Electrochemical sensor data prediction method based on GA-BP neural network

Through the GA-BP neural network model optimized based on genetic algorithm, the amplitude-bias voltage characteristic curve of the EISCAP sensor is used to predict data, which solves the problems of slow measurement speed and low accuracy of traditional methods, and achieves efficient and accurate biochemistry Data detection.

CN118917357APending Publication Date: 2024-11-08XIAN UNIV OF POSTS & TELECOMM
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202410953927.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-17
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Traditional methods require expensive equipment and are slow in measuring biochemical data, and interpolation methods lead to oscillation near data points, affecting measurement accuracy and efficiency.

Method used

The GA-BP neural network model optimized based on genetic algorithm is adopted. By adjusting the weight and threshold of the neural network, the amplitude-bias voltage characteristic curve of the EISCAP sensor is used to predict data, shortening the measurement time and improving accuracy.

Benefits of technology

High-precision prediction is performed without multiple measurement data sets, shortening actual measurement time, improving detection accuracy and efficiency, and having good applicability and flexibility.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917357A_ABST
    Figure CN118917357A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of sensor data prediction, and discloses an electrochemical sensor data prediction method based on a GA-BP neural network. According to the specific technical scheme, the method comprises the following steps: constructing a GA-BP neural network model optimized by a genetic algorithm; determining a corresponding relation between the amplitude and the bias voltage, performing data prediction on amplitude detection data of the electrolyte-insulator-semiconductor capacitor applied to the pH buffer solution by using a GA-BP neural network model, and drawing an amplitude-bias voltage characteristic curve; the method comprises the following steps of: drawing an EISCAP amplitude-bias voltage characteristic curve by utilizing a corresponding relation between the output amplitude of an EISCAP sensor and the bias voltage; according to the method, the working voltage point of the output signal amplitude of the EISCAP sensor is predicted and calculated, the practical application problem is further solved, and the method has good applicability and flexibility, is easy to implement and facilitates subsequent popularization and application.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of sensor data prediction, and in particular relates to a sensor data prediction method based on genetic algorithm optimization of neural network. Background Art

[0002] Electrolyte-Insulator-Silicon Capacitor (EISCAP) is an electrochemical sensor that organically combines biochemical sensitive materials with semiconductor devices. It is widely used in the field of biochemical sensors because of its smooth surface, small size, fast response speed, low power consumption and high sensitivity. However, with the increasing amount of measurement data, the use of traditional measurement methods not only requires expensive and bulky laboratory equipment, such as impedance analyzers or electrochemical workstations, but also has a slow measurement speed and takes a long time. In addition, when using general interpolation methods to fit predicted data, oscillations near the data points will occur, resulting in overfitting, affecting the stability of the interpolation results, and thus affecting the measurement accuracy and detection efficiency of the sensor. Summary of the invention

[0003] In order to solve the technical problems existing in the prior art, the present invention provides an EISCAP sensor data prediction method based on genetic algorithm optimized neural network. The weights and thresholds of the neural network are adjusted and optimized by genetic algorithm on the traditional neural network, and an efficient and time-saving EISCAP sensor data prediction method based on GA-BP is designed.

[0004] To achieve the above purpose, the technical solution adopted by the present invention is: a method for predicting electrochemical sensor data based on GA-BP neural network, the specific steps are as follows:

[0005] Step 1: Construct a GA-BP neural network model optimized by genetic algorithm;

[0006] Step 2. The DC bias voltage extracted from the experimental data sample after data fitting is used as the input feature quantity, and the amplitude detected by the electrolyte-insulator-semiconductor capacitor detection system is used as the output feature quantity to determine the corresponding relationship between the amplitude and the bias voltage. The GA-BP neural network model is used to predict the amplitude detection data of the electrolyte-insulator-semiconductor capacitor applied to the pH buffer solution, and the amplitude-bias voltage characteristic curve is drawn.

[0007] The electrolyte-insulator-semiconductor capacitor detection system includes an electrolyte-insulator-semiconductor capacitor, a data acquisition card and an amplitude detection circuit. Based on the semiconductor field effect, the pH value corresponding to the specific point of the silicon nitride film is converted into an electrical signal, and the electrical signal is input into the data acquisition card to extract the output signal amplitude and draw an amplitude-bias voltage curve.

[0008] The GA-BP neural network model includes an input layer, an output layer, the number of hidden layers, the number of nodes in the hidden layer, network weights and network thresholds. The number of hidden layers Q is determined by formula (1):

[0009]

[0010] The number of nodes q in the hidden layer is determined by formula (2):

[0011]

[0012] Among them, m is the number of input layer nodes, n is the number of output layer nodes, K is the number of standard samples, and a is a constant between 1 and 10.

[0013] The number of nodes in the hidden layer is determined by different pH buffer solutions. The specific steps are as follows:

[0014] Obtaining training samples: measuring the amplitudes of different pH buffer solutions through an electrolyte-insulator-semiconductor capacitor detection system, and using the measured amplitude data as training set samples;

[0015] Extracting feature quantities: Using the measured amplitude data to extract features, using the bias voltage as the input feature quantity, using the amplitude detected by the electrolyte-insulator-semiconductor capacitor detection system as the output feature quantity, and using the normalized inflection point voltage offset as the feature quantity of the buffer concentration;

[0016] Training set samples: Select the fixed-step amplitude measurement data from the data measured for each pH buffer as the training set;

[0017] Test set samples: fixed-step amplitude measurement data is used as the test set;

[0018] Establish a data prediction model: The neural network parameters are constructed through the training network function newff in the GA-BP neural network model. The activation function in the network model maps the nonlinear relationship between the output amplitude and the bias voltage of the electrolyte-insulator-semiconductor capacitor detection system to a high-dimensional feature space. The activation function of the hidden layer is the tansig function, and the activation function of the output layer is the purelin function. The training function trainlm function is used to train the network performance. Finally, the genetic algorithm is used to optimize the weights and thresholds of each neuron node in the neural network, and the entire network is trained, tested and learned.

[0019] The specific steps to establish a data prediction model are as follows:

[0020] Initialize the population: randomly generate a group of individuals and form a population with each individual representing a potential solution. The individual length L is determined according to the structure of the BP neural network. Each individual contains all the weights and thresholds of the neural network.

[0021] The threshold number formula b is:

[0022] b=q+n (3)

[0023] The formula for the number of weights w is:

[0024] w=m×q+q×n (4)

[0025] The individual code length L is:

[0026] L=b+w=q+n+m×q+q×n (5)

[0027] Among them, m is the number of input layer nodes, q is the number of hidden layer nodes, and n is the number of output layer nodes;

[0028] Determine the fitness function: Calculate the fitness value. The specific calculation is shown in formula (6):

[0029]

[0030] Among them, T is the true value, P is the predicted value, and K is the number of samples;

[0031] Selection operation: The selection strategy adopted is the roulette method, whose basic principle is to select based on the fitness ratio of individuals. i The probability of being selected is P i for:

[0032]

[0033] Among them, M is the sum of the fitness values ​​of all individuals in the group, and N is the number of individuals in the population;

[0034] Crossover operation: replace and recombine parts of the structures of two parent individuals to generate new individuals;

[0035] Mutation operation: change the gene value of individual strings in the population;

[0036] Find the optimal solution: decode the optimal solution to obtain the optimal initial weights and thresholds, and train the BP neural network with the optimal initial weights and thresholds again to obtain the fitting model.

[0037] The update formula of weights and thresholds in GA-BP neural network is as follows:

[0038] The weight W connecting the input layer and the hidden layer ij , the update formula is:

[0039]

[0040] in, is the updated weight, is the original weight, α is the learning rate, δ j is the error term of hidden layer neuron j, x i is the output of neuron i in the input layer;

[0041] The threshold b of the input layer and hidden layer j , the update formula is:

[0042]

[0043] in, is the updated threshold, is the original threshold, α is the learning rate, δ j is the error term of hidden layer neuron j;

[0044] The weight W connecting the hidden layer and the output layer jk , the update formula is:

[0045]

[0046] in, is the updated weight, is the original weight, α is the learning rate, δ k is the error term of neuron k in the output layer, y j is the output of hidden layer neuron j;

[0047] The threshold b of the hidden layer and the output layer k , the update formula is:

[0048]

[0049] in, is the updated threshold, is the original threshold, α is the learning rate, δ k is the error term of neuron k in the output layer.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] 1. The present invention utilizes the corresponding relationship between the output amplitude of the EISCAP sensor and the bias voltage to draw an EISCAP amplitude-bias voltage characteristic curve, and performs fitting prediction on the EISCAP sensor measurement data based on the characteristic curve.

[0052] 2. The present invention adopts the GA-BP neural network algorithm model to fit and predict the EISCAP sensor measurement data. There is no need to measure the data set multiple times. The network model can directly perform data fitting and prediction on the DC bias voltage conditions with different steps, shortening the actual measurement time. Compared with the traditional interpolation method, the measurement error is small, which can further improve the detection accuracy and efficiency of the EISCAP sensor.

[0053] 3. The present invention uses a small amount of sensor data to establish a data prediction model through GA-BP neural network algorithm training, accurately predicts EISCAP sensor data, and obtains high-precision detection results using a small amount of sampling data.

[0054] 4. The GA-BP neural network model established in the present invention can predict and calculate the working voltage point of the output signal amplitude of the EISCAP sensor, further solve practical application problems, and has good applicability and flexibility, is easy to implement, and is conducive to subsequent promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 Diagram of the sensor system detection of electrolyte-insulator-semiconductor capacitor.

[0056] Figure 2 This is the topological structure diagram of the GA-BP neural network algorithm.

[0057] Figure 3 This is the overall flow chart of the GA-BP neural network algorithm.

[0058] Figure 4 This is a comparison chart of the predicted and measured curves of BP neural network, GA-BP neural network and interpolation method.

[0059] Figure 5 This is the prediction error curve of BP neural network, GA-BP neural network and interpolation method.

[0060] Figure 6This is a comparison chart of the working voltage points of BP neural network, GA-BP neural network and interpolation method. DETAILED DESCRIPTION

[0061] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0062] like Figure 1 As shown, a data prediction method for an electrochemical sensor based on a GA-BP neural network is used to measure standard buffer solutions of different pH values ​​(pH = 4.0, 6.86, 7.0, 10.01, 12.45) at room temperature using a sinusoidal AC voltage with a frequency of 4 kHz and an amplitude of 20 mV and a DC bias voltage with an amplitude of -2.5 V to 2 V and a step of 0.1 V.

[0063] In this embodiment, several important network parameters in the neural network model must first be determined, including the number of layers and nodes of the input layer, output layer, and hidden layer, as well as the setting of network weights and thresholds. Generally speaking, the number of layers of the input layer and the output layer is usually only one layer, and the number of nodes of the input and output layers is determined by the number of input and output parameters. The number of layers and nodes of the hidden layer are key parts in the design of the neural network. The number of layers of the hidden layer is generally determined by formula (1), and the number of nodes is determined by formula (2):

[0064]

[0065] Among them, m is the number of input layer nodes, n is the number of output layer nodes, K is the number of standard samples, and a is a constant between 1 and 10.

[0066] like Figure 2 As shown, in this model, the number of nodes in the input layer is set to 1, the number of nodes in the output layer is set to 1, and the number of nodes in the hidden layer is determined according to different pH buffer solutions.

[0067] The specific steps are as follows:

[0068] 1) Obtaining training samples: through Figure 1 The EISCAP detection system shown measures the output signal amplitude of five buffer solutions with different pH values ​​(pH=4.0, 6.86, 7.0, 10.01, 12.45), and uses the data measured with a step size of 0.2 as training set samples.

[0069] 2) Feature extraction: The above experimental data are used for feature extraction, with the DC bias voltage as the input feature, the amplitude accurately detected by EISCAP as the output feature, and the normalized inflection point voltage offset as the feature of the buffer concentration.

[0070] 3) Training set samples and test set samples: The amplitude measurement data with a step size of 0.2 from the data measured for each pH buffer solution is selected as the training set, and the amplitude measurement data with a step size of 0.01 is selected as the test set.

[0071] 4) Establishment of data prediction model: The EISCAP data prediction model is established in the MATLAB environment. The nonlinear relationship between the EISCAP output amplitude and the bias voltage is mapped to the high-dimensional feature space through the activation function in the neural network, so that the neural network has the ability to learn nonlinear relationships. Then, the genetic algorithm is used to optimize the weights and thresholds of each neuron node in the neural network to further improve the performance of the network. Finally, the training, testing and learning of the entire network are realized to improve the fitting prediction and generalization capabilities of the model.

[0072] The overall flow chart of the GA-BP neural network algorithm in the present invention is as follows: Figure 3 As shown, the specific implementation steps are as follows:

[0073] Step 1, initialize the population: randomly generate a group of individuals in the EISCAP sensor data prediction model, and form the generated individuals into a population. Each individual in the population represents a potential solution. The individual encoding length L is determined according to the structure of the BP neural network. Each individual contains all the weights and thresholds of the neural network.

[0074] The threshold number formula b is:

[0075] b=q+n (3)

[0076] The formula for the number of weights w is:

[0077] w=m×q+q×n (4)

[0078] The individual code length L is:

[0079] L=b+w=q+n+m×q+q×n (5)

[0080] Among them, m is the number of input layer nodes, q is the number of hidden layer nodes, and n is the number of output layer nodes.

[0081] Step 2, determine the fitness function, which is also called the evaluation function. It is an indicator used to judge the quality of individuals in a group and is evaluated according to the objective function of the problem being solved. In the process of solving the problem, the genetic algorithm basically does not use external information, but only uses the fitness value of individuals in the group to search. The fitness function is usually expressed in the form of an objective function or a cost function. The fitness function of the present invention is shown in formula (7). The fitness value is calculated and the minimum mean square error is solved to achieve the best training effect.

[0082] δ=PT (6)

[0083]

[0084] Among them, T is the true value, P is the predicted value, K is the number of samples, and δ is the sample prediction error.

[0085] Step 3: The selection operation is the process of selecting some individuals that can produce offspring from the current population. The selection strategy adopted in the present invention is the roulette method, which is based on the fitness ratio of the individuals. Each individual X i The probability of being selected is P i for:

[0086]

[0087] Among them, M is the sum of the fitness values ​​of all individuals in the group, and N is the number of individuals in the population.

[0088] Step 4: Crossover operation refers to the operation of replacing and recombining part of the structure of two parent individuals to generate new individuals. The purpose of crossover is to exchange information between individuals. Through crossover, the search ability of genetic algorithm can be rapidly improved.

[0089] Step 5: The basic content of the mutation operation is to change some gene values ​​of individual strings in the group. The mutation operation can enable the genetic algorithm to have local random search capabilities, further maintain the diversity of the group, and prevent premature convergence.

[0090] Step 6: Find the optimal solution. The initial population evolves for several generations, and steps 3-5 are continuously performed until the genetic algorithm converges to the individual with the highest fitness value, that is, the optimal solution. The optimal solution is decoded to obtain the optimal initial weights and thresholds, and the BP neural network with the optimal initial weights and thresholds is trained again to obtain the EISCAP sensor data prediction model based on GA-BP.

[0091] Evaluation of data prediction effect: The training of GA-BP neural network model takes the minimum mean square error between the output value of the model and the actual value as the goal, and uses GA-BP neural network algorithm to update the weights and thresholds in training. The weights and thresholds are continuously adjusted during training until the number of training times is reached or the prediction effect no longer increases, so as to achieve the training and learning of the neural network model. The comparison of the prediction curve and the actual measurement curve of BP neural network model, GA-BP neural network model and polynomial interpolation method is shown in the figure below. Figure 4 As shown in Figure 2, the prediction errors of the BP neural network model, GA-BP neural network model and polynomial interpolation method are as follows: Figure 5 As shown in Figure 2, the working voltage points of the BP neural network model, GA-BP neural network model and polynomial interpolation method are compared with the actual measurement data. Figure 6 As shown in the figure, the GA-BP neural network model has better prediction effect and relatively smaller prediction error. Its operating voltage point is closer to the operating voltage point of the actual measurement data, and can be used in the data fitting prediction of the EISCAP sensor. Furthermore, it can effectively improve the measurement speed and measurement accuracy of the EISCAP sensor.

[0092] Specifically, the update formulas for weights and thresholds in the GA-BP neural network are as follows:

[0093] 1) The weight W connecting the input layer and the hidden layer ij , the update formula is:

[0094]

[0095] in, is the updated weight, is the original weight, α is the learning rate, δ j is the error term of hidden layer neuron j, x i is the output of neuron i in the input layer.

[0096] 2) Threshold b of input layer and hidden layer j , the update formula is:

[0097]

[0098] in, is the updated threshold, is the original threshold, α is the learning rate, δ j is the error term of hidden layer neuron j.

[0099] 3) The weight W connecting the hidden layer and the output layer jk , the update formula is:

[0100]

[0101] in, is the updated weight, is the original weight, α is the learning rate, δ k is the error term of neuron k in the output layer, y j is the output of neuron j in the hidden layer.

[0102] 4) The threshold b of the hidden layer and the output layer k , the update formula is:

[0103]

[0104] in, is the updated threshold, is the original threshold, α is the learning rate, δ k is the error term of neuron k in the output layer.

[0105] The present invention adopts a GA-BP neural network algorithm model to fit EISCAP measurement data, and does not need to measure the data set multiple times. The network model is used to directly fit the data of the bias voltage conditions with different steps, which shortens the actual measurement time, improves the detection accuracy and detection efficiency of EISCAP, has good applicability and flexibility, is easy to implement, and is conducive to subsequent promotion and application.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims

1. A method for predicting electrochemical sensor data based on GA-BP neural network, characterized in that: The specific steps are as follows: Step 1: Construct a GA-BP neural network model optimized by genetic algorithm; Step 2. The DC bias voltage extracted from the experimental data sample after data fitting is used as the input feature quantity, and the amplitude detected by the electrolyte-insulator-semiconductor capacitor detection system is used as the output feature quantity to determine the corresponding relationship between the amplitude and the bias voltage. The GA-BP neural network model is used to predict the amplitude detection data of the electrolyte-insulator-semiconductor capacitor applied to the pH buffer solution, and the amplitude-bias voltage characteristic curve is drawn.

2. The electrochemical sensor data prediction method based on GA-BP neural network according to claim 1 is characterized in that: The electrolyte-insulator-semiconductor capacitor detection system includes an electrolyte-insulator-semiconductor capacitor, a data acquisition card and an amplitude detection circuit. Based on the semiconductor field effect, the pH value corresponding to the specific point of the silicon nitride film is converted into an electrical signal, and the electrical signal is input into the data acquisition card to extract the output signal amplitude and draw an amplitude-bias voltage curve.

3. The electrochemical sensor data prediction method based on GA-BP neural network according to claim 2 is characterized in that: The GA-BP neural network model includes an input layer, an output layer, the number of hidden layers, the number of nodes in the hidden layers, network weights and network thresholds. The number of hidden layers Q is determined by formula (1): The number of nodes q in the hidden layer is determined by formula (2): Among them, m is the number of input layer nodes, n is the number of output layer nodes, K is the number of standard samples, and a is a constant between 1 and 10.

4. The electrochemical sensor data prediction method based on GA-BP neural network according to claim 3 is characterized in that: The number of nodes in the hidden layer is determined by different pH buffer solutions. The specific steps are as follows: Obtaining training samples: measuring the amplitudes of different pH buffer solutions through an electrolyte-insulator-semiconductor capacitor detection system, and using the measured amplitude data as training set samples; Extracting feature quantities: Using the measured amplitude data to extract features, using the bias voltage as the input feature quantity, using the amplitude detected by the electrolyte-insulator-semiconductor capacitor detection system as the output feature quantity, and using the normalized inflection point voltage offset as the feature quantity of the buffer concentration; Training set samples: Select the fixed-step amplitude measurement data from the data measured for each pH buffer as the training set; Test set samples: fixed-step amplitude measurement data is used as the test set; Establish a data prediction model: The neural network parameters are constructed through the training network function newff in the GA-BP neural network model. The activation function in the network model maps the nonlinear relationship between the output amplitude and the bias voltage of the electrolyte-insulator-semiconductor capacitor detection system to a high-dimensional feature space. The activation function of the hidden layer is the tansig function, and the activation function of the output layer is the purelin function. The training function trainlm is used to train the network performance. Finally, the genetic algorithm is used to optimize the weights and thresholds of each neuron node of the neural network, and the entire network is trained, tested and learned.

5. The electrochemical sensor data prediction method based on GA-BP neural network according to claim 4 is characterized in that: The specific steps to establish a data prediction model are as follows: Initialize the population: randomly generate a group of individuals and form a population with each individual representing a potential solution. The individual length L is determined according to the structure of the BP neural network. Each individual contains all the weights and thresholds of the neural network. The threshold number formula b is: b=q+n (3) The formula for the number of weights w is: w=m×q+q×n (4) The individual code length L is: L=b+w=q+n+m×q+q×n (5) Among them, m is the number of input layer nodes, q is the number of hidden layer nodes, and n is the number of output layer nodes; Determine the fitness function: Calculate the fitness value. The specific calculation is shown in formula (6): Among them, T is the true value, P is the predicted value, and K is the number of samples; Selection operation: The selection strategy adopted is roulette method. i The probability of being selected is P i for: Among them, M is the sum of the fitness values ​​of all individuals in the group, and N is the number of individuals in the population; Crossover operation: replace and recombine parts of the structures of two parent individuals to generate new individuals; Mutation operation: change the gene value of individual strings in the population; Find the optimal solution: decode the optimal solution to obtain the optimal initial weights and thresholds, and train the BP neural network with the optimal initial weights and thresholds again to obtain the fitting model.

6. The electrochemical sensor data prediction method based on GA-BP neural network according to claim 3 is characterized in that: The update formula of weights and thresholds in GA-BP neural network is as follows: The weight W connecting the input layer and the hidden layer ij , the update formula is: in, is the updated weight, is the original weight, α is the learning rate, δ j is the error term of hidden layer neuron j, x i is the output of neuron i in the input layer; The threshold b of the input layer and hidden layer j , the update formula is: in, is the updated threshold, is the original threshold, α is the learning rate, δ j is the error term of hidden layer neuron j; The weight W connecting the hidden layer and the output layer jk , the update formula is: in, is the updated weight, is the original weight, α is the learning rate, δ k is the error term of neuron k in the output layer, y j is the output of hidden layer neuron j; The threshold b of the hidden layer and the output layer k , the update formula is: in, is the updated threshold, is the original threshold, α is the learning rate, δ k is the error term of neuron k in the output layer.

Citation Information

Patent Citations

  • Battery SOC estimation method applying multiple models parallelly

    CN108226809A

  • Silicon carbide field effect tube model based on neural network

    CN108875172A

  • Power flow calculation method based on GA-BP neural network

    CN117335425A