BCG信号的波形图 模式识别神经网络模型，作为一种抽象的、参考人脑结构

Waveform diagram of BCG signal The <br><br>pattern recognition <br>neural network model, as an abstract mathematical model that refers to the structure of the human brain, was initiated and developed by modern neuroscience. The application of neural network model is very extensive, for example, it is used in classification and prediction problems whose rules are difficult to describe. BP neural network belongs to one of the more mature and widely used neural network models [17]. The BP neural network model is usually composed of an input layer, a hidden layer and an output layer. It can complete any n-dimensional to m-dimensional mapping, and then realize complex and tedious recognition and classification functions. <br>With the BP neural network model, samples with different characteristics can be classified into pre-calibrated categories. <br>For the purpose of improving the accuracy of the model, the sample usually needs to be divided into a training set and a validation set. <br><br>By inputting the training set, the BP neural network can learn. According to the difference between the calculated output result and the expected value, the weight value of the network is adjusted by back propagation, and then a new round of training is performed on the BP neural network until the training result reaches the pre-error range or Suspend to the preset number of repeated learning [18]. After the above steps, a mature neural network model can be obtained. The quality of the model and whether it needs further adjustment can be judged to pass the validation set. <br>Assuming that each sample is p, the sample error criterion function Ep is [19]: <br><br>and the total error criterion function E of the system is [19]: <br><br>Back propagation is a more critical step in the training process of BP neural network. <br><br>Based on the difference between the calculated result of the output layer and the expected value, the output error of the neurons of each layer of neural network is calculated in reverse layer by layer. <br><br>According to the gradient descent method, the new weight value and threshold value of each layer are calculated. Then, a new round of output value is recalculated according to the new weight value and threshold value, so that the calculated result of the updated neural network is closer to the expected value [20].<br>In order to avoid falling into the local optimal solution, genetic algorithm is introduced on the basis of the original BP neural network. The genetic algorithm first encodes the real problem, simulating the concept of biological population reproduction, and iteratively evolves the population through crossover mutation, and finally obtains the most suitable group, that is, the optimal solution to the real problem [21]. Figure 3 is a flowchart of genetic algorithm-BP neural network.

Wave chart of the BCG signal.<br><br>Pattern recognition.<br>Neural network model, as an abstract mathematical model that refers to the structure of the human brain, was initiated and developed by modern neuroscience. Neural network models are widely used, for example, in classification and prediction problems that are difficult to describe by law. BP neural network belongs to one of the more mature and widely used neural network models. BP neural network model is usually composed of input layer, implicit layer and output layer, which can complete the mapping of any n-dimensional to m-dimensional, and thus realize the function of complex and cumbersome identification and classification.<br>With BP neural network models, samples with different characteristics can be classified into predetermined categories.<br>Samples often need to be divided into training sets and validation sets for the purpose of improving the accuracy of the model.<br><br>By entering the training set, the BP neural network can learn, adjust the weight value of the network according to the difference between the calculated output results and expectations, and then carry out a new round of training on the BP neural network until the training results reach the pre-error range or abort to the pre-set number of repeated learnings. After these steps, a mature neural network model can be obtained. The model is good or bad, and whether further adjustments are required can be judged through the validation set.<br>Assuming that each sample is p, the sample error criterion function Ep is 19:<br><br>And the system's total error criterion function E is 19:<br><br>Reverse propagation is a key step in the training of BP neural networks.<br><br>Based on the difference between the calculation results of the output layer and the expected value, the output error of neurons in each layer of neural network is calculated in reverse layer by layer.<br><br>According to the gradient drop method, the new weight values and thresholds for each layer are calculated. The new round of output values is then recalculated according to the new weight values and thresholds, and the results of the updated neural network are closer to the expected values.<br>In order to avoid getting caught up in the local optimal solution, the genetic algorithm is introduced on the basis of the original BP neural network. The genetic algorithm first encodes the real problem, simulates the concept of biological population reproduction, evolves the population iteratively through cross-mutation, and finally obtains the most suitable population, that is, obtains the optimal solution to the real problem. Figure 3 shows a flowchart of the genetic algorithm-BP neural network.

Waveform of BCG signal<br>pattern recognition<br>Neural network model, as an abstract mathematical model referring to the structure of human brain, is initiated and developed by modern neuroscience. Neural network model is widely used, for example, in the classification and prediction problems which are difficult to describe. BP neural network is one of the more mature and widely used neural network models [17]. BP neural network model is usually composed of input layer, hidden layer and output layer. It can complete the mapping from n-dimension to m-dimension, and then realize the complex recognition and classification functions.<br>With BP neural network model, samples with different characteristics can be classified into pre calibrated categories.<br>In order to improve the accuracy of the model, samples usually need to be divided into training set and verification set.<br>By inputting the training set, the BP neural network can learn. According to the difference between the calculated output and the expected value, the weight value of the network is adjusted by back propagation, and then a new round of training is carried out on the BP neural network until the training result reaches the pre-set error range or stops to the preset repeated learning times [18]. After the above steps, a mature neural network model can be obtained. Whether the model is good or bad and whether it needs further adjustment can be judged by the verification set.<br>Suppose each sample is p, then the sample error criterion function EP is [19]:<br>And the total error criterion function e of the system is [19]:<br>Back propagation is a key step in the process of BP neural network training.<br>Based on the difference between the calculated results of the output layer and the expected value, the output error of each layer of neural network is calculated layer by layer.<br>According to the gradient descent method, the new weight value and threshold value of each layer are calculated. Then, a new round of output value is recalculated according to the new weight value and threshold value, so that the calculation result of the updated neural network is closer to the expected value [20].<br>In order to avoid falling into the local optimal solution, genetic algorithm is introduced based on the original BP neural network. The genetic algorithm first encodes the real problem, simulates the concept of biological population reproduction, iteratively evolves the population through cross mutation, and finally obtains the most suitable population, that is, to obtain the optimal solution of the real problem [21]. Figure 3 is the flow chart of genetic algorithm BP neural network.<br>