2 is an improved prediction network structure based on a brain emotion learning model;
FIG. 3 is an improved classification network structure based on a brain emotion learning model;
FIG. 4 is a flow chart for optimizing a brain emotion learning network based on an adaptive genetic algorithm;
5(a) -5 (c) are classification results of the Breastcancer data set AGA-BEL algorithm; FIG. 5(a) is a fitness function output curve; FIG. 5(b) is a training set classification confusion matrix; FIG. 5(c) is a test set classification confusion matrix;
6(a) -6 (c) show the regression prediction results of the AGA-BEL algorithm for the time series Dst data set; FIG. 6(a) is a mean square error output curve; FIG. 6(b) is a linear correlation analysis of training data; FIG. 6(c) is a linear correlation analysis of test data.
Detailed Description
The present invention includes various examples of classification embodiments and prediction embodiments. The classification test set selects a Breastcancer Data set in a standard machine learning database UCI (https:// architecture. ics. UCI. edu), and the prediction test set selects a Dst Data set of a World Data Center (World Data Center for geomaganetism, Kyoto, Japan) geomagnetic Data sharing platform. The data analysis method proposed in the present invention is explained below with reference to the drawings.
The basic structure of the brain emotion learning model is shown in fig. 1, and the model mainly comprises four main parts, namely thalamus, sensory cortex, orbital-frontal cortex and amygdala, wherein the amygdala and the orbital-frontal cortex are two main parts. Each input node in the almond body receives sensory input signal SI (sensory input), reward signal REW and thalamic signal A simultaneouslyth(ii) a The orbital-frontal cortex receives the sensory input signal SI and the reward signal REW for regulating output. In the process of learning the cerebral emotion, the learning process of the cerebral emotion mainly occurs in the almond body, and the orbital frontal cortex plays a role in monitoring the emotion learning process occurring in the almond body, so that the phenomenon of over-learning and under-learning of the almond body is avoided. The rewarding signal REW determines the change rule of the weight value, and plays an important role in the effect of brain emotion learning. In order to improve the recognition capability of the brain emotion learning model in data analysis, the invention adopts the self-adaptive genetic algorithm to replace the function of a reward signal, and realizes the optimization and adjustment of the weights of the almond body and the orbital frontal cortex in the brain emotion learning network. In addition, according to an interaction mechanism between the almond body and the orbital and frontal cortex in the brain, a threshold value between the almond body and the orbital and frontal cortex is added, and the accuracy of the brain emotion learning network is further improved. FIG. 2 is a diagram of a multi-input single-output improved brain emotion learning network structure for prediction. By extension, a multiple-input multiple-output network structure can be formed for solving the multiple classification problem, as shown in fig. 3.
The brain emotion learning algorithm has the following calculation formula:
in the brain emotion learning network shown in fig. 2, sensory input signals are set as follows:
SI＝[S1,S2,...,Sm]
where m represents the number of input signals. The maximum value of the sensory input signal is transmitted through the thalamus to the amygdala body and can be expressed as:
Ath＝max(S1,S2,...,Sm)
for each input signal SiThe almond bodies all have a corresponding node AiTo receive, expressed as:
wherein v isiAnd representing the weight among the nodes of the almond body, and then the overall output of the almond body is represented as:
for each input signal SiThere is also a corresponding node O in the orbital-frontal cortexiTo receive, expressed as:
Oi＝Si·wi,i＝1,2,...,m
wherein, wiRepresenting the weight between the nodes of the orbital-frontal cortex, the overall output E of the orbital-frontal cortexoExpressed as:
under the combined action of the almond body and the orbital and frontal cortex, the brain emotion learning network output is generated, and the output expression is as follows:
E＝Ea-Eo
(1)
as can be seen from the above description, the learning of cerebral emotion mainly includes the learning of the amygdala and the orbitofrontal cortex. The learning process of the almond body is a dynamic weight value adjusting process; the study of the orbital and frontal cortex is to realize the correction of the study of the almond body by dynamically adjusting all the node weights, so that the almond body learns towards an expected value, and once the system output reaches a target value, the weight adjustment is finished.
The invention adopts the self-adaptive genetic algorithm to optimize and update the weight and the threshold in the brain emotion learning neural network. In the chromosome coding design of the genetic algorithm, a real number coding mode is adopted, and the weight and the threshold value in the brain emotion learning network are distributed on a chromosome gene sequence in sequence. Taking the single-output brain emotion learning network shown in fig. 2 as an example, if m is the number of nodes in the input layer, the chromosome coding form is:
Ch＝[v1,...,vm+1,w1,...,wm,bo,ba]
wherein v is1,...,vm+1Represents the weight between the nodes of the almond body, baIs the amygdaloid neuron threshold, w1,...,wmRepresents the weight between the orbital and frontal cortex nodes, boThe orbital-frontal cortical neuron threshold, each chromosome contains a base factor of 2m + 3.
If the brain emotion learning network is used for solving the multi-classification problem, a single-output neural network needs to be expanded into a multi-output network, as shown in fig. 3.
An adaptation function is used to evaluate the extent to which the chromosome is likely to reach or approach the optimal solution in the optimization calculation. And taking the training samples as the mode input and the expected output of the network, calculating the network output according to the brain emotion learning algorithm, and calculating the error between the actual output and the expected output of the network. Defining the fitness function as:
wherein, ChkRepresenting the weight assignment in the chromosome in the kth input mode, EkThe actual output of the brain emotion learning network in the kth input mode can be calculated according to equation (1). T iskN represents the number of samples for the desired output corresponding to the kth input mode. The problem of solving the minimum value due to the defined fitness functionTherefore, the minimum function output means the best learning effect.
In order to avoid the local convergence of the genetic algorithm, namely the phenomenon of premature, the invention generates new individuals by utilizing the operations of selection, adaptive intersection, adaptive variation and the like, and improves the diversity of the population. In the selection operation, a roulette method and an optimal individual keeping strategy are employed according to fitness values of all individuals. The optimal individuals in the parents are reserved to directly enter the next generation, and then the other individuals are selected by using a roulette method. Therefore, the optimal individuals are guaranteed to participate in genetic operation all the time, the searching speed of the algorithm is improved, the possibility that the individuals with lower fitness are selected is also guaranteed, and the diversity of the individuals is guaranteed. Probability of selection piCan be expressed as:
in the formula (f)iExpressing fitness value of the ith chromosome, FiIs a result calculated by the fitness function shown in equation (2), M represents the population size, and k represents the correlation coefficient.
In the cross operation, firstly two parents for carrying out the cross operation are selected, and then the cross probability p is determined according to the fitness of the parentscThe calculation formula is as follows:
in the formula, pcmaxAnd pcminRespectively a preset maximum cross probability and a preset minimum cross probability. f is the one with high fitness in parents before crossing, fmaxAnd favgThe maximum fitness and the average fitness of the population are respectively. Because the invention adopts real number coding, the arithmetic crossover operator suitable for real number coding is adopted to carry out crossover operation, namely two new individuals are generated through the linear combination of the two individuals, and the definition is as follows:
in the formula, C1And C2Representing two parents, C1' and C2' represents two sub-individuals, alpha is a random number and has a value range of [0,1]。
In the mutation operation, firstly, judging whether each chromosome needs mutation or not, and if so, determining the mutation probability p according to the fitness of a father individualmThe definition is as follows:
in the formula, pmaxAnd pminRespectively, a maximum mutation probability and a minimum mutation probability which are preset. f' is the fitness of the individual needing variation, fmaxAnd favgThe maximum fitness and the average fitness of the population are respectively. And judging whether the individual needs to carry out mutation operation or not according to the mutation probability, and if so, carrying out mutation on the selected gene position by adopting a uniform mutation mode to generate a new individual.
Through the operations of selection, adaptive intersection, adaptive variation and the like, judging whether the output result meets the termination condition, if so, outputting a target solution, namely an optimal weight combination; otherwise, a new round of genetic evolution operation is carried out. And updating the brain emotion learning neural network by using the optimal weight and threshold combination, inputting the test sample into the trained network, and respectively calculating the output of the almond body, the output of the orbital and frontal cortex and the overall output of the network generated under the combined action of the almond body and the orbital and frontal cortex by using a brain emotion learning algorithm to obtain a data analysis result. The specific program flow is shown in FIG. 4.
1. Classification examples
Classification experiments were performed using the BreastCancer data set, which contained 9 features, 699 samples, and 2 classes. The parameter design comprises brain emotion learning neural network design and genetic algorithm design. Number of sensory input signals in brain emotion learning neural network designThe number of the input nodes is determined by the characteristic number of the input training sample, namely the number of the input nodes is set to be 9, the number of the almond body-orbital and frontal cortex output units is determined by the classifiable number of the input sample, namely the number of the output nodes is set to be 2, and the number of the hidden layer nodes is set to be 6. Initializing network weight and threshold value, and the value range is [ -1,1]. In 699 samples of the Breastcancer data set, 70% of sample data is divided into a training set, and 30% of sample data is divided into a test set for classification testing. In the genetic algorithm design, the size of a population is 200, the maximum evolutionary algebra is 80, the length of a chromosome is 21, and the maximum cross probability and the minimum cross probability p are setcmax＝0.8，pcminSet the maximum mutation probability and the minimum mutation probability p as 0.2max＝0.1，pmin0.008. And obtaining a classification result through the brain emotion learning and the genetic algorithm learning.
The classification performance is evaluated from two aspects of rapidity and accuracy. The quickness evaluation includes: calculating time of the training sample and the test sample; the accuracy evaluation comprises the following steps: and the classification accuracy of the training samples and the test samples. In the field of machine learning, evaluation criteria for unbalanced data classification often employ several metrics based on confusion matrices. In the present invention, the classification output result is represented by a confusion matrix shown in table 1, which includes Accuracy (Accuracy), Recall (Recall) and Precision (Precision).
TABLE 1 Classification confusion matrix
Wherein TP (true Positives) represents the number of true positive classes, TN (true negative) represents the number of true negative classes, FN (false negative) represents the number of false negative classes, FP (false Positives) represents the number of false positive classes. In the invention, the classification Accuracy (Accuracy) is a main evaluation index, and the calculation formula is as follows:
the classification output for the Breastcancer dataset is shown in FIG. 5, where FIG. 5a represents the fitness function output curve and the network converges completely when evolving to the 80 th generation. Fig. 5b and 5c show the classification confusion matrix of the training set and the test set, respectively, and the 3 rd row and the 3 rd column of each matrix show the classification correctness, from which it can be seen that the classification correctness of the training set and the test set is 96.1% and 97.6%, respectively.
In order to compare the performance of the BEL model (AGA-BEL) optimized based on the adaptive genetic algorithm, the basic BEL algorithm, the AGA-BEL algorithm and the BP algorithm are respectively adopted to carry out classification experiments on the Breastcancer data set under the same experimental conditions, the average classification accuracy and the average calculation time are obtained through statistics after 50 times of repeated experiments, and the results are shown in Table 2.
TABLE 2 Breastcancer dataset Classification result comparison
From the comparison results in table 2, it can be seen that, in the aspect of rapidity, compared with the conventional BP algorithm, the calculation time of BEL and AGA-BEL is shorter, because the BEL method simulates a mechanism that emotional stimuli cause rapid emotional response in the brain short reflex pathway, the calculation complexity is low, and the operation speed is high. In the aspect of classification accuracy, the AGA-BEL algorithm provided by the invention has obvious advantages, because the BEL model is optimized by adopting the adaptive genetic algorithm in the AGA-BEL, and the classification capability of the model is enhanced.
2. Predictive embodiment
The method provided by the invention is adopted to predict the geomagnetic index describing the geomagnetic activity level. The solar wind disturbance can cause the drastic change of the earth magnetic field, and when the geomagnetic disturbance reaches a peak, a geostationary satellite, radar observation, radio wave communication, power equipment and the like can be influenced, so that various disastrous space environment events are induced. Therefore, the research work of predicting geomagnetic disturbance is carried out, and the establishment of a corresponding early warning system plays an important role in aviation, industry, agriculture and other departments.
The parameter for describing the geomagnetic activity level is mainly a magnetic storm loop current index Dst, which is formed by averaging the offsets of the magnetic field level components measured by the foundation magnetometers near a plurality of magnetic equators. The Dst data set of the invention is derived from a geomagnetic data sharing platform of a world data center. 1000 data of magnetic storm loop current indexes Dst occurring in 2000 years are selected, and the time interval is 1 hour. The 5 data are divided into a plurality of groups as a group, the first 4 data of each group of data are used as training data, the 5 th data are used as target data, and the training data and the target data are input into a brain emotion learning network.
Let the time series representing the Dst index be: dst1,Dst2,Dst3,…,Dstt…, the predicted value Dst at time t + 1t+1The calculation formula is as follows:
Dst(t+1)＝E(Dst(t),Dst(t-1),Dst(t-2),Dst(t-3))
wherein, Dst (t), Dst (t-1), Dst (t-2), Dst (t-3) represents the input of the brain emotion learning network, E represents the output of the brain emotion learning network, and the calculation is carried out according to the formula (1).
The prediction performance evaluation index comprises Mean Square Error (MSE) and linear correlation Coefficient (COR), and the calculation formula is as follows:
where n represents the number of samples, P represents the mode input, T represents the target value, mean () is used for averaging, std () is used for standard deviation.
In the design of a brain emotion learning neural network, the number of input nodes is set to be 4, the number of output nodes is set to be 1, the number of hidden layer nodes is set to be 5, a network weight is initialized, and the value range is [ -1,1]. In the genetic algorithm design, the size of a population is 100, the maximum evolution algebra is 100, the length of a chromosome is 11, and the maximum cross probability and the minimum cross probability are respectively set as pcmax0.7 and pcmin0.1, the maximum and minimum mutation probabilities are pmax0.1 and pmin0.003. The prediction result obtained by learning of the brain emotion and the genetic algorithm is shown in fig. 6, wherein fig. 6(a) represents the mean square error, the network is completely converged when the evolution reaches the 100 th generation, and the mean square error MSE is 0.0014. Fig. 6(b) and 6(c) show the linear correlation coefficients of the training set and the test set, respectively, from which it can be seen that the linear correlation coefficients of the training set and the test set are 0.96163 and 0.97128, respectively.
In order to compare the performance of the AGA-BEL algorithm, the Dst index is predicted by adopting a basic BEL algorithm, an AGA-BEL algorithm and a traditional BP algorithm under the same experimental condition. After 50 times of repeated experiments, the average values of the running algebra, the mean square error and the correlation coefficient are obtained through statistics, and the comparison result is shown in table 3. As can be seen from Table 3, in terms of convergence speed, the number of steps required for convergence is far less for both BEL and AGA-BEL based methods than for BP algorithm; in the aspect of prediction accuracy, the AGA-BEL algorithm has obvious advantages in prediction performance from the aspects of mean square error and linear correlation.
TABLE 3 comparison of prediction results for Dst index dataset
Claims (4)
1. A breast cancer data classification method for optimizing a brain emotion learning model is characterized by comprising the following steps:
1) establishing an improved neural network based on a brain emotion learning model according to the characteristics of training samples, wherein the training samples are 70% of sample data of a breast cancer data set;
2) initializing weights and thresholds of an almond body unit and an orbital-frontal cortex unit in the improved neural network;
3) selecting, self-adaptive crossing and self-adaptive variation operations are carried out by adopting a roulette method and an optimal individual storage strategy in a genetic algorithm, and the weight and the threshold of the almond body and the orbital frontal cortex in the neural network are optimized and improved;
the specific implementation process of the step 3) comprises the following steps:
A. distributing the weight value and the threshold value sequence of the improved neural network on the chromosome gene sequence; the chromosome codes for the form: ch ═ v1,...,vm+1,w1,...,wm,bo,ba]Wherein v is1,...,vm+1Represents the weight between the nodes of the almond body, baIs the amygdaloid neuron threshold, w1,...,wmRepresents the weight between the orbital and frontal cortex nodes, boThe threshold value of the orbital-frontal cortex neuron is defined, m is the number of pathological features of the training sample, and each chromosome contains 2m +3 of basis factors;
B. initializing weight and threshold values in the chromosome, and calculating improved brain emotion learning network output E by adopting a brain emotion learning algorithm, wherein the calculation formula is as follows:
E＝Ea-Eo
wherein E isaIndicating the output of the almond body, EoIndicating orbital and frontal cortex output, SiRepresenting an input training sample, i.e. a pathological feature of breast cancer, AthSignals transmitted to the amygdala body through the thalamus, Ath＝max(S1,S2,...,Sm)；
C. The chromosome is evaluated using a fitness function defined as:
wherein, ChkRepresenting the distribution of weight values and threshold values in chromosomes under the kth input mode; ekThe actual output of the brain emotion learning network in the kth input mode; t iskN represents the number of training samples for the desired output corresponding to the kth input pattern;
D. actual output E using modified neural networkkAnd a target value TkCalculating a chromosome fitness value F, if the error between the current fitness value F and the last iteration fitness value F is greater than an allowable error threshold value, and the requirement of fitness is not met, changing the chromosome by adopting selection, self-adaptive intersection and variation operations in a genetic algorithm, and realizing the updating of the weight and the threshold value on the locus;
calculating and improving the output E of the neural network according to the formula in the step B, and evaluating the fitness value of the chromosome by using the fitness function in the step C until the error between the current iteration result and the last iteration result is less than an error threshold value, so that the optimal chromosome, namely the combination of the optimized weight and the threshold value is obtained;
4) updating the optimized weight and the threshold value into an improved neural network to obtain an optimal network structure;
5) and respectively calculating the output of the almond body, the output of the orbital-frontal cortex and the optimal network overall output under the combined action of the almond body and the orbital-frontal cortex by adopting a brain emotion learning algorithm to obtain a breast cancer data classification result.
2. The method for classifying the breast cancer data for optimizing the brain emotion learning model according to claim 1, wherein the specific implementation process of the step 1) comprises:
1) establishing a three-layer neural network based on a brain emotion learning model, wherein the node number m of an input layer is determined by the pathological feature number of training samples, the node number n of an output layer is determined by the classifiable number of the input samples, and the node number n of an initial hidden layer is determined1From empirical formulas
To obtain a mixture of, among others,a is [1,10 ]]A constant;
2) removing the reward signal in the brain emotion learning model, adopting adaptive genetic algorithm learning to replace reinforcement learning based on the reward signal, and realizing optimization and adjustment of the weight and the threshold of the almond body and the orbital-frontal cortex in the three-layer neural network of the brain emotion learning model: forming chromosomes by weights and thresholds of a neural network input layer, a hidden layer and an output layer, and selecting an optimal chromosome by a training network of a self-adaptive genetic algorithm, wherein values on a chromosome locus correspond to parameters of the improved neural network;
3) on the basis of removing the reward signal of the brain emotion learning model, threshold values are added to the amygdala body and the orbital frontal cortex neurons of the hidden layer of the three-layer neural network according to an interaction mechanism between the amygdala body and the orbital frontal cortex in the brain, and the improved neural network based on the brain emotion learning model is obtained.
3. The method for classifying breast cancer data by optimizing a brain emotion learning model according to claim 1, wherein the threshold value of amygdala and orbitofrontal cortex neurons is in the range of [ -1,1].
4. The method for classifying the breast cancer data for optimizing the brain emotion learning model according to claim 1, wherein the error threshold value is 0.0001-0.001.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710550139.7A CN107392315B (en) | 2017-07-07 | 2017-07-07 | Breast cancer data classification method for optimizing brain emotion learning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710550139.7A CN107392315B (en) | 2017-07-07 | 2017-07-07 | Breast cancer data classification method for optimizing brain emotion learning model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107392315A CN107392315A (en) | 2017-11-24 |
| CN107392315B true CN107392315B (en) | 2021-04-09 |
Family
ID=60334337
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710550139.7A Active CN107392315B (en) | 2017-07-07 | 2017-07-07 | Breast cancer data classification method for optimizing brain emotion learning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107392315B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108958037B (en) * | 2018-08-15 | 2021-06-15 | 厦门理工学院 | Wavelet fuzzy brain emotion learning control method, device, equipment and storage medium |
| CN110442953B (en) * | 2019-07-31 | 2022-11-25 | 东北大学 | Q & P steel design method based on machine learning under guidance of physical metallurgy |
| CN111624880B (en) * | 2020-05-21 | 2021-05-18 | 大连理工大学 | Variable cycle engine multivariable control algorithm based on brain emotion learning model |
| CN111844049B (en) * | 2020-08-04 | 2021-08-17 | 河北省科学院应用数学研究所 | Dexterous hand grabbing control method and device and terminal equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1579383A4 (en) * | 2002-10-24 | 2006-12-13 | Univ Duke | Binary prediction tree modeling with many predictors and its uses in clinical and genomic applications |
| WO2013122617A1 (en) * | 2012-02-15 | 2013-08-22 | Amunix Operating Inc. | Factor viii compositions and methods of making and using same |
| CN104991446A (en) * | 2015-05-21 | 2015-10-21 | 南京航空航天大学 | Unmanned plane thrust direction-changing intelligent control method based on brain emotion learning |
| CN105303252A (en) * | 2015-10-12 | 2016-02-03 | 国家计算机网络与信息安全管理中心 | Multi-stage nerve network model training method based on genetic algorithm |
-
2017
- 2017-07-07 CN CN201710550139.7A patent/CN107392315B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1579383A4 (en) * | 2002-10-24 | 2006-12-13 | Univ Duke | Binary prediction tree modeling with many predictors and its uses in clinical and genomic applications |
| WO2013122617A1 (en) * | 2012-02-15 | 2013-08-22 | Amunix Operating Inc. | Factor viii compositions and methods of making and using same |
| CN104991446A (en) * | 2015-05-21 | 2015-10-21 | 南京航空航天大学 | Unmanned plane thrust direction-changing intelligent control method based on brain emotion learning |
| CN105303252A (en) * | 2015-10-12 | 2016-02-03 | 国家计算机网络与信息安全管理中心 | Multi-stage nerve network model training method based on genetic algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107392315A (en) | 2017-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107392315B (en) | Breast cancer data classification method for optimizing brain emotion learning model | |
| CN105160444B (en) | Electrical equipment failure rate determining method and system | |
| CN109243172A (en) | Traffic flow forecasting method based on genetic algorithm optimization LSTM neural network | |
| CN103971160B (en) | particle swarm optimization method based on complex network | |
| CN110472778A (en) | A kind of short-term load forecasting method based on Blending integrated study | |
| CN102622418B (en) | Prediction device and equipment based on BP (Back Propagation) nerve network | |
| CN110824915B (en) | GA-DBN network-based intelligent monitoring method and system for wastewater treatment | |
| CN106650920A (en) | Prediction model based on optimized extreme learning machine (ELM) | |
| CN107037373A (en) | Battery dump energy Forecasting Methodology based on neutral net | |
| CN114357852A (en) | Layered water injection optimization method based on long-short term memory neural network and particle swarm optimization algorithm | |
| CN109934422A (en) | Neural network wind speed prediction method based on time series data analysis | |
| CN110147890A (en) | A kind of method and system based on lion group's algorithm optimization extreme learning machine integrated study | |
| CN104656620A (en) | Comprehensive evaluation system for remanufacturing of heavy-duty machine tool | |
| CN105404142B (en) | Aluminium electroloysis multi parameters control method based on BP neural network Yu MBFO algorithms | |
| CN106651001A (en) | Needle mushroom yield prediction method based on improved neural network and implementation system | |
| CN114219139A (en) | DWT-LSTM power load prediction method based on attention mechanism | |
| Rozaqi et al. | Comparison between RLS-GA and RLS-PSO for Li-ion battery SOC and SOH estimation: a simulation study | |
| CN116502455A (en) | Process parameter determination method and system for laser selective melting technology | |
| CN113807040B (en) | Optimized design method for microwave circuit | |
| CN107511823A (en) | The method of robot manipulating task track optimizing analysis | |
| CN105389614A (en) | Implementation method for neural network self-updating process | |
| CN105334730B (en) | The IGA optimization T S of heating furnace oxygen content obscure ARX modeling methods | |
| CN109408896B (en) | Multi-element intelligent real-time monitoring method for anaerobic sewage treatment gas production | |
| CN106204314A (en) | A kind of southeast Pacific jumbo flying squid cental fishing ground Forecasting Methodology | |
| CN106250980A (en) | The sliding squid cental fishing ground Forecasting Methodology of a kind of Argentina |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |