CN113176022B - Segmented neural network pressure sensor pressure detection method and system - Google Patents
Segmented neural network pressure sensor pressure detection method and system Download PDFInfo
- Publication number
- CN113176022B CN113176022B CN202110516026.1A CN202110516026A CN113176022B CN 113176022 B CN113176022 B CN 113176022B CN 202110516026 A CN202110516026 A CN 202110516026A CN 113176022 B CN113176022 B CN 113176022B
- Authority
- CN
- China
- Prior art keywords
- neural network
- pressure
- pressure sensor
- training
- section
- 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
- G01—MEASURING; TESTING
- G01L—MEASURING FORCE, STRESS, TORQUE, WORK, MECHANICAL POWER, MECHANICAL EFFICIENCY, OR FLUID PRESSURE
- G01L1/00—Measuring force or stress, in general
- G01L1/20—Measuring force or stress, in general by measuring variations in ohmic resistance of solid materials or of electrically-conductive fluids; by making use of electrokinetic cells, i.e. liquid-containing cells wherein an electrical potential is produced or varied upon the application of stress
- G01L1/22—Measuring force or stress, in general by measuring variations in ohmic resistance of solid materials or of electrically-conductive fluids; by making use of electrokinetic cells, i.e. liquid-containing cells wherein an electrical potential is produced or varied upon the application of stress using resistance strain gauges
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01L—MEASURING FORCE, STRESS, TORQUE, WORK, MECHANICAL POWER, MECHANICAL EFFICIENCY, OR FLUID PRESSURE
- G01L9/00—Measuring steady of quasi-steady pressure of fluid or fluent solid material by electric or magnetic pressure-sensitive elements; Transmitting or indicating the displacement of mechanical pressure-sensitive elements, used to measure the steady or quasi-steady pressure of a fluid or fluent solid material, by electric or magnetic means
- G01L9/02—Measuring steady of quasi-steady pressure of fluid or fluent solid material by electric or magnetic pressure-sensitive elements; Transmitting or indicating the displacement of mechanical pressure-sensitive elements, used to measure the steady or quasi-steady pressure of a fluid or fluent solid material, by electric or magnetic means by making use of variations in ohmic resistance, e.g. of potentiometers, electric circuits therefor, e.g. bridges, amplifiers or signal conditioning
- G01L9/04—Measuring steady of quasi-steady pressure of fluid or fluent solid material by electric or magnetic pressure-sensitive elements; Transmitting or indicating the displacement of mechanical pressure-sensitive elements, used to measure the steady or quasi-steady pressure of a fluid or fluent solid material, by electric or magnetic means by making use of variations in ohmic resistance, e.g. of potentiometers, electric circuits therefor, e.g. bridges, amplifiers or signal conditioning of resistance-strain gauges
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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 and a system for detecting pressure of a segmented neural network pressure sensor, which comprises the steps of setting a BP neural network in a segmented manner, determining a training function, determining an optimal hidden layer and optimizing a genetic algorithm; and sequentially dividing the detection range of the piezoresistive pressure sensor into a low section and a high section according to the detection range of the piezoresistive pressure sensor, respectively setting BP neural networks, determining a proper neural network training function and the optimal number of hidden layers, improving the identification accuracy, optimizing the initial weight and the threshold of the BP neural network with the optimal number of hidden layers through a genetic algorithm, and determining a segmented neural network model optimized based on the genetic algorithm. The advantages are that: the method has the advantages that the local search capability of the BP neural network is fully exerted, the stability of the algorithm is improved, the algorithm is prevented from falling into a local optimal value, and the output readability, the pressure identification precision, the pressure identification speed and the pressure identification stability of the piezoresistive pressure sensor are effectively improved.
Description
Technical Field
The invention relates to a segmented neural network pressure sensor pressure detection method and system, and belongs to the technical field of sensors.
Background
With the advent of the world of everything interconnection, human beings begin to use various sensors to sense external changes on a large scale to realize informatization and intellectualization. The pressure identification is used as the core of the interconnection of everything and plays an extremely important role in the Internet of things. In the intelligent robot, pressure identification is the key to realize accurate completion of various interaction tasks. In an intelligent automobile, pressure identification is the core of key functions such as tire pressure detection and the like. Pressure identification is also widely applied in the fields of intelligent medical treatment, biological measurement and the like.
The pressure recognition means that a pressure sensor applies an arbitrary pressure and recognizes a specific applied pressure value from an output of the pressure sensor. When pressure is applied to a piezoresistive pressure sensor, the resistance of the strained material in the sensor changes due to its piezoresistive effect, and its resistance change is related to the amount of pressure applied. With the progress of research on pressure sensors, more and more excellent pressure sensors having high detection sensitivity, a large detection range and strong stability have been proposed. However, the input/output relationship of these pressure sensors is nonlinear, and it is difficult to read the external pressure value in actual use, which results in poor readability and practicability. Although various pressure sensors having excellent performance have been proposed, it is difficult to apply them to production and life because of poor readability of their output.
In the field of neural networks for pressure sensing, there is a lack of research on neural networks for piezoresistive pressure sensor pressure sensing. For example, a method (patent application No. 201910226605.5) for realizing piezoelectric type pressure detection touch screen piezoelectric response homogenization by using a neural network establishes a mapping relation between an inhomogeneous piezoelectric response and a user touch screen position and force magnitude by training the neural network, and maps the piezoelectric response to the user touch screen position and the touch screen pressure magnitude by using the trained neural network. The piezoelectric pressure detection is realized by combining the neural network with the piezoelectric response, and the combination of the neural network and the pressure detection of the piezoresistive pressure sensor is not researched. In the existing research of combining the neural network with the pressure detection, the BP neural network is mostly directly used for the pressure detection, the optimization of the BP neural network is lacked, the network learning is easy to fall into a local minimum value, the stability of the network is reduced, and the network learning time is greatly prolonged. For example, a pulsating pressure prediction method based on an artificial neural network (patent application number 202011306306879.4), the pulsating pressure on the surface of an aircraft under the condition of incoming flow is directly predicted through a BP neural network, and the prediction precision is low and the stability is poor. The segmented neural network pressure sensor pressure detection method based on genetic algorithm optimization not only fills the blank of the research field that a neural network is used for piezoresistive pressure sensor pressure detection, but also optimizes the initial weight and the threshold of a BP neural network through the genetic algorithm so as to give full play to the local search capability of the BP neural network, improve the stability of the algorithm and avoid the algorithm from falling into a local optimal value.
In a piezoresistive pressure sensor, effective identification of pressure is a complex and challenging detection problem, and the main difficulty is three aspects, on one hand, because the piezoresistive effect of a strain material in the pressure sensor is nonlinear along with the change of an external force, the output readability of the strain material is poor, and although the strain material has excellent performance, the strain material cannot be used. On one hand, due to the lack of an effective and quick output reading method of the piezoresistive pressure sensor, the identification speed is slow in the actual use process, and the time delay of using the pressure sensor system is increased. On the other hand, since the pressure applied to the pressure sensor may fluctuate to a certain extent during the actual use of the pressure sensor, such disturbances increase the difficulty of pressure identification.
The BP (Back Propagation) neural network is one of the most widely used artificial neural network models at present, is a multi-layer feedforward network trained under the guidance of an error Back Propagation algorithm, and has a three-layer structure of an input layer, a hidden layer and an output layer. By training and learning given data, the relationship between unknown input and output can be approximated and simulated. The network takes a steepest descent method as a learning rule, the output error of the network reversely propagates along the network layer by layer, each layer of neuron continuously adjusts the connection weight and the threshold according to the error relation, the aim of continuously reducing the output error of the network is taken, the minimum value of the error square sum is finally reached, and the high-precision input and output mapping relation is obtained. The BP neural network exhibits good characteristics in classification and recognition, and thus, is used by researchers in the fields of vehicle recognition, face detection and the like. The BP neural network is used as an efficient information processing algorithm, can model the internal connection of data, has good comprehensive capability of processing disordered data, can be used for operation classification and prediction of the data, and is widely applied to the fields of intelligent interaction, biology, medical treatment and the like.
The genetic algorithm is a highly parallel global probability search algorithm established by simulating the heredity and evolution processes of a population in the nature, and reflects a natural law of excellence and rejection and survival of suitable persons. The genetic algorithm has high operation efficiency, can process problems in parallel, searches from the global aspect, can actively learn and accumulate the spatial knowledge in the search, and continuously reduces the search range to finally reach the optimal solution. The genetic algorithm maps the problem to be solved into the natural environment of the population, each potential solution of the problem represents one individual in the population, then the set of all potential solutions of the problem forms the population, and the algorithm calculates the fitness value of each individual based on the fitness function to measure the adaptability of the individual to the environment. During algorithm operation, starting from an initial population generated randomly, generating a new population with stronger survival capability in the environment by selecting, crossing and mutating three genetic operators according to the fitness value of each individual in the population evolution process, and gradually reducing the population search space to better and better areas. Along with the iteration of the algorithm, the population fitness value is continuously improved, and finally, the population fitness value is converged to the individual with the highest environmental adaptability and the strongest survival ability, so that the optimal solution of the problem is obtained.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a method and a system for detecting the pressure of a segmented neural network pressure sensor.
In order to solve the technical problem, the invention provides a pressure detection method for a segmented neural network pressure sensor, which comprises the following steps:
acquiring resistance change data of a strain material based on a pressure applied to the piezoresistive pressure sensor;
inputting the resistance change data into a predetermined segmented neural network model optimized based on a genetic algorithm, and outputting accurate pressure data;
the determination process of the segmented neural network model based on genetic algorithm optimization comprises the following steps:
according to the pressure identification range of the piezoresistive pressure sensor, the low-section pressure identification range and the high-section pressure identification range are sequentially divided from the middle point of the pressure identification range, and a corresponding low-section BP neural network and a corresponding high-section BP neural network are respectively arranged in the low-section pressure identification range and the high-section pressure identification range and are respectively used for identifying pressure data in the low-section pressure identification range and pressure data in the high-section pressure identification range of the piezoresistive pressure sensor;
acquiring a training function;
acquiring the predetermined optimal number of hidden layers of a low-section BP neural network and the predetermined optimal number of hidden layers of a high-section BP neural network;
and respectively optimizing the low-segment BP neural network and the high-segment BP neural network which determine the training function and the optimal number of hidden layers by utilizing a genetic algorithm to determine an optimal BP neural network model.
Further, the training function is one of a Bayesian rule method, a Levenberg-Marquardt algorithm, a normalized conjugate gradient method and an elastic BP algorithm.
Further, the determination process of the training function includes:
acquiring sample data of the piezoresistive pressure sensor of the type to be tested, and dividing the sample data into a training set, a verification set and a test set;
acquiring different training functions, including a Bayesian rule method, a Levenberg-Marquardt algorithm, a normalized conjugate gradient method and an elastic BP algorithm;
and training the neural network according to the training set and each training function, and comprehensively considering the iteration times and the error to determine the final training function.
Further, the determining process of the optimal number of hidden layers of the low-segment BP neural network and the optimal number of hidden layers of the high-segment BP neural network includes:
circularly training the BP neural network under different number hidden layers for a plurality of times;
and respectively determining the optimal number of hidden layers of the low-section BP neural network and the optimal number of hidden layers of the high-section BP neural network by using the principle of minimum error according to the result of the cyclic training processing.
Further, the optimal BP neural network model further includes determination of an optimal genetic algebra, and the process includes:
and respectively optimizing the low-segment BP neural network and the high-segment BP neural network which determine the training function and the optimal number of hidden layers by utilizing a genetic algorithm under different genetic algebras, and determining an optimal BP neural network model with the optimal number of hidden layers.
A segmented neural network pressure sensor pressure detection system, comprising:
a first acquisition module that acquires resistance change data of a strain material based on pressure applied to the piezoresistive pressure sensor;
the processing module is used for inputting resistance change data of a strain material based on pressure applied to the piezoresistive pressure sensor into a predetermined segmented neural network model optimized based on a genetic algorithm and outputting accurate pressure data;
the processing module comprises a model determination module comprising:
the model segmentation module is used for sequentially dividing the pressure identification range of the piezoresistive pressure sensor into a low-section pressure identification range and a high-section pressure identification range from the middle point of the pressure identification range according to the pressure identification range of the piezoresistive pressure sensor, setting a corresponding low-section BP neural network and a corresponding high-section BP neural network for the low-section pressure identification range and the high-section pressure identification range respectively, and identifying pressure data of the low-section pressure identification range and pressure data of the high-section pressure identification range of the piezoresistive pressure sensor respectively;
the second acquisition module is used for acquiring a training function;
a third obtaining module, configured to obtain the predetermined optimal number of hidden layers of the low-segment BP neural network and the predetermined optimal number of hidden layers of the high-segment BP neural network;
and the optimization module is used for respectively optimizing the low-segment BP neural network and the high-segment BP neural network which determine the training function and the optimal number of the hidden layers by utilizing a genetic algorithm to determine an optimal BP neural network model.
Further, the training function is one of a Bayesian rule method, a Levenberg-Marquardt algorithm, a normalized conjugate gradient method and an elastic BP algorithm.
Further, the second obtaining module includes:
acquiring sample data of the piezoresistive pressure sensor of the type to be tested, and dividing the sample data into a training set, a verification set and a test set;
acquiring different training functions, including a Bayes rule method, a Levenberg-Marquardt algorithm, a normalized conjugate gradient method and an elastic BP algorithm;
and training the neural network according to the training set and each training function, and comprehensively considering the iteration times and the error to determine the final training function.
Further, the third obtaining module includes:
the cyclic training module is used for circularly training the BP neural network under different numbers of hidden layers for a plurality of times;
and respectively determining the optimal number of hidden layers of the low-section BP neural network and the optimal number of hidden layers of the high-section BP neural network by using the principle of minimum error according to the result of the circular training process.
Further, the optimization module also comprises a genetic algebra determination module,
the method is used for respectively optimizing the low-segment BP neural network and the high-segment BP neural network which determine the training function and the optimal number of hidden layers by utilizing a genetic algorithm under different genetic algebras, and determining the optimal BP neural network model with the optimal number of hidden layers.
The invention has the following beneficial effects:
according to the pressure detection method of the segmented neural network pressure sensor based on genetic algorithm optimization, pressure identification is carried out through the BP neural network, the output readability of the piezoresistive pressure sensor is greatly improved, and the problem that the piezoresistive pressure sensor cannot be used in actual production life due to poor readability is solved.
The invention provides a segmented neural network pressure sensor pressure detection method based on genetic algorithm optimization, and provides a method for quickly and accurately reading the output of a piezoresistive pressure sensor.
The pressure detection method of the segmented neural network pressure sensor based on genetic algorithm optimization provided by the invention effectively improves the pressure identification precision and speed by dividing the pressure identification range into a low segment and a high segment for segmented neural network identification.
The pressure detection method of the segmented neural network pressure sensor based on genetic algorithm optimization optimizes initial weights and threshold values of the segmented neural network with the optimal number of hidden layers through a genetic algorithm, so that the local search capability of a BP (back propagation) neural network is fully exerted, the stability of the algorithm is improved, and the algorithm is prevented from falling into a local optimal value.
According to the pressure detection method of the segmented neural network pressure sensor based on genetic algorithm optimization, when the initial weight and the threshold of the segmented neural network with the optimal number of hidden layers are optimized through a genetic algorithm, the neural network under different genetic algebra is repeatedly trained for a plurality of times, the genetic algebra and errors are comprehensively considered, the optimal genetic algebra is selected, the accuracy of the calculation result of the neural network can be guaranteed, and the calculation time of the neural network can be reduced.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a graph of deformation of a graphene film in a graphene pressure sensor as a function of applied pressure;
FIG. 3 is a schematic diagram of a BP neural network segment of the present invention;
FIG. 4 is a comparison graph of a BP neural network using different training functions;
FIG. 5 is a graph comparing the performance of BP neural network corresponding to low-stage pressure identification in different hidden layer numbers in the segmented neural network of the present invention;
FIG. 6 is a graph comparing the performance of BP neural network corresponding to high-stage pressure identification in different hidden layer numbers in the segmented neural network of the present invention;
FIG. 7 is a flow chart of the present invention for optimizing a segmented neural network by a genetic algorithm;
FIG. 8 is a graph comparing the performance of BP neural network under different genetic algorithms of the present invention;
FIG. 9 is a graph comparing error of calculation results of a segmented neural network and a segmented neural network optimized by a genetic algorithm under different levels of interference.
Detailed Description
In order to make the objects, features and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the embodiments described below are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
FIG. 1 is a flow chart of the invention, and the method for detecting pressure of a segmented neural network pressure sensor based on genetic algorithm optimization comprises the steps of setting a BP neural network in a segmented manner, determining a training function, determining an optimal hidden layer and optimizing a genetic algorithm.
In the piezoresistive effect of graphene, an external force causes the graphene film to deform, the graphene deformation causes the resistance change of the graphene film, and the deformation of the graphene film is related to the resistance change of the graphene film. The relation between the deformation of the graphene film in the graphene pressure sensor and the applied pressure is shown in fig. 2, along with the increase of the pressure, the change rate of the deformation of the graphene film is reduced, the sensitivity of the deformation of the graphene film along with the change of the pressure is reduced, and the input/output change curve of the graphene pressure sensor is nonlinear. Firstly, the pressure identification range of the piezoresistive pressure sensor is divided into a low section and a high section in sequence, BP neural network identification is carried out in a segmentation mode to improve the pressure identification performance, and fig. 3 is a segmentation schematic diagram of the pressure identification range.
In the BP neural network training process, 200 groups of samples are obtained by simulating the deformation displacement of the graphene film within the pressure range of 0-1000kPa by using the established high-precision graphene pressure sensor simulation model with the step length of 5 kPa. Wherein 140 groups are used as training set, 30 groups are used as verification set, and 30 groups are used as test set.
BP neural network-based pressure identification is essentially a mapping of pressure sensor inputs to outputs that is able to learn a large number of mapping relationships between pressure sensor inputs and outputs without any precise mathematical expressions between inputs and outputs, and the network has the ability to map pressure sensor inputs to outputs as long as the neural network is trained with the appropriate methods. Before training is started, all weights should be initialized with some different small random number. The small random number is used for ensuring that the network does not enter a saturation state due to overlarge weight value, so that training fails; "different" is used to ensure that the network can learn properly. In fact, if the weight matrix is initialized with the same number, there is symmetry, resulting in the same for each layer, and the network is not able to learn.
The training process of the BP neural network comprises the following two steps: (1) forward propagation: the input signal acts on the output node through the hidden layer, the output signal is generated through nonlinear transformation, and if the actual output does not accord with the expected output, the process of error back propagation is carried out; (2) back propagation of the error: and reversely transmitting the output error to the input layer by layer through the hidden layer, and distributing the error to all units of each layer, wherein error signals obtained from each layer are used as a basis for adjusting the weight of each unit.
The BP neural network can obtain a change relation curve of the pressure sensor according to the input and output samples of the pressure sensor. The training function of the network will affect its recognition accuracy. Therefore, a proper training function needs to be selected to obtain the BP neural network with excellent performance. Due to the fact that differences among different samples are large, training functions need to be selected according to specific samples, the neural network used for the graphene piezoresistive pressure sensor is trained through a Bayes rule method, a Levenberg-Marquardt algorithm, a normalized conjugate gradient method and an elastic BP algorithm, the training result is shown in figure 4, the abscissa of the training result is the training times of the BP neural network, and the ordinate of the training result is the mean square error. As can be seen from fig. 4, as the number of training times increases, the mean square error of the neural network calculation result decreases, and the recognition accuracy significantly improves. And training the BP neural network by respectively adopting a Bayes rule method, a Levenberg-Marquardt algorithm, a normalized conjugate gradient method and an elastic BP algorithm. And (3) training the neural network by adopting a Levenberg-Marquardt algorithm, wherein the iteration number is 1000, the error is 0.0000442, and comprehensively considering that the Levenberg-Marquardt algorithm is high in training speed and small in error, the Levenberg-Marquardt algorithm is selected as the neural network for the graphene piezoresistive pressure sensor.
The number of the hidden layers has great influence on the recognition performance of the BP neural network, the recognition accuracy of the neural network is too low due to the fact that the number of the hidden layers is too small, and the complexity of the network is increased due to the fact that the number of the hidden layers is too large. Therefore, choosing the appropriate number of hidden layers is crucial for the BP neural network. Before the BP neural network begins to train, the weight is initialized by random numbers, and the recognition performance of the BP neural network fluctuates within a certain range. The proper number of hidden layers are determined by training the BP neural network under different numbers of hidden layers circularly for 5 times and comparing the average recognition performance of the BP neural network, and the comparison result is shown in fig. 5 and fig. 6.
As shown in fig. 5 and 6, in the segmented neural network, the BP neural network under different number of hidden layers is trained circularly 5 times. The result shows that when the number of the hidden layers is 3, the error of the BP neural network corresponding to the low-stage pressure identification is only 0.00000595; when the number of hidden layers is 5, the error of the BP neural network corresponding to the high-section pressure identification is only 0.00004770; therefore, in the segmented neural network for the graphene piezoresistive pressure sensor, 3 layers of hidden layers are selected from the BP neural network corresponding to low-stage pressure identification, and 5 layers of hidden layers are selected from the BP neural network corresponding to high-stage pressure identification, so that the identification rate of the BP neural network can be improved.
The comparison result of the pressure identification performance of the segmented and whole-segmented BP neural networks is shown in the table 1. It can be known from the observation of table 1 that the pressure recognition rate can be improved by performing segmented neural network recognition on the piezoresistive pressure sensor, and the pressure recognition rate can be improved by performing genetic algorithm optimization on the segmented neural network.
TABLE 1 comparison table of BP neural network pressure identification performance before and after optimization
The flow chart of the genetic algorithm is shown in fig. 7, firstly, a parameter coding rule is determined according to the topological structure design of the BP neural network; randomly generating long strings of N parameters according to a designed coding rule to form an initial population of the algorithm; the algorithm starts from an initial population, new individuals are generated by continuously using genetic operators, and the overall population fitness is improved until the algorithm meets a termination condition. After the genetic algorithm part is terminated, the optimal individuals are decoded into a group of distribution of BP neural network connection weight values and threshold values by contrasting the coding rules, the distribution is used as an initial parameter of the network, and the algorithm enters the BP neural network part. And starting from the initial parameters, the BP neural network calculates the network output error by using the training data set, reversely transmitting the error if the termination condition is not met, and correcting the weight and the threshold until the error meets the condition to terminate the network learning, thereby obtaining a final calculation model.
The genetic algorithm simulates the problem into a biological evolution process, generates a next generation solution through operations such as copying, crossing, mutation and the like, gradually eliminates the solution with a low fitness function value, and increases the solution with a high fitness function value. Therefore, the individual with high fitness function value is likely to be evolved after evolution for several generations. The more genetic algebra, the smaller the optimized network output error, but the longer the computation time. And determining the optimal genetic algebra by comparing the fluctuation errors of the calculation results of the BP neural network after 1-20 generations of genetic algorithm optimization. As shown in fig. 8, when the number of genetic generations is 10, the fluctuation error of the calculation result of the optimized BP neural network is stabilized around 0.002. Therefore, the genetic algebra is selected to be 10 generations, so that the accuracy of the calculation result of the neural network can be ensured, and the calculation time of the neural network can be reduced. The segmented neural network optimized by the genetic algorithm has stronger stability compared with the prior segmented neural network, and the error of the calculation results of the two neural networks under different levels of interference is shown in FIG. 9.
Correspondingly, the invention also provides a segmented neural network pressure sensor pressure detection system, which comprises:
a first acquisition module that acquires resistance change data of a strain material based on a pressure applied to the piezoresistive pressure sensor;
the processing module is used for inputting resistance change data of a strain material based on pressure applied to the piezoresistive pressure sensor into a predetermined segmented neural network model optimized based on a genetic algorithm and outputting accurate pressure data;
the model segmentation module is used for sequentially dividing the pressure identification range into a low-section pressure identification range and a high-section pressure identification range from the midpoint of the pressure identification range according to the pressure identification range of the piezoresistive pressure sensor, and respectively setting a corresponding low-section BP neural network and a corresponding high-section BP neural network for the low-section pressure identification range and the high-section pressure identification range, and respectively identifying pressure data of the low-section pressure identification range and pressure data of the high-section pressure identification range of the piezoresistive pressure sensor;
the second acquisition module is used for acquiring a training function;
the third acquisition module is used for acquiring the predetermined optimal number of hidden layers of the low-section BP neural network and the predetermined optimal number of hidden layers of the high-section BP neural network;
and the optimization module is used for respectively optimizing the low-segment BP neural network and the high-segment BP neural network which determine the training function and the optimal number of the hidden layers by utilizing a genetic algorithm to determine an optimal BP neural network model.
The training function is one of a Bayes rule method, a Levenberg-Marquardt algorithm, a normalized conjugate gradient method and an elastic BP algorithm.
The second acquisition module comprises:
the sample data and training function acquisition module is used for acquiring sample data of the piezoresistive pressure sensor to be tested and dividing the sample data into a training set, a verification set and a test set; acquiring different training functions, including a Bayesian rule method, a Levenberg-Marquardt algorithm, a normalized conjugate gradient method and an elastic BP algorithm;
and the training module is used for training the neural network according to the training set and each training function, and comprehensively considering the iteration times and the errors to determine the final training function.
The process for determining the optimal number of hidden layers of the low-segment BP neural network and the optimal number of hidden layers of the high-segment BP neural network comprises the following steps:
circularly training the BP neural network under different number hidden layers for a plurality of times;
and respectively determining the optimal number of hidden layers of the low-section BP neural network and the optimal number of hidden layers of the high-section BP neural network by using the principle of minimum error according to the result of the circular training process.
The third obtaining module includes:
the cyclic training module is used for circularly training the BP neural network under different numbers of hidden layers for a plurality of times;
and the comparison module is used for respectively determining the optimal hidden layer number of the low-section BP neural network and the optimal hidden layer number of the high-section BP neural network according to the result of the cyclic training processing by using the principle of minimum error.
The optimization module further comprises a genetic algebra determination module,
the method is used for optimizing the low-segment BP neural network and the high-segment BP neural network which determine the training function and the optimal number of the hidden layers by utilizing a genetic algorithm under different genetic algebras respectively, and determining the optimal BP neural network model with the optimal number of the hidden layers.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above-mentioned embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the same; although the present invention has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A segmented neural network pressure sensor pressure detection method is characterized by comprising the following steps:
acquiring resistance change data of a strain material based on pressure applied to a piezoresistive pressure sensor;
inputting the resistance change data into a predetermined segmented neural network model optimized based on a genetic algorithm, and outputting accurate pressure data;
the determination process of the segmented neural network model based on genetic algorithm optimization comprises the following steps:
according to the pressure identification range of the piezoresistive pressure sensor, the low-section pressure identification range and the high-section pressure identification range are sequentially divided from the middle point of the pressure identification range, and a corresponding low-section BP neural network and a corresponding high-section BP neural network are respectively arranged in the low-section pressure identification range and the high-section pressure identification range and are respectively used for identifying pressure data in the low-section pressure identification range and pressure data in the high-section pressure identification range of the piezoresistive pressure sensor;
acquiring a training function;
acquiring the predetermined optimal number of hidden layers of a low-section BP neural network and the predetermined optimal number of hidden layers of a high-section BP neural network;
respectively optimizing the low-segment BP neural network and the high-segment BP neural network which determine the training function and the optimal number of hidden layers by utilizing a genetic algorithm to determine an optimal BP neural network model;
the optimization process of the genetic algorithm comprises the following steps:
determining a parameter coding rule according to the structure of the segmented neural network model; randomly generating long strings of N parameters according to a parameter coding rule to form an initial population of the algorithm; starting from the initial population, continuously using genetic operators to generate new individuals, and improving the overall fitness of the population until the algorithm meets the termination condition; after the termination condition is met, decoding the optimal individual according to a parameter coding rule into a group of distribution of a sectional BP neural network connection weight and a threshold value, and taking the distribution as an initial parameter of the sectional BP neural network; starting from initial parameters, the segmented BP neural network calculates network output errors by using a training data set, reversely transmits the errors and corrects the weight and the threshold if the termination condition is not met, and terminates the network learning until the errors meet the condition, thereby obtaining an optimal BP neural network model.
2. The segmented neural network pressure sensor pressure detection method of claim 1, wherein the training function is one of a Bayesian rule method, a Levenberg-Marquardt algorithm, a normalized conjugate gradient method, and an elastic BP algorithm.
3. The segmented neural network pressure sensor pressure sensing method of claim 2, wherein the training function determination process comprises:
acquiring sample data of the piezoresistive pressure sensor of the type to be tested, and dividing the sample data into a training set, a verification set and a test set;
acquiring different training functions, including a Bayesian rule method, a Levenberg-Marquardt algorithm, a normalized conjugate gradient method and an elastic BP algorithm;
and training the neural network according to the training set and each training function, and comprehensively considering the iteration times and the error to determine the final training function.
4. The segmented neural network pressure sensor pressure detection method of claim 1, wherein the determination process of the optimal number of hidden layers of the low-segment BP neural network and the optimal number of hidden layers of the high-segment BP neural network comprises:
circularly training BP neural networks under different numbers of hidden layers for a plurality of times;
and respectively determining the optimal number of hidden layers of the low-section BP neural network and the optimal number of hidden layers of the high-section BP neural network by using the principle of minimum error according to the result of the cyclic training processing.
5. The segmented neural network pressure sensor pressure sensing system of claim 1, wherein the optimal BP neural network model further comprises determination of optimal genetic algebra, and the process comprises:
and respectively optimizing the low-segment BP neural network and the high-segment BP neural network which determine the training function and the optimal number of hidden layers by utilizing a genetic algorithm under different genetic algebras, and determining an optimal BP neural network model with the optimal number of hidden layers.
6. A segmented neural network pressure sensor pressure detection system, comprising:
a first acquisition module that acquires resistance change data of a strain material based on a pressure applied to the piezoresistive pressure sensor;
the processing module is used for inputting resistance change data of the strain material based on the pressure applied to the piezoresistive pressure sensor into a predetermined segmented neural network model optimized based on a genetic algorithm and outputting accurate pressure data;
the processing module comprises a model determination module comprising:
the model segmentation module is used for sequentially dividing the pressure identification range of the piezoresistive pressure sensor into a low-section pressure identification range and a high-section pressure identification range from the middle point of the pressure identification range according to the pressure identification range of the piezoresistive pressure sensor, setting a corresponding low-section BP neural network and a corresponding high-section BP neural network for the low-section pressure identification range and the high-section pressure identification range respectively, and identifying pressure data of the low-section pressure identification range and pressure data of the high-section pressure identification range of the piezoresistive pressure sensor respectively;
the second acquisition module is used for acquiring a training function;
the third acquisition module is used for acquiring the predetermined optimal number of hidden layers of the low-section BP neural network and the predetermined optimal number of hidden layers of the high-section BP neural network;
the optimization module is used for respectively optimizing the low-segment BP neural network and the high-segment BP neural network which determine the training function and the optimal number of hidden layers by utilizing a genetic algorithm to determine an optimal BP neural network model;
the optimization process of the genetic algorithm comprises the following steps:
determining a parameter coding rule according to the structure of the segmented neural network model; randomly generating long strings of N parameters according to a parameter coding rule to form an initial population of the algorithm; starting from the initial population, continuously using genetic operators to generate new individuals, and improving the overall fitness of the population until the algorithm meets the termination condition; after the termination condition is met, decoding the optimal individual according to a parameter coding rule into a group of distribution of a sectional BP neural network connection weight and a threshold value, and taking the distribution as an initial parameter of the sectional BP neural network; starting from initial parameters, the segmented BP neural network calculates network output errors by using a training data set, reversely transmits the errors and corrects the weight and the threshold if the termination condition is not met, and terminates the network learning until the errors meet the condition, thereby obtaining an optimal BP neural network model.
7. The segmented neural network pressure sensor pressure sensing system of claim 6, wherein the training function is one of a Bayesian rule method, a Levenberg-Marquardt algorithm, a normalized conjugate gradient method, an elastic BP algorithm.
8. The segmented neural network pressure sensor pressure detection system of claim 7, wherein the second acquisition module comprises:
acquiring sample data of the piezoresistive pressure sensor of the type to be tested, and dividing the sample data into a training set, a verification set and a test set;
acquiring different training functions, including a Bayesian rule method, a Levenberg-Marquardt algorithm, a normalized conjugate gradient method and an elastic BP algorithm;
and training the neural network according to the training set and each training function, and comprehensively considering the iteration times and the error to determine the final training function.
9. The segmented neural network pressure sensor pressure detection system of claim 6, wherein the third acquisition module comprises:
the cyclic training module is used for circularly training the BP neural network under different numbers of hidden layers for a plurality of times;
and respectively determining the optimal number of hidden layers of the low-section BP neural network and the optimal number of hidden layers of the high-section BP neural network by using the principle of minimum error according to the result of the cyclic training processing.
10. The segmented neural network pressure sensor pressure detection system of claim 6, wherein the optimization module further comprises a genetic algebra determination module,
the method is used for respectively optimizing the low-segment BP neural network and the high-segment BP neural network which determine the training function and the optimal number of hidden layers by utilizing a genetic algorithm under different genetic algebras, and determining the optimal BP neural network model with the optimal number of hidden layers.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110516026.1A CN113176022B (en) | 2021-05-12 | 2021-05-12 | Segmented neural network pressure sensor pressure detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110516026.1A CN113176022B (en) | 2021-05-12 | 2021-05-12 | Segmented neural network pressure sensor pressure detection method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113176022A CN113176022A (en) | 2021-07-27 |
| CN113176022B true CN113176022B (en) | 2022-10-14 |
Family
ID=76929144
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110516026.1A Active CN113176022B (en) | 2021-05-12 | 2021-05-12 | Segmented neural network pressure sensor pressure detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113176022B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113397500B (en) * | 2021-08-03 | 2022-06-28 | 华东师范大学 | Pulse monitoring device |
| JP2023140059A (en) * | 2022-03-22 | 2023-10-04 | 株式会社ブリヂストン | Estimation device, estimation method, estimation program, and learning model generator |
| CN115265881B (en) * | 2022-09-28 | 2022-12-20 | 宁波普瑞均胜汽车电子有限公司 | Pressure detection method and device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102004031006A1 (en) * | 2003-07-23 | 2005-04-28 | Daimler Chrysler Ag | Combustion monitoring system for common rail direct injection diesel engine has pressure readings from sensor in common rail filtered through neural net |
| CN103971162A (en) * | 2014-04-04 | 2014-08-06 | 华南理工大学 | Method for improving BP (back propagation) neutral network and based on genetic algorithm |
| CN106983504A (en) * | 2016-09-21 | 2017-07-28 | 李军 | A kind of method based on convolutional neural networks algorithm intracranial pressure |
| CN110044522A (en) * | 2019-03-25 | 2019-07-23 | 北京航空航天大学 | A method of it is homogenized using neural fusion piezoelectric pressure detection touch screen piezoelectric response |
| CN112274129A (en) * | 2020-10-11 | 2021-01-29 | 山东大学 | Intracranial pressure measuring device of BP neural network model based on genetic algorithm optimization and working method thereof |
| CN112446091A (en) * | 2020-11-19 | 2021-03-05 | 中国航天空气动力技术研究院 | Artificial neural network-based pulsating pressure prediction method |
-
2021
- 2021-05-12 CN CN202110516026.1A patent/CN113176022B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102004031006A1 (en) * | 2003-07-23 | 2005-04-28 | Daimler Chrysler Ag | Combustion monitoring system for common rail direct injection diesel engine has pressure readings from sensor in common rail filtered through neural net |
| CN103971162A (en) * | 2014-04-04 | 2014-08-06 | 华南理工大学 | Method for improving BP (back propagation) neutral network and based on genetic algorithm |
| CN106983504A (en) * | 2016-09-21 | 2017-07-28 | 李军 | A kind of method based on convolutional neural networks algorithm intracranial pressure |
| CN110044522A (en) * | 2019-03-25 | 2019-07-23 | 北京航空航天大学 | A method of it is homogenized using neural fusion piezoelectric pressure detection touch screen piezoelectric response |
| CN112274129A (en) * | 2020-10-11 | 2021-01-29 | 山东大学 | Intracranial pressure measuring device of BP neural network model based on genetic algorithm optimization and working method thereof |
| CN112446091A (en) * | 2020-11-19 | 2021-03-05 | 中国航天空气动力技术研究院 | Artificial neural network-based pulsating pressure prediction method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113176022A (en) | 2021-07-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113176022B (en) | Segmented neural network pressure sensor pressure detection method and system | |
| US20230252327A1 (en) | Neural architecture search for convolutional neural networks | |
| CN111192631B (en) | Methods and systems for constructing models for predicting protein-RNA interaction binding sites | |
| KR102492318B1 (en) | Model training method and apparatus, and data recognizing method | |
| US20190228312A1 (en) | Unsupervised model building for clustering and anomaly detection | |
| CN108596327B (en) | Seismic velocity spectrum artificial intelligence picking method based on deep learning | |
| CN108664690A (en) | Long-life electron device reliability lifetime estimation method under more stress based on depth belief network | |
| US20220121934A1 (en) | Identifying neural networks that generate disentangled representations | |
| KR20210107491A (en) | Method for generating anoamlous data | |
| CN112215259B (en) | Gene selection method and apparatus | |
| Tian et al. | Joint learning model for underwater acoustic target recognition | |
| CN108846200B (en) | Quasi-static bridge influence line identification method based on iteration method | |
| CN113742983A (en) | Long-span structural damage identification method based on depth self-encoder neural network | |
| CN111639194B (en) | Knowledge graph query method and system based on sentence vector | |
| CN113257425A (en) | Influenza prediction system, storage medium and device for optimizing LSTM and LightGBM parameters | |
| CN110489800B (en) | Structural dynamic load sparse identification method based on matrix regularization | |
| CN108280485A (en) | A kind of non-rigid method for searching three-dimension model based on spectrogram Wavelet Descriptor | |
| CN109187898B (en) | Soft measurement method and device for ammonia nitrogen content of water in aquaculture environment | |
| Bondarenko et al. | Inverse problems of anomalous diffusion theory: An artificial neural network approach | |
| US20240120022A1 (en) | Predicting protein amino acid sequences using generative models conditioned on protein structure embeddings | |
| CN115423045A (en) | System log detection method and system based on GAN network and meta learning | |
| Long et al. | Comparative study between the magic formula and the neural network tire model based on genetic algorithm | |
| CN115062769A (en) | Knowledge distillation-based model training method, device, equipment and storage medium | |
| US20230214668A1 (en) | Hyperparameter adjustment device, non-transitory recording medium in which hyperparameter adjustment program is recorded, and hyperparameter adjustment program | |
| CN114821322A (en) | Small sample remote sensing image classification method and system based on attention mechanism |
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 |