CN111816404B - Demagnetization method and system - Google Patents
Demagnetization method and system Download PDFInfo
- Publication number
- CN111816404B CN111816404B CN202010679178.9A CN202010679178A CN111816404B CN 111816404 B CN111816404 B CN 111816404B CN 202010679178 A CN202010679178 A CN 202010679178A CN 111816404 B CN111816404 B CN 111816404B
- Authority
- CN
- China
- Prior art keywords
- induction signal
- medium
- signal
- magnetic field
- induction
- 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
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01F—MAGNETS; INDUCTANCES; TRANSFORMERS; SELECTION OF MATERIALS FOR THEIR MAGNETIC PROPERTIES
- H01F13/00—Apparatus or processes for magnetising or demagnetising
- H01F13/006—Methods and devices for demagnetising of magnetic bodies, e.g. workpieces, sheet material
Abstract
The invention relates to a demagnetization method and a demagnetization system, wherein the method comprises the following steps: transmitting a detection signal of the degaussing coil array to the medium to generate an original induction signal of the medium; preprocessing an original induction signal to obtain an induction signal; taking the induction signal as input, taking the medium type corresponding to the induction signal as output, and training a medium classification model by adopting a machine learning algorithm; taking the induction signal and the medium type corresponding to the induction signal as input, taking the magnetic field intensity required by demagnetization corresponding to the induction signal as output, and training a magnetic field intensity prediction model by adopting a machine learning algorithm; inputting the induction signal to be demagnetized into a medium classification model to obtain a medium classification result of the induction signal to be demagnetized; inputting the medium classification result and the induction signal to be demagnetized into a magnetic field intensity prediction model to obtain the magnetic field intensity required by medium demagnetization; and demagnetizing the medium according to the magnetic field intensity. The invention demagnetizes the magnetic storage medium according to the acquired magnetic field intensity, thereby reducing the energy consumption required by demagnetization.
Description
Technical Field
The invention relates to the technical field of electromagnetism, in particular to a demagnetization method and a demagnetization system.
Background
With the progress of national information construction, the recording, storing and transmitting technology of information is developed unprecedentedly. Magnetic storage media are widely used in enterprises, individuals and government agencies as storage media with large capacity and high transmission speed, and the main magnetic carriers of the magnetic storage media comprise magnetic storage media such as hard disks, magnetic tapes and floppy disks. The magnetic storage medium usually has a certain service life, and the continuous improvement of the use requirement further accelerates the updating and updating of the magnetic storage medium, so that a large amount of scrapped magnetic storage media are generated, and meanwhile, the problems of information leakage, illegal information stealing and incomplete safety elimination of a large amount of sensitive information of the scrapped magnetic storage media are brought.
The electromagnetic degaussing system is a degaussing technology for destroying information in a storage medium by using a strong magnetic field, and the magnetic storage media are various in types and different in forms, so that the intensity of a degaussing field required by different types of magnetic storage media for achieving thorough degaussing and destroying is greatly different, and the difficulty is brought to the use and the upgrade of the electromagnetic degaussing system. The existing electromagnetic demagnetizing system does not usually have the capacity of identifying the type of the storage medium, the maximum power of the demagnetizer is uniformly used for information destruction on the magnetic storage medium, the phenomena of electric energy waste and service life shortening of the demagnetizer exist, and the method of manually judging the type of the storage medium has extra management overhead and reduces the operation efficiency of the demagnetizer.
Disclosure of Invention
Based on this, the invention aims to provide a demagnetization method and a demagnetization system, which are used for acquiring the magnetic field intensity required by demagnetization of a magnetic storage medium to be demagnetized, demagnetizing the magnetic storage medium according to the acquired magnetic field intensity and reducing the energy consumption required by demagnetization.
In order to achieve the purpose, the invention provides the following scheme:
a method of degaussing, the method comprising:
transmitting a detection signal of a degaussing coil array to a medium to generate an original induction signal of the medium;
receiving the original sensing signals of a plurality of types of media;
preprocessing the original induction signal to obtain an induction signal;
taking the induction signal as sample data, taking the induction signal as input, taking the medium type corresponding to the induction signal as output, and training a medium classification model by adopting a machine learning algorithm;
taking the induction signal and the type of the medium corresponding to the induction signal as input, taking the magnetic field strength required by demagnetization corresponding to the induction signal as output, and training a magnetic field strength prediction model by adopting a machine learning algorithm;
inputting the induction signal to be demagnetized into the medium classification model to obtain a medium classification result of the induction signal to be demagnetized; the induction signal to be demagnetized is an induction signal obtained by transmitting a detection signal to a medium to be demagnetized;
inputting the medium classification result and the induction signal to be demagnetized into the magnetic field strength prediction model to obtain the magnetic field strength required by medium demagnetization;
and demagnetizing the medium according to the magnetic field intensity.
Optionally, the preprocessing the original sensing signal to obtain the sensing signal specifically includes:
rejecting abnormal data in the original induction signal by adopting a 3 sigma criterion to obtain a first induction signal;
adding the maximum value, the minimum value, the mean value and the variance of the first induction signal into the first induction signal as structural characteristics to obtain a second induction signal;
performing dimensionality reduction processing on the second induction signal by using a principal component analysis method to obtain a third induction signal;
and normalizing the third induction signal by using a linear function normalization method or a zero mean normalization method to obtain the induction signal.
Optionally, the original sensing signal is an induced current signal obtained by sampling the received current signal for 2 seconds for 100 times at equal intervals.
Optionally, the machine learning algorithm comprises any one of XGBoost, decision tree, support vector machine and neural network.
Optionally, before the medium classification model is trained by using a machine learning algorithm, the sensing signals are randomly divided into a training set, a verification set and a test set according to a ratio of 6:2: 2.
The invention also provides a demagnetization system, comprising:
the information acquisition module: the detection signal of the degaussing coil array is transmitted to a medium, and an original induction signal of the medium is generated; receiving the original sensing signals of a plurality of types of media; preprocessing the original induction signal to obtain an induction signal;
a medium classification model modeling module: the medium classification model is trained by a machine learning algorithm by taking the induction signal as sample data, taking the induction signal as input and taking the medium type corresponding to the induction signal as output;
a magnetic field strength prediction model modeling module: the magnetic field strength prediction model is trained by adopting a machine learning algorithm by taking the induction signal as sample data, taking the induction signal and the type of the medium corresponding to the induction signal as input and taking the magnetic field strength required by demagnetization corresponding to the induction signal as output;
the magnetic field intensity prediction module is used for inputting the induction signal to be demagnetized into the medium classification model to obtain a medium classification result of the induction signal to be demagnetized; the induction signal to be demagnetized is an induction signal obtained by transmitting a detection signal to a medium to be demagnetized; inputting the medium classification result and the induction signal to be demagnetized into the magnetic field strength prediction model to obtain the magnetic field strength required by medium demagnetization;
and the demagnetizing module is used for demagnetizing the medium according to the magnetic field intensity.
Optionally, the information acquisition module includes a data processing unit, and specifically includes:
the abnormal data processing subunit is used for eliminating abnormal data in the original induction signal by adopting a 3 sigma criterion to obtain a first induction signal;
the characteristic construction subunit is used for adding the maximum value, the minimum value, the mean value and the variance of the first induction signal into the first induction signal as construction characteristics to obtain a second induction signal;
the dimensionality reduction processing subunit is used for performing dimensionality reduction processing on the second induction signal by using a principal component analysis method to obtain a third induction signal;
and the normalization subunit is used for normalizing the third sensing signal by using a linear function normalization method or a zero-mean normalization method to obtain the sensing signal.
Optionally, the original sensing signal is an induced current signal obtained by sampling the received current signal for 2 seconds for 100 times at equal intervals.
Optionally, the machine learning algorithm comprises any one of XGBoost, decision tree, support vector machine and neural network.
Optionally, the medium classification model modeling module includes:
and the data set processing unit is used for randomly dividing the induction signals into a training set, a verification set and a test set according to the ratio of 6:2: 2.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention trains a medium classification model and a magnetic field intensity module by taking induction signals of different types of media as a data set, obtains the magnetic field intensity required by demagnetization of unknown media through the medium classification model and the magnetic field intensity module, and adjusts the magnetic field intensity required by demagnetization aiming at different types of media, thereby reducing the energy consumption of demagnetization.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a schematic flow chart of a demagnetization method according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a demagnetization system according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a demagnetization method and a demagnetization system, which are used for acquiring the magnetic field intensity required by demagnetization of a magnetic storage medium to be demagnetized, demagnetizing the magnetic storage medium according to the acquired magnetic field intensity and reducing the energy consumption required by demagnetization.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Fig. 1 is a schematic flow chart of a degaussing method, and as shown in fig. 1, the degaussing method includes:
step 101: the detection signal of the degaussing coil array is transmitted to the medium, and the original induction signal of the medium is generated.
Wherein, step 101 specifically includes: transmitting a pulse magnetic field detection signal of a degaussing coil array to various media, receiving an induced magnetic field signal of the media and generating an induced current signal, and sampling the received current signal within 2 seconds for 100 times at equal intervals to obtain an induced current signal I ═ xjAnd j equals to 1,2,3 …,100}, and the induced current signal is the original induced signal.
Step 102: receiving the raw sensing signals for multiple types of media.
Step 103: and preprocessing the original induction signal to obtain an induction signal.
Wherein, step 103 specifically comprises: preprocessing the original induction signal to obtain an induction signal I' ═ { x ═ xi1,2,3 …, n }, where n is the number of sensing signals, and specifically includes:
calculating the average value mu and the standard deviation sigma of the collected induced current signals I, judging the induced current signals with the difference between the signal value and the average value mu of more than 3 mu in each collected induced current signal as abnormal data according to a 3 sigma criterion, replacing the abnormal data with the average value mu, and obtaining the induced current signals I after data cleaning1And is denoted as a first sensing signal.
Adding the maximum value, the minimum value, the mean value and the variance of the first induction signal into the first induction signal as structural characteristics to obtain a second induction signal I2＝{xi1,2,3 …,104}, where x is101、x102、x103And x104Respectively, the maximum value, the minimum value, the mean value and the variance of the first induction signal.
Performing dimensionality reduction on the second induction signal by using a Principal Component Analysis (PCA), solving a data dimensionality N after dimensionality reduction by using a grid search (GridSearchCV) with a goal of maximizing the Euclidean distance between classes, setting an output data dimensionality of the PCA algorithm to be N, and setting the second induction current signal I to be N2After the PCA algorithm is input, the PCA outputs a reduced-dimension N-dimension third induced current signal I3。
The third sensing signal is normalized using either linear function Normalization (Min-Max Scaling) or zero mean Normalization (Z-Score Normalization) to obtain a sensing signal I'.
The expression of the linear function normalization method for processing the induced current signals is as follows:
xnormis xiNormalized induced current signal, xiFor normalized induced current signal, xmaxIs the third induced current signal I3Maximum value of (1), xminIs the third induced current signal I3Minimum value of (1).
The expression of the zero-mean normalization method for processing the induced current signals is as follows:
wherein z is xiNormalized signal, μ3Is the third induced current signal I3Mean value of (a)3Is the third induced current signal I3Standard deviation of (2).
Step 104: and taking the induction signal as sample data, taking the induction signal as input, taking the medium type corresponding to the induction signal as output, and training a medium classification model by adopting a machine learning algorithm.
Wherein, step 104 specifically includes: the medium type label of the medium sample data is labeled manually, for example, the number 1 represents a first type of medium, the number 2 represents a second type of medium, … …, and so on. By the leave-out method, the data set is randomly divided into training sets X according to the ratio of 6:2:2trainVerification set XvalAnd test set XtestAnd the verification set is used for optimizing the parameters of the machine learning model and preventing the machine learning model from being over-fitted.
Constructing a neural network model: the number of neurons of the input layer of the neural network is equal to the dimensionality of the sensing signal, no activation function is used, the number of neurons of the output layer of the neural network is equal to the number of classes of the medium, a normalized exponential function (softmax function) is used as the activation function, the process of determining the number of hidden layers of the neural network, the number of neurons contained in each layer and the activation function used is used as a parameter optimization process, and a grid search (GridSearchCV) method is used to maximize the verification set XvalThe accuracy of the method is the optimal parameter of the target for acquiring the hidden layer. And finishing the construction of the medium classification model.
Will train set XtrainInputting a medium classification model and optimizing weight parameters of a neural network by using a Stochastic Gradient Descent (SGD) algorithm to minimize cross entropy (cross entropy), and iterating 1000 times to obtain the trained medium classification model and storing the weight parameters. To ensure that the medium classification model has good classification capability on unknown media, the generalization error of the medium classification model needs to be evaluated, namely the trained medium classification model is used for testing the set XtestAnd (3) predicting the sample, calculating the classification accuracy, considering that the medium classification model has good generalization performance when the accuracy is more than 99%, and reconstructing and training the medium classification model otherwise. When the medium classification model is tested, the evaluation indexes of the generalized errors comprise an error rate, an accuracy rate and a recall rate besides the accuracy rate.
Step 105: and training a magnetic field intensity prediction model by adopting a machine learning algorithm by taking the induction signal and the type of the medium corresponding to the induction signal as input and the magnetic field intensity required by demagnetization corresponding to the induction signal as output.
Wherein, step 105 specifically includes: and manually marking the medium type label of the medium sample data and the magnetic field intensity required by demagnetization of the medium sample. By the leave-out method, the data set is randomly divided into training sets X according to the ratio of 6:2:2trainVerification set XvalAnd test set Xtest。
Constructing a neural network model: the number of neurons of the input layer of the neural network is equal to the dimension of the induced current signal, no activation function is used, the output layer of the neural network comprises 1 neuron, no activation function is used, the process of determining the number of hidden layers of the neural network, the number of neurons contained in each layer and the used activation function is taken as a parameter optimization process, and a grid search (GridSearchCV) method is used for minimizing a verification set XvaAnd l, taking Mean Square Error (MSE) as an optimal parameter of the target acquisition hidden layer to complete the construction of the magnetic field strength prediction model.
Will train set XtrainInputting a magnetic field strength prediction model, solving an optimal weight parameter of a neural network by using a Stochastic Gradient Descent (SGD) algorithm with a minimum training set mean square error as a target, iterating for 1000 times to obtain the trained magnetic field strength prediction model, and storing the weight parameter. In order to ensure that the magnetic field strength prediction model has good prediction capability on the demagnetizing magnetic field strength required by an unknown medium, the generalization error of the magnetic field strength prediction model needs to be evaluated, namely the trained magnetic field strength prediction model is used for testing the set XtestAnd predicting the demagnetized magnetic field intensity required by the sample, calculating the mean square error, and if the mean square error is less than 1, determining that the magnetic field intensity prediction model has good generalization performance, otherwise, reconstructing and training the magnetic field intensity prediction model. When the magnetic field intensity prediction model is tested, the evaluation indexes of the generalized error comprise a root mean square error and an absolute error besides a mean square error.
Step 106: the method comprises the steps of transmitting a detection signal of a degaussing coil array to a medium to be degaussed to generate an original induction signal to the medium to be degaussed, preprocessing the original induction signal of the medium to be degaussed to obtain the induction signal to be degaussed, inputting the induction signal to be degaussed into a medium classification model, and obtaining a medium classification result of the induction signal to be degaussed.
Wherein, step 106 specifically includes: and preprocessing the induction signal to be demagnetized, inputting the preprocessed induction signal into a medium classification model, and identifying the class label C of the medium through forward propagation.
Step 107: and inputting the medium classification result and the induction signal to be demagnetized into the magnetic field intensity prediction model to obtain the magnetic field intensity required by medium demagnetization.
Wherein, step 107 specifically comprises: and inputting the preprocessed induction signal and the medium type label C into a magnetic field strength prediction model, and obtaining the magnetic field strength H required by medium demagnetization through forward propagation.
Step 108: and demagnetizing the medium according to the magnetic field intensity.
Wherein, step 108 specifically comprises:
setting the current I to be injected into the degaussing coil according to the magnetic field intensity H required by medium degaussingt. The current I of the degaussing coil can be obtained by inputting the magnetic field strength H through establishing a function mapping relation between the magnetic field strength and the current of the degaussing coil in advancet. The function mapping relation determination mode is as follows: and adjusting the current of the demagnetizing coil, testing and recording the corresponding magnetic field intensity, obtaining a plurality of groups of test data, and then performing polynomial fitting to obtain a function mapping relation between the current of the demagnetizing coil and the magnetic field intensity.
Fig. 2 is a schematic structural diagram of a demagnetization system, as shown in fig. 2, the demagnetization system includes:
the information acquisition module 201: the detection signal of the degaussing coil array is transmitted to a medium, and an original induction signal of the medium is generated; receiving the original sensing signals of a plurality of types of media; and preprocessing the original induction signal to obtain an induction signal.
Wherein, the information collecting module 201 specifically includes:
and the transmitting unit comprises a transmitting circuit for transmitting the pulse magnetic field detection signal of the degaussing coil array.
And the receiving unit comprises a receiving circuit and is used for receiving the electromagnetic induction signal intensity of the medium and generating induction current information.
Electromagnetic induction signal Bs(ω) can be expressed as:
where m (ω) is the induced dipole moment at the location of the medium at an angular frequency ω, which is related to the magnetic field to be detected and the material, shape, and size of the medium, r is the distance from the identification coil to the center of the medium,
is a unit vector with the center of the medium pointing to the center of the identification coil,
is an identity matrix. And transmits the collected signals to the data processing unit.
The data processing unit includes:
a data sampling subunit, configured to sample the received current signal within 2 seconds for 100 times at equal intervals to obtain an induced current signal I ═ xj|j＝1,2,3…,100}。
And the abnormal data processing subunit is used for setting a normal value range of a statistical square tolerance method (Root-Sum-Squares RSS) and eliminating abnormal data in the original induction signal by adopting a 3 sigma criterion. The process of removing the abnormal data specifically comprises the following steps: calculating the average value mu and the standard deviation sigma of the acquired induced current signals I, judging the induced current signals with the difference between the signal value and the average value mu of each acquired induced current signal larger than 3 mu as abnormal data according to a 3 sigma criterion, replacing the abnormal data with the average value mu or a fixed value, and acquiring the induced current signals I after data cleaning1And is denoted as a first sensing signal.
A feature construction subunit, configured to perform feature construction on the signal using an automatic encoder (AutoEncoder); or adding the maximum, minimum, mean and variance of the first sensing signal as a constructive feature to the first sensingOf the signals, a second induced signal I is obtained2＝{xi1,2,3 …,104}, where x is101、x102、x103And x104Respectively, the maximum value, the minimum value, the mean value and the variance of the first induction signal. The feature construction highlights data features of the sensing signal, and enhances information expression capability of the sensing signal data set.
And the dimension reduction processing subunit is used for performing dimension reduction processing on the second induction signal by using a principal component analysis method. Performing dimensionality reduction on the second induction signal by using a Principal Component Analysis (PCA), solving a data dimensionality N after dimensionality reduction by using a grid search (GridSearchCV) with a goal of maximizing inter-class Euclidean distance, setting an output data dimensionality of the PCA algorithm to be N, and setting the second induction current signal I to be N2After the PCA algorithm is input, the PCA outputs a reduced-dimension N-dimension third induced current signal I3. And the information redundancy of the induction information is reduced through dimension reduction processing, and the model training and predicting process is accelerated. The dimension reduction processing can also select a feature extraction and feature selection method. The feature extraction method maps high-dimensional attributes in a feature set to a low-dimensional subspace through mathematical transformation, so that the dimensionality of a data set is reduced, and the feature extraction method comprises but is not limited to a principal component analysis method, a kernel principal component analysis method and local linear embedding; the feature selection method selects K-dimensional feature subsets from M-dimensional feature sets, wherein K is<M, including but not limited to genetic algorithm-based feature selection and correlation vector machine-based feature selection. In addition, data dimensionality is reduced, noise interference can be reduced, and separability of sample data is enhanced.
And the normalizing subunit is used for normalizing the third induction signal by using linear function normalization (Min-Max Scaling) or zero-mean normalization (Z-Scorenomalization) to obtain an induction signal I'.
The expression of the linear function normalization method for processing the induced current signals is as follows:
xnormis xiNormalized induced current signal, xiFor normalized induced current signal, xmaxIs the third induced current signal I3Maximum value of (1), xminIs the third induced current signal I3Minimum value of (1).
The expression of the zero-mean normalization method for processing the induced current signals is as follows:
wherein z is xiNormalized signal, μ3Is the third induced current signal I3Mean value of (a)3Is the third induced current signal I3Standard deviation of (2).
The data processing unit comprises at least one microprocessor which sends the received signals to a medium classification model modelling module 202 and a magnetic field strength prediction model modelling module 203.
Media classification model modeling module 202: the method is used for training a medium classification model by using a machine learning algorithm with historical induction signals as sample data, the induction signals as input and the medium types corresponding to the induction signals as output. And evaluating the generalization error of the medium classification model, and selecting the medium classification model with good generalization performance. The evaluation indexes of the medium classification model comprise accuracy, error rate, precision rate, recall rate and the like.
The medium classification model modeling module 202 specifically includes: the medium type label of the medium sample data is labeled manually, for example, the number 1 represents a first type of medium, the number 2 represents a second type of medium, … …, and so on. Taking the induction signals and the medium types corresponding to the induction signals as data sets, and randomly dividing the data sets into training sets X according to the ratio of 6:2:2 by any one of a leave-out method, a cross-validation method and a self-help methodtrainVerification set XvalAnd test set Xtest。
Constructing a neural network model: the number of neurons of the input layer of the neural network is equal to the dimensionality of the sensing signal, without using an activation function, the number of neurons of the output layer of the neural network is equal to the number of classes of the medium, with softmax as the activation function, the determination being madeThe process of the number of hidden layers of the neural network, the number of neurons contained in each layer and the activation function used is a parameter optimization process, and a grid search (GridSearchCV) method is used to maximize the verification set XvalThe accuracy is the best parameter for acquiring the hidden layer of the target. And finishing the construction of the medium classification model.
Will train set XtrainInputting a medium classification model and optimizing weight parameters of a neural network by using a Stochastic Gradient Descent (SGD) algorithm to minimize cross entropy (cross entropy), and iterating 1000 times to obtain the trained medium classification model and storing the weight parameters. To ensure that the medium classification model has good classification capability on unknown media, the generalization error of the medium classification model needs to be evaluated, namely the trained medium classification model is used for testing the set XtestAnd (3) predicting the sample, calculating the classification accuracy, considering that the medium classification model has good generalization performance when the accuracy is more than 99%, and reconstructing and training the medium classification model otherwise.
Magnetic field strength prediction model modeling module 203: and the magnetic field strength prediction model is trained by adopting a machine learning regression algorithm by taking the induction signal as sample data, taking the induction signal and the type of the medium corresponding to the induction signal as input and taking the magnetic field strength required by demagnetization corresponding to the induction signal as output.
The magnetic field strength prediction model modeling module 203 specifically includes: and manually marking the medium type label of the medium sample data and the magnetic field intensity required by demagnetization of the medium sample. Taking the induction signal, the type of the medium corresponding to the induction signal and the magnetic field intensity required by demagnetization corresponding to the induction signal as a data set, and randomly dividing the data set into a training set X according to the ratio of 6:2:2 by any one of a reservation method, a cross-validation method and a self-help methodtrainVerification set XvalAnd test set Xtest。
Constructing a neural network model: the number of neurons in the input layer of the neural network is equal to the dimension of the induced current signal without using the activation function, and the output layer of the neural network comprises 1 neuron without using the activation functionThe process of the number of hidden layers of the neural network, the number of neurons contained in each layer and the activation function used is a parameter optimization process, and a grid search (GridSearchCV) method is used to minimize the verification set XvalAnd the Mean Square Error (MSE) is the target to obtain the optimal parameter of the hidden layer, so that the magnetic field intensity prediction model construction is completed.
Will train set XtrainInputting a magnetic field strength prediction model, solving an optimal weight parameter of a neural network by using a Stochastic Gradient Descent (SGD) algorithm with a minimum training set mean square error as a target, iterating for 1000 times to obtain the trained magnetic field strength prediction model, and storing the weight parameter. In order to ensure that the magnetic field strength prediction model has good prediction capability on the demagnetizing magnetic field strength required by an unknown medium, the generalization error of the magnetic field strength prediction model needs to be evaluated, namely the trained magnetic field strength prediction model is used for testing the set XtestAnd predicting the demagnetized magnetic field intensity required by the sample, calculating the mean square error, and if the mean square error is less than 1, determining that the magnetic field intensity prediction model has good generalization performance, otherwise, reconstructing and training the magnetic field intensity prediction model.
The magnetic field strength prediction module 204 is configured to input the induction signal to be demagnetized into the medium classification model, and obtain a medium classification result of the induction signal to be demagnetized; the induction signal to be demagnetized is an induction signal obtained by transmitting a detection signal to a medium to be demagnetized; and inputting the medium classification result and the induction signal to be demagnetized into the magnetic field intensity prediction model to obtain the magnetic field intensity required by medium demagnetization. And evaluating the generalization error of the magnetic field strength prediction model, and selecting the magnetic field strength prediction model with good generalization performance. The evaluation indexes of the magnetic field strength prediction model comprise mean square error, root mean square error, absolute error and the like.
And the medium classification model and the magnetic field intensity prediction model are evaluated, so that overfitting of the models is prevented, and the models are ensured to keep good prediction performance on unknown media.
The magnetic field strength predicting module 204 specifically includes: and the induction signal to be demagnetized is an induction signal of the medium to be demagnetized, the induction signal to be demagnetized is preprocessed and then is input into the medium classification model in real time, and the class label C of the medium is identified through forward propagation. And inputting the preprocessed induction signal and the medium type label C into a magnetic field intensity prediction model in real time, and obtaining the magnetic field intensity H required by medium demagnetization in real time through forward propagation.
And the demagnetizing module 205 is used for demagnetizing the medium according to the magnetic field strength.
Wherein, the demagnetization module 205 specifically includes:
according to the magnetic field intensity H required by medium degaussing, the degaussing control unit sets the current I required to be injected into the degaussing coilt. The current I of the degaussing coil can be obtained in real time by inputting the magnetic field intensity H through establishing a function mapping relation between the magnetic field intensity and the current of the degaussing coil in advancet. The function mapping relation determination mode is as follows: and adjusting the current of the demagnetizing coil, testing and recording the corresponding magnetic field intensity, obtaining a plurality of groups of test data, and then performing polynomial fitting to obtain a function mapping relation between the current of the demagnetizing coil and the magnetic field intensity. In the process of demagnetizing various types of media, the demagnetizing system adaptively adjusts the current of the demagnetizing coil, drives the transmitting circuit to generate a demagnetizing magnetic field, completes the demagnetization of the media and reduces the energy consumption of the demagnetizing system.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (8)
1. A method of degaussing, the method comprising:
transmitting a detection signal of a degaussing coil array to a medium to generate an original induction signal of the medium;
receiving the original sensing signals of a plurality of types of media;
preprocessing the original induction signal to obtain an induction signal;
taking the induction signal as sample data, taking the induction signal as input, taking the medium type corresponding to the induction signal as output, and training a medium classification model by adopting a machine learning algorithm;
taking the induction signal and the type of the medium corresponding to the induction signal as input, taking the magnetic field strength required by demagnetization corresponding to the induction signal as output, and training a magnetic field strength prediction model by adopting a machine learning algorithm;
inputting the induction signal to be demagnetized into the medium classification model to obtain a medium classification result of the induction signal to be demagnetized, wherein the induction signal to be demagnetized is an induction signal obtained by transmitting a detection signal to a medium to be demagnetized;
inputting the medium classification result and the induction signal to be demagnetized into the magnetic field strength prediction model to obtain the magnetic field strength required by medium demagnetization;
demagnetizing the medium according to the magnetic field intensity;
preprocessing the original sensing signal to obtain a sensing signal, which specifically comprises:
rejecting abnormal data in the original induction signal by adopting a 3 sigma criterion to obtain a first induction signal;
adding the maximum value, the minimum value, the mean value and the variance of the first induction signal into the first induction signal as structural characteristics to obtain a second induction signal;
performing dimensionality reduction processing on the second induction signal by using a principal component analysis method to obtain a third induction signal;
and normalizing the third induction signal by using a linear function normalization method or a zero mean normalization method to obtain the induction signal.
2. The method of claim 1, wherein the original induced signal is an induced current signal obtained by sampling the received current signal 100 times at equal intervals over 2 seconds.
3. The degaussing method of claim 1, wherein the machine learning algorithm comprises any one of an XGBoost, a decision tree, a support vector machine, and a neural network.
4. The method of claim 1, wherein the induction signal is randomly divided into a training set, a validation set, and a test set at a ratio of 6:2:2 before the machine learning algorithm is used to train the medium classification model.
5. A demagnetization system, the demagnetization system comprising:
the information acquisition module: the detection signal of the degaussing coil array is transmitted to a medium, and an original induction signal of the medium is generated; receiving the original sensing signals of a plurality of types of media; preprocessing the original induction signal to obtain an induction signal;
a medium classification model modeling module: the medium classification model is trained by a machine learning algorithm by taking the induction signal as sample data, taking the induction signal as input and taking the medium type corresponding to the induction signal as output;
a magnetic field strength prediction model modeling module: the magnetic field strength prediction model is trained by adopting a machine learning algorithm by taking the induction signal as sample data, taking the induction signal and the type of the medium corresponding to the induction signal as input and taking the magnetic field strength required by demagnetization corresponding to the induction signal as output;
the magnetic field intensity prediction module is used for inputting the induction signal to be demagnetized into the medium classification model to obtain a medium classification result of the induction signal to be demagnetized; the induction signal to be demagnetized is an induction signal obtained by transmitting a detection signal to a medium to be demagnetized; inputting the medium classification result and the induction signal to be demagnetized into the magnetic field strength prediction model to obtain the magnetic field strength required by medium demagnetization;
the demagnetizing module is used for demagnetizing the medium according to the magnetic field intensity;
the information acquisition module comprises a data processing unit, and specifically comprises:
the abnormal data processing subunit is used for eliminating abnormal data in the original induction signal by adopting a 3 sigma criterion to obtain a first induction signal;
the characteristic construction subunit is used for adding the maximum value, the minimum value, the mean value and the variance of the first induction signal into the first induction signal as construction characteristics to obtain a second induction signal;
the dimensionality reduction processing subunit is used for performing dimensionality reduction processing on the second induction signal by using a principal component analysis method to obtain a third induction signal;
and the normalization subunit is used for normalizing the third sensing signal by using a linear function normalization method or a zero-mean normalization method to obtain the sensing signal.
6. The demagnetizing system of claim 5, wherein the original induced signal is an induced current signal obtained by sampling the received current signal 100 times at equal intervals for 2 seconds.
7. The demagnetization system of claim 5, wherein the machine learning algorithm comprises any one of an XGboost, a decision tree, a support vector machine, and a neural network.
8. The demagnetizing system of claim 5, wherein the medium classification model modeling module comprises:
and the data set processing unit is used for randomly dividing the induction signals into a training set, a verification set and a test set according to the ratio of 6:2: 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010679178.9A CN111816404B (en) | 2020-07-15 | 2020-07-15 | Demagnetization method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010679178.9A CN111816404B (en) | 2020-07-15 | 2020-07-15 | Demagnetization method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111816404A CN111816404A (en) | 2020-10-23 |
| CN111816404B true CN111816404B (en) | 2021-02-12 |
Family
ID=72866322
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010679178.9A Active CN111816404B (en) | 2020-07-15 | 2020-07-15 | Demagnetization method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111816404B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112837705B (en) * | 2021-02-25 | 2022-05-17 | 精密电产(北京)科技有限公司 | Demagnetizing method with magnetic field capable of being inverted |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103294960A (en) * | 2013-06-06 | 2013-09-11 | 国家电网公司 | Portable data wiping device based on U disk |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPWO2006070434A1 (en) * | 2004-12-27 | 2008-06-12 | オリエント測器コンピュータ株式会社 | Data recording medium disposal method |
| JP2006313587A (en) * | 2005-05-06 | 2006-11-16 | Hitachi Global Storage Technologies Netherlands Bv | Demagnetization method of magnetic recording medium and demagnetizer |
| CN103295587A (en) * | 2013-06-06 | 2013-09-11 | 国家电网公司 | Centralized synchronous electronic storage medium data scrubbing method |
| CN103400586B (en) * | 2013-07-05 | 2016-04-27 | 中国电子科技集团公司第三十研究所 | A kind of magnetic storage medium data destroying effect detection method |
| CN110517844A (en) * | 2019-09-03 | 2019-11-29 | 中国建设银行股份有限公司 | A kind of self-service article demagnetization method, device, equipment and storage medium |
| CN110797052B (en) * | 2019-10-21 | 2021-10-01 | 北京工业大学 | Rapid demagnetization method based on magnetic medium characteristics |
-
2020
- 2020-07-15 CN CN202010679178.9A patent/CN111816404B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103294960A (en) * | 2013-06-06 | 2013-09-11 | 国家电网公司 | Portable data wiping device based on U disk |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111816404A (en) | 2020-10-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110132598B (en) | Fault noise diagnosis algorithm for rolling bearing of rotating equipment | |
| Zhang et al. | Machine learning based bearing fault diagnosis using the case western reserve university data: A review | |
| CN111914873A (en) | Two-stage cloud server unsupervised anomaly prediction method | |
| CN109302410B (en) | Method and system for detecting abnormal behavior of internal user and computer storage medium | |
| CN111753881B (en) | Concept sensitivity-based quantitative recognition defending method against attacks | |
| CN109902564B (en) | Abnormal event detection method based on structural similarity sparse self-coding network | |
| CN113673346B (en) | Motor vibration data processing and state identification method based on multiscale SE-Resnet | |
| CN112039903B (en) | Network security situation assessment method based on deep self-coding neural network model | |
| CN111353373A (en) | Correlation alignment domain adaptive fault diagnosis method | |
| CN109903053B (en) | Anti-fraud method for behavior recognition based on sensor data | |
| KR102272573B1 (en) | Method for nonintrusive load monitoring of energy usage data | |
| CN115277189B (en) | Unsupervised intrusion flow detection and identification method based on generation type countermeasure network | |
| CN111816404B (en) | Demagnetization method and system | |
| Kim et al. | AnoGAN-based anomaly filtering for intelligent edge device in smart factory | |
| CN110020868B (en) | Anti-fraud module decision fusion method based on online transaction characteristics | |
| CN113822336A (en) | Cloud hard disk fault prediction method, device and system and readable storage medium | |
| Kopčan et al. | Anomaly detection using Autoencoders and Deep Convolution Generative Adversarial Networks | |
| CN117154263A (en) | Lithium battery cascade utilization charging and discharging system and control method | |
| CN116861331A (en) | Expert model decision-fused data identification method and system | |
| Aquize et al. | Self-organizing maps for anomaly detection in fuel consumption. Case study: Illegal fuel storage in Bolivia | |
| CN116662899A (en) | Noise-containing data anomaly detection method based on self-adaptive strategy | |
| CN111667000B (en) | Earthquake early warning method of adaptive field deep neural network | |
| CN115130599A (en) | Semi-supervision method for strip mine card state recognition under time series GAN data enhancement | |
| CN113935413A (en) | Distribution network wave recording file waveform identification method based on convolutional neural network | |
| Amodio et al. | Automatic vehicle model recognition and lateral position estimation based on magnetic sensors |
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 | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A degaussing method and system Effective date of registration: 20220304 Granted publication date: 20210212 Pledgee: Haidian Beijing science and technology enterprise financing Company limited by guarantee Pledgor: Beijing China super industry information security technology Limited by Share Ltd. Registration number: Y2022110000042 |
|
| PC01 | Cancellation of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Date of cancellation: 20230907 Granted publication date: 20210212 Pledgee: Haidian Beijing science and technology enterprise financing Company limited by guarantee Pledgor: Beijing China super industry information security technology Limited by Share Ltd. Registration number: Y2022110000042 |