CN114330553B - Digital acquisition system calibration method based on supervised learning - Google Patents
Digital acquisition system calibration method based on supervised learning Download PDFInfo
- Publication number
- CN114330553B CN114330553B CN202111639271.8A CN202111639271A CN114330553B CN 114330553 B CN114330553 B CN 114330553B CN 202111639271 A CN202111639271 A CN 202111639271A CN 114330553 B CN114330553 B CN 114330553B
- Authority
- CN
- China
- Prior art keywords
- data
- calibrated
- acquisition system
- digital acquisition
- calibration
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Landscapes
- Testing Or Calibration Of Command Recording Devices (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a digital acquisition system calibration method based on an integrated supervised learning algorithm, which comprises the steps of constructing a calibration system comprising an upper computer, a high-precision standard source and a high-precision digital acquisition system, constructing a calibration model, controlling the digital acquisition system to be calibrated and the high-precision digital acquisition system to acquire a source signal of the high-precision standard source by the upper computer, acquiring working operation parameter data of the digital acquisition system to be calibrated simultaneously, processing the data to be calibrated, reference data and the working operation parameter data by the upper computer to obtain a training sample data set, training and testing the calibration model, and calibrating the data to be calibrated of the digital acquisition system to be calibrated by adopting the trained calibration model after the calibration accuracy reaches a target. The invention improves the data processing method, comprehensively considers the signal data and the working operation parameter data of the digital acquisition system to be calibrated, and improves the calibration speed and the calibration precision of the digital acquisition system.
Description
Technical Field
The invention belongs to the technical field of high-precision data acquisition modules, and particularly relates to a digital acquisition system calibration method based on supervised learning.
Background
In modern industry, accuracy of a data acquisition system is very important in a very widely used electronic system. The calibration work of the digital acquisition system usually utilizes a high-precision multifunctional calibrator to complete the precision regression analysis of any quantity value by comparing a relative true value output by the calibrator with an actually measured quantity value of the digital acquisition system. In the traditional digital acquisition system calibration work, the acquired original data is subjected to simple abnormal value processing and then is transferred to a hardware storage medium for persistence, and a model is repeatedly imported to complete the calibration process. There are also a number of problems with such data processing approaches:
a. preprocessing rules of abnormal data cannot be updated and learned in real time, so that the situation of error processing after environmental factors are changed occurs;
b. the characteristics of the types (such as measurement types, measurement ranges, speed/frequency gears and the like) are not considered or processed, and the characteristics cannot be effectively coded and reasonably utilized;
c. the method has the advantages that the lost or abnormal processing vacant characteristic data during acquisition can not be processed, so that the local precision loss caused by incomplete data is caused;
e. mapping of probability distribution is not considered on the collected data so as to be better suitable for a nonlinear regression model;
e. the distribution of the sampling point sequence is not subjected to any preprocessing or feedback processing, so that the sampling points are not representative, and the model interpretability is lost;
f. screening and extracting the characteristics without analyzing the cause/correlation of the characteristics in the training data set, and establishing a reasonable regression model;
g. there is no strategy to update or learn the training data, resulting in inefficient hardware persistence and frequent data failures.
In addition, the hardware design of the digital acquisition system usually has some factors which are difficult to control, mainly including:
a. the thermal noise of the device is inconsistent with the working temperature curve, so that the working characteristics of the equipment at different temperatures are different;
b. the amplitude of the same alternating current signal has a remarkable correlation with the frequency; and under the condition of alternating current small signals, random noise is extremely high due to loss of sampling precision, and the regression curve shows obvious nonlinearity.
The calibration work of the traditional digital acquisition system usually does not consider the factors, so that the final calibration has larger errors.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a digital acquisition system calibration method based on supervised learning, and an improved data processing method is adopted to improve the calibration speed and the calibration precision.
In order to achieve the above purpose, the digital acquisition system calibration method based on supervised learning of the present invention comprises the following steps:
s1: the method comprises the following steps of constructing a calibration system, wherein the calibration system comprises an upper computer, a high-precision standard source and a high-precision digital acquisition system, wherein:
the high-precision standard source is used for outputting source signals to the digital acquisition system to be calibrated and the high-precision digital acquisition system under the control of the upper computer;
the high-precision digital acquisition system is used as a calibration reference, the digital acquisition system to be calibrated and the high-precision digital acquisition system simultaneously acquire source signals under the control of the upper computer, and the acquired sample data is sent to the upper computer;
the upper computer is used for controlling the high-precision standard source, the digital acquisition system to be calibrated and the high-precision digital acquisition system, sampling working operation parameters of the digital acquisition system to be calibrated, receiving sample data of the digital acquisition system to be calibrated and the high-precision digital acquisition system and calibrating the digital acquisition system to be calibrated;
s2: establishing a calibration model, wherein the input of the calibration model is the characteristic data corresponding to the data to be calibrated of the digital acquisition system to be calibrated, and the output of the calibration model is the calibration data corresponding to the data to be calibrated;
s3: the upper computer controls the high-precision standard source to output source signals to the digital acquisition system to be calibrated and the high-precision digital acquisition system, the digital acquisition system to be calibrated and the high-precision digital acquisition system are adjusted to the same gear and simultaneously acquire the source signals, and the digital acquisition system to be calibrated and the high-precision digital acquisition system feed back the acquired data to be calibrated and the reference data to the upper computer;
meanwhile, the upper computer collects the working operation parameters of the digital acquisition system to be calibrated in the working operation parameter set phi by adopting the same sampling rate to obtain working operation parameter data, and the working operation parameters contained in the initial working operation parameter set phi are set according to actual needs;
s4: the upper computer processes the signal data of the calibration digital acquisition system and the high-precision digital acquisition system to obtain a training sample data set, and the method specifically comprises the following steps:
s4.1: the upper computer detects abnormal data of the data to be calibrated, the reference data and the working operation parameter data and marks the detected abnormal data;
s4.2: for the data to be calibrated, the reference data and the working operation parameter data, if the data of a certain sampling point is missing or marked as abnormal data, interpolating the data of the sampling point;
s4.3: segmenting data to be calibrated, reference data and working operation parameter data which are subjected to abnormal data detection and interpolation by adopting a sliding window with the length of L; noting the number of the obtained sample data segments as N, the nth data to be calibrated as c n N =1,2, \ 8230, N, corresponding to the reference data of the same time period is r n Each operating parameter data is w n,m M =1,2, \ 8230, | φ |, and | φ | represents the number of selected operating parameters;
the upper computer determines gear parameters corresponding to each section of data according to control signals of a digital acquisition system to be calibrated and a high-precision digital acquisition system, wherein the gear parameters comprise a measurement type, a measurement range, a speed gear and a frequency gear, each gear parameter is subjected to characteristic coding, a numerical gear parameter is obtained, and gear data g are obtained by splicing n Recording the dimension as G;
corresponding time period of data c to be calibrated n M number of operating parameter data w n,m And gear data g n Splicing to obtain L + G dimensional characteristic data d with dimension of (M + 1) n ；
S4.4: for the characteristic data d n Performing dimensionless processing on each bit of feature data to obtain feature data d' n ；
S4.5: according to N characteristic data d' n Carrying out feature screening, then carrying out dimension reduction on feature data, and recording the obtained feature data as D n The dimension is K;
s4.6: data c to be calibrated n Corresponding characteristic data D n As input, reference data r of the corresponding time period n As data c to be calibrated n Forming a training sample; then dividing the N training samples into two sets which are respectively used as a training set and a test set;
s5: setting the input dimension of the calibration model as K and the output dimension as L, and training the calibration model by adopting a training set;
s6: testing the calibration model by adopting a test set, and calculating to obtain calibration accuracy;
s7: judging whether the calibration accuracy reaches a preset target or not, if not, entering a step S8, otherwise, entering a step S9;
s8: according to the feature screening result of the step S4.5, determining the working operation parameters related in the feature data after the dimension reduction, and updating a working operation parameter set phi; returning to the step S3;
s9: the upper computer controls the digital acquisition system to be calibrated to adjust to a required gear to acquire a signal to be acquired, and the digital acquisition system to be calibrated feeds back acquired data to be calibrated to the upper computer; simultaneously, the upper computer collects the working operation parameters of the digital collection system to be calibrated in the current working operation parameter set phi by adopting the same sampling rate;
the upper computer records the data to be calibrated of the current sampling point and the front L-1 sampling points as
Each operating parameter datum is recorded as->
Simultaneously acquiring the current gear parameter, and performing characteristic coding on the gear parameter by the same method in the step SS4.3 to obtain gear data->
Data to be calibrated>
M working operation parameter data->
And shift position data->
Splicing to obtain characteristic data with dimension of (M + 1) L + G->
Characteristic data are analysed in the same way as in step S4.4->
Each bit of feature data in the database is subjected to dimensionless processing to obtain feature data->
Based on the current feature screening result, the feature data is/are based on>
Performing dimensionality reduction to obtain feature data D';
inputting the characteristic data D' into the trained calibration model to obtain the data to be calibrated
Is corrected result of->
The invention relates to a digital acquisition system calibration method based on an integrated supervised learning algorithm, which comprises the steps of constructing a calibration system comprising an upper computer, a high-precision standard source and a high-precision digital acquisition system, constructing a calibration model, controlling the digital acquisition system to be calibrated and the high-precision digital acquisition system to acquire a source signal of the high-precision standard source by the upper computer, simultaneously acquiring working operation parameter data of the digital acquisition system to be calibrated, processing the data to be calibrated, reference data and the working operation parameter data by the upper computer to obtain a training sample data set to train and test the calibration model, and calibrating the data to be calibrated of the digital acquisition system to be calibrated by adopting the trained calibration model after the calibration accuracy reaches a target.
The invention improves the data processing method, comprehensively considers the signal data and the working operation parameter data of the digital acquisition system to be calibrated, and improves the calibration speed and the calibration precision of the digital acquisition system.
Drawings
FIG. 1 is a flow chart of an embodiment of a supervised learning based digital acquisition system calibration method of the present invention;
FIG. 2 is a block diagram of a calibration system of the present invention;
FIG. 3 is a flowchart of a method for determining a sampling interval during sampling at a varying interval in the present embodiment;
FIG. 4 is an exemplary diagram of a round of data optimization;
fig. 5 is a flow chart of data processing in the present invention. .
Detailed Description
The following description of the embodiments of the present invention is provided in order to better understand the present invention for those skilled in the art with reference to the accompanying drawings. It is to be expressly noted that in the following description, a detailed description of known functions and designs will be omitted when it may obscure the subject matter of the present invention.
Examples
Fig. 1 is a flowchart of an embodiment of a calibration method of a digital acquisition system based on supervised learning according to the present invention. As shown in fig. 1, the method for calibrating a digital acquisition system based on supervised learning of the present invention comprises the following specific steps:
s101: constructing a calibration system:
in order to realize the calibration of the digital acquisition system, the invention adopts an upper computer, a high-precision standard source and the high-precision digital acquisition system to build the calibration system. Fig. 2 is a block diagram of a calibration system in the present invention. As shown in fig. 2, the calibration system of the present invention includes an upper computer, a high precision standard source and a high precision digital acquisition system, wherein:
and the high-precision standard source is used for outputting source signals to the digital acquisition system to be calibrated and the high-precision digital acquisition system under the control of the upper computer.
The high-precision digital acquisition system is used as a calibration reference, and the precision of the high-precision digital acquisition system is higher than that of the digital acquisition system to be calibrated. And the digital acquisition system to be calibrated and the high-precision digital acquisition system simultaneously acquire source signals under the control of the upper computer, and send acquired sample data to the upper computer.
The upper computer is used for controlling the high-precision standard source, the digital acquisition system to be calibrated and the high-precision digital acquisition system, sampling working operation parameters of the digital acquisition system to be calibrated, receiving signal data of the digital acquisition system to be calibrated and the high-precision digital acquisition system and calibrating the digital acquisition system to be calibrated.
S102: constructing a calibration model:
and constructing a calibration model, wherein the input of the calibration model is the characteristic data corresponding to the data to be calibrated of the digital acquisition system to be calibrated, and the output of the calibration model is the calibration data corresponding to the data to be calibrated.
In the calibration work of the existing digital acquisition system, the calibration strategy is generally to perform simpler segmentation interpolation and high-order expansion on the basis of least square linear regression, and the calibration mode has a plurality of problems:
a. the high-order expansion of the linear regression is carried out blindly, so that the phenomenon of overfitting is easily caused, when the number of training points is equal to or insufficient than the order of the model, the training data is local and unilateral, and the high-dimensional model excessively memorizes the noise carried by the training data and ignores the real input/output relation. Reflecting the results, it appears that the model can regress the training values rather accurately, while the performance on the test values is not satisfactory.
b. Least squares linear regression uses the minimization of the mean square error as a regression criterion, resulting in its excessive sensitivity to subtle changes in input values, while the model itself is completely unmanageable for anomalous inputs (coarse values), which makes the stability of the model a great challenge.
c. The segmentation interpolation mode is suitable for the condition that the target regression value has obvious segmentation phenomenon, but the method can not accurately select and process the segmentation point, so that the model is complex and the precision is reduced.
d. The least square linear regression mode cannot make timely and accurate feedback on other characteristic factors (temperature, signal frequency and the like), and even if a multidimensional multiple linear regression mode is adopted, a regression model with interpretability and high precision cannot be perfectly trained;
e. the biggest defect of the traditional model is that point analysis is simply carried out on the regression model, noise distribution carried by training data and a target value of the training data and distribution conditions of regression coefficients are not considered, and interpretable nonlinear regression prediction can not be made according to a Bayesian method by using a plurality of prior conditions of the training data.
f. In addition, the least square linear regression has many defects that the characteristic dependence cannot be processed, the model is difficult to interpret, the regression model is single, the new environment cannot be adapted, the regression computation complexity is high, the incremental processing cannot be realized, and the like.
In view of the above problems, in this embodiment, according to the characteristics of the data to be calibrated of the digital acquisition system, a plurality of regression models are selected and integrated to obtain the calibration model.
Among the regression models, penalized linear regression models are commonly used to control the phenomenon of high-order model overfitting, such as Ridge regression model, lasso model, and Elastic-Net model. They may provide hyper-parameters for adjusting the feature weight distribution of the linear model. Because each numerical characteristic associated with the method can only ensure relative stability within a certain precision range, the actually measured value of each sampling point carries system noise, the punishment linear regression theory can relieve the interference of the precision noise on the robustness of the model to a certain extent, and can properly inhibit the overfitting phenomenon of the regression result.
And the high-robustness linear regression can control outlier interference, such as RANSAC model, huber model and Theil-Sen model. The negative influence of the abnormal value on the linear model is extremely severe, and the high-robustness linear regression model gradually identifies the bad value and constructs a global regression model in an iterative fitting mode of dividing sampling points.
Nonlinear models can describe real problems relatively accurately and improve model interpretability, and the models are approximately reconstructed by introducing kernel techniques on the basis of other known models, such as support vector regression, kernel ridge regression and Gaussian process regression. Due to the introduction of the kernel method, the model can well provide reasonable prior distribution hypothesis for the known quantity (training set) and the unknown quantity (characteristic weight coefficient), and the optimal regression value of the unknown quantity is obtained through cross validation and parameter optimization.
Therefore, in this embodiment, a Ridge model, a Lasso model, an Elastic-Net model, an ARD model, a RANSAC model, and a Huber model are preferably obtained from regression models, a GPR (gaussian process regression) model, a KRR (kernel Ridge regression) model, and an SVR (support vector regression) model are preferably obtained from nonlinear models, and the calibration models are obtained by integrating the models. As for the integration algorithm, the stacking algorithm is adopted in the embodiment, and the algorithm is a common integration algorithm, and the detailed process thereof is not described herein again.
S103: collecting sample data:
the upper computer controls the high-precision standard source to output source signals to the digital acquisition system to be calibrated and the high-precision digital acquisition system, the digital acquisition system to be calibrated and the high-precision digital acquisition system are adjusted to the same gear and simultaneously acquire the source signals, and the digital acquisition system to be calibrated and the high-precision digital acquisition system feed back the acquired sample data to the upper computer.
And simultaneously, the upper computer collects the working operation parameters of the digital acquisition system to be calibrated in the working operation parameter set phi by adopting the same sampling rate, and the working operation parameters contained in the initial working operation parameter set phi are set according to actual needs. Generally, the operating parameters include device temperature, boot time, ac signal frequency, etc.
The quality of data is crucial to subsequent processing and calibration, and good sampling point distribution design can ensure the rationality of data processing and save data processing time. Therefore, variable-interval sampling can be adopted during data acquisition, so that the data is more reasonable, and the setting of each sampling interval is particularly important. Fig. 3 is a flowchart of a method for determining a sampling interval during sampling at a variable interval in the present embodiment. As shown in fig. 3, in this embodiment, the sampling interval during sampling at the varying interval is determined by the following method:
s301: acquiring sample data:
in order to enable the sampling data to be more accurate, a high-precision digital acquisition system is adopted to acquire source signals at preset sampling intervals to obtain original data, and the number of sampling points is counted as H. Taking the original data as a data sequence X, wherein the h-th sampling data is X h Corresponding to a sampling time t h ，h＝1,2,…,H。
S302: calculating the original specific gravity:
calculating data X of each sampling point in data sequence X h Original specific gravity p of h ：
Wherein, the first and the second end of the pipe are connected with each other,
representing the f-th sample data x derived from the source signal f The theoretical value of (1).
S303: judging H original specific gravities p h Whether the preset linear distribution is satisfied, if not, the step S304 is executed, otherwise, the step S308 is executed.
S304: calculating the weight:
calculate each sample point data x h Weight w of h ：
S305: calculating an absolute weight:
calculate each sample point data x h Absolute weight W of h ：
S306: and (3) calculating quantization stepping:
calculate each sample point data x h And its next data x h+1 Quantized step λ in between h The specific method comprises the following steps:
1) Let h =1 and the cumulative error Δ =0.
2) Judging whether H =1 or H = H, if so, stepping the quantization by lambda h =1, step 6) is entered, otherwise step 3) is entered.
3) Determine if an absolute weight W h < 1, if yes, go to step 4), otherwise, the absolute weight W h Rounding as quantization step lambda h Proceed to step 6).
In this embodiment, a banker rounding method, also called a four-house six-in-five even-taking (also called a four-house six-in-five-left-two) method, is adopted for the rounding, and the error is smaller than that of a common rounding method.
4) Let Δ = Δ + W h 。
5) Judging whether the accumulated error delta is less than 1, if so, making quantization step lambda h =0, otherwise step quantization by λ h =1, update accumulated error Δ = Δ -1. Then step 6) is entered.
6) And judging whether H is less than H, if so, entering a step 7), and otherwise, entering a step 8).
7) Let h = h +1, return to step 2).
8) Judging whether to use
If so, the quantization step calculation is ended, otherwise step 9) is entered.
9) Quantizing step lambda from H h Middle screening out maximum value lambda h * Let us order
I.e. by the maximum value lambda h * The difference between the quantization step and the sampling point is compensated, so that the number of sampling points of the optimized data is the same as that of the original data. The quantization step calculation ends.
S307: generating optimized data:
for each quantization step λ in turn h Making a judgment if λ h =0, then sample point data x is discarded h If λ is h =1, then sample point data x h Addition to optimizationData and saving its sampling time t h If λ is h If > 1, then sample point data x is added h Adding to optimized data while interpolating lambda between the data and the previous optimized data using mean interpolation h -1 optimization datum, while determining the sampling instant of each inserted optimization datum on the basis of the sampling instants of this datum and of the previous optimization datum.
And taking the optimized data as a data sequence X, and returning to the step S302.
S308: determining a sampling interval:
and determining the sampling interval of each data in the data to be acquired according to the sampling time of each sampling data in the data sequence X.
FIG. 4 is an exemplary diagram of a round of data optimization. As shown in fig. 4, the original data includes 42 sampling points, and the optimized data is finally obtained by sequentially calculating the original specific gravity, the weight, the absolute weight, and the quantization step. Therefore, the head data and the tail data of the optimized data are kept unchanged as anchor points, the middle sampling point data is redistributed, the linear characteristic of the optimized data is better than that of the original data, and the sampling interval meeting the requirement can be determined from the final optimized data through a plurality of rounds of iterative optimization.
S104: sample data processing:
and the upper computer processes the data to be calibrated, the reference data and the working operation parameter data to obtain a training sample data set. Fig. 5 is a flow chart of data processing in the present invention. As shown in fig. 5, the specific steps of the data processing in the present invention include:
s501: and (3) exception data processing:
in both linear and nonlinear models, the negative effects caused by abnormal characteristic data are destructive, so the upper computer of the invention needs to detect abnormal data of the calibration data and the reference data first and mark the detected abnormal data. The specific method for detecting abnormal data in this embodiment is as follows:
and calculating the similarity between the data value of each sampling point in the data to be calibrated and the reference data and the theoretical true value and the similarity between the data value of each sampling point in the reference data and the data values of the left and right adjacent sampling points, and marking the data value of each sampling point as abnormal data when any one of the similarities is greater than a corresponding preset threshold.
In practical application, because the data to be calibrated and the reference data obtained by the upper computer are fed back by the corresponding digital acquisition systems, and there may be a case of data transmission error, when abnormal data occurs, the upper computer may require the corresponding digital acquisition systems to retransmit, so as to eliminate the data transmission problem.
S502: data interpolation:
and for the data to be calibrated, the reference data and the working operation parameter data, if the data of a certain sampling point is missing or marked as abnormal data, interpolating the data of the sampling point. The specific interpolation method can select different feature types and regression models, and generally comprises zero interpolation, mean interpolation, neighbor interpolation, iterative interpolation and the like. In practical applications, the optimal interpolation strategy may be determined experimentally.
S503: acquiring characteristic data:
and segmenting the data to be calibrated, the reference data and the working operation parameter data which are subjected to abnormal data detection and interpolation by adopting a sliding window with the length of L. Recording the number of the obtained sample data segments as N, and the nth data to be calibrated as c n N =1,2, \8230, N, and the reference data of the corresponding same time period is r n Each operating parameter data is w n,m M =1,2, \8230 |, | φ | represents the number of selected operating parameters.
And the upper computer determines gear parameters corresponding to each section of data according to control signals of the digital acquisition system to be calibrated and the high-precision digital acquisition system, wherein the gear parameters comprise a measurement type, a measurement range, a speed gear and a frequency gear. Because the feature data depended on by any regression model in machine learning should be numerical type, and the gear parameters include type data, each gear parameter needs to be subjected to feature coding, numerical gear parameters are obtained, and the gear number is obtained by splicingAccording to g n Let it be dimension G. The common characteristic codes comprise unique hot codes and label codes, and can be selected according to actual requirements.
Corresponding time period of data c to be calibrated n M number of operating parameter data w n,m And gear data g n Splicing to obtain characteristic data d with dimension of (M + 1) L + G n 。
S504: dimensionless processing:
due to characteristic data d n The data to be calibrated and the gear data are combined to obtain, and in order to eliminate the influence of dimension, the characteristic data d needs to be subjected to n Performing dimensionless processing on each bit of feature data to obtain feature data d' n 。
S505: and (3) feature screening:
according to N characteristic data d' n Carrying out feature screening, then carrying out dimension reduction on feature data, and recording the obtained feature data as D n The dimension is K.
The specific method of feature screening can be selected according to specific needs. In the embodiment, N pieces of feature data d' n The method comprises the steps of extracting main features through principal component analysis or independent component analysis, and removing the features with weak influence through correlation analysis or variance thresholding analysis, so that feature screening is achieved. In general, the feature data D n The dimension of (a) is gradually reduced with the number of rounds of training of the calibration model,
s506: generating a training sample:
data c to be calibrated n Corresponding characteristic data D n As input, reference data r of the corresponding time period n As data c to be calibrated n Constitutes a training sample. Then, the N training samples are divided into two sets which are respectively used as a training set and a test set.
S105: training a calibration model:
and setting the input dimension of the calibration model as K and the output dimension as L, and training the calibration model by adopting a training set.
S106: and (3) testing a calibration model:
and testing the calibration model by adopting a test set, and calculating to obtain the calibration accuracy. In this embodiment, the calibration accuracy is the euclidean distance between the calibration result obtained by inputting the feature data and the corresponding reference data.
S107: judging whether the calibration accuracy reaches a preset target or not, if not, entering step S108, otherwise, entering step S109;
s108: and (3) updating a working operation parameter set:
and determining the working operation parameters related in the feature data after the dimension reduction according to the feature screening result of the step S505, and updating the working operation parameter set phi. The process returns to step S103.
S109: and (3) data calibration:
the upper computer controls the digital acquisition system to be calibrated to adjust to a required gear to acquire a signal to be acquired, and the digital acquisition system to be calibrated feeds back acquired data to be calibrated to the upper computer; and simultaneously, the upper computer collects the working operation parameters of the digital collection system to be calibrated in the current working operation parameter set phi by adopting the same sampling rate.
The upper computer records the data to be calibrated of the current sampling point and the front L-1 sampling points as
Each operating parameter datum is recorded as->
Meanwhile, the current gear parameter is obtained, the gear parameter is subjected to feature coding by adopting the same method in the step S503, and gear data is obtained>
Data to be calibrated>
M working operation parameter data->
And shift position data->
Splicing to obtain characteristic data with dimension of (M + 1) L + G->
The same method is used for the feature data ≥ in step S504>
Each bit of feature data in the database is subjected to dimensionless processing to obtain feature data->
Based on the current feature screening result, the feature data is/are based on>
Reducing the dimension to obtain characteristic data D';
inputting the characteristic data D' into the trained calibration model to obtain the data to be calibrated
Is corrected result of->
Although the illustrative embodiments of the present invention have been described in order to facilitate those skilled in the art to understand the present invention, it is to be understood that the present invention is not limited to the scope of the embodiments, and that various changes may be made apparent to those skilled in the art as long as they are within the spirit and scope of the present invention as defined and defined in the appended claims, and all matters of the invention using the inventive concepts are protected.
Claims (5)
1. A digital acquisition system calibration method based on supervised learning is characterized by comprising the following steps:
s1: the method comprises the following steps of constructing a calibration system, wherein the calibration system comprises an upper computer, a high-precision standard source and a high-precision digital acquisition system, wherein:
the high-precision standard source is used for outputting source signals to the digital acquisition system to be calibrated and the high-precision digital acquisition system under the control of the upper computer;
the high-precision digital acquisition system is used as a calibration reference, the digital acquisition system to be calibrated and the high-precision digital acquisition system simultaneously acquire source signals under the control of the upper computer, and the acquired sample data is sent to the upper computer;
the upper computer is used for controlling the high-precision standard source, the digital acquisition system to be calibrated and the high-precision digital acquisition system, sampling working operation parameters of the digital acquisition system to be calibrated, receiving sample data of the digital acquisition system to be calibrated and the high-precision digital acquisition system and calibrating the digital acquisition system to be calibrated;
s2: establishing a calibration model, wherein the input of the calibration model is the characteristic data corresponding to the data to be calibrated of the digital acquisition system to be calibrated, and the output of the calibration model is the calibration data corresponding to the data to be calibrated;
s3: the upper computer controls the high-precision standard source to output source signals to the digital acquisition system to be calibrated and the high-precision digital acquisition system, the digital acquisition system to be calibrated and the high-precision digital acquisition system are adjusted to the same gear and simultaneously acquire the source signals, and the digital acquisition system to be calibrated and the high-precision digital acquisition system feed back the acquired data to be calibrated and the reference data to the upper computer;
meanwhile, the upper computer collects the working operation parameters of the digital acquisition system to be calibrated in the working operation parameter set phi by adopting the same sampling rate to obtain working operation parameter data, and the working operation parameters contained in the initial working operation parameter set phi are set according to actual needs;
s4: the upper computer processes the data to be calibrated, the reference data and the working operation parameter data to obtain a training sample data set, and the method specifically comprises the following steps:
s4.1: the upper computer detects abnormal data of the data to be calibrated, the reference data and the working operation parameter data and marks the detected abnormal data;
s4.2: for data to be calibrated, reference data and working operation parameter data, if data of a certain sampling point is missing or marked as abnormal data, interpolating the data of the sampling point;
s4.3: segmenting data to be calibrated, reference data and working operation parameter data which are subjected to abnormal data detection and interpolation by adopting a sliding window with the length of L; noting the number of the obtained sample data segments as N, the nth data to be calibrated as c n N =1,2, \ 8230, N, corresponding to the reference data of the same time period is r n Each operating parameter data is w n,m M =1,2, \8230, | phi | representing the number of selected operating parameters;
the upper computer determines gear parameters corresponding to each section of data according to control signals of a digital acquisition system to be calibrated and a high-precision digital acquisition system, wherein the gear parameters comprise a measurement type, a measurement range, a speed gear and a frequency gear, each gear parameter is subjected to characteristic coding, a numerical gear parameter is obtained, and gear data g are obtained by splicing n Recording the dimension as G;
corresponding time period of data c to be calibrated n M number of operating parameter data w n,m And gear data g n Splicing to obtain L + G dimensional characteristic data d with dimension of (M + 1) n ；
S4.4: for the feature data d n Performing dimensionless processing on each bit of feature data to obtain feature data d' n ；
S4.5: according to N characteristic data d' n Carrying out feature screening, then carrying out dimension reduction on feature data, and recording the obtained feature data as D n The dimension is K;
s4.6: data c to be calibrated n Corresponding characteristic data D n As input, reference data r of the corresponding time period n As data c to be calibrated n Forming a training sample; then dividing the N training samples into two sets which are respectively used as a training set and a test set;
s5: setting the input dimension of the calibration model as K and the output dimension as L, and training the calibration model by adopting a training set;
s6: testing the calibration model by adopting a test set, and calculating to obtain calibration accuracy;
s7: judging whether the calibration accuracy reaches a preset target or not, if not, entering a step S8, otherwise, entering a step S9;
s8: according to the feature screening result of the step S4.5, determining the working operation parameters related in the feature data after dimension reduction, and updating a working operation parameter set phi; returning to the step S3;
s9: the upper computer controls the digital acquisition system to be calibrated to adjust to a required gear to acquire a signal to be acquired, and the digital acquisition system to be calibrated feeds back acquired data to be calibrated to the upper computer; meanwhile, the upper computer collects the working operation parameters of the digital collection system to be calibrated in the current working operation parameter set phi by adopting the same sampling rate;
the upper computer records the data to be calibrated of the current sampling point and the front L-1 sampling points as
Each operating parameter datum is recorded as->
Simultaneously acquiring a current gear parameter, and performing characteristic coding on the gear parameter by adopting the same method in the step SS4.3 to obtain gear data->
Data to be calibrated>
M working operation parameter data->
And shift position data->
Splicing to obtain characteristic data (M + 1) L + G dimension>
The same method is used for the characteristic data ≥ in step S4.4>
Each bit of feature data in the database is subjected to dimensionless processing to obtain feature data->
Based on the current feature screening result, the feature data is/are based on>
Reducing the dimension to obtain characteristic data D';
inputting the characteristic data D' into the trained calibration model to obtain the data to be calibrated
Is corrected result of->
2. The method of claim 1, wherein the calibration model is integrated from a Ridge model, a Lasso model, an Elastic-Net model, an ARD model, a RANSAC model, a Huber model, a GPR model, a KRR model, and an SVR model.
3. The method for calibrating a digital acquisition system according to claim 1, wherein the data acquisition in step S3 uses variable-interval sampling, and the sampling interval is determined by the following method:
S3.1：acquiring a source signal by a high-precision data acquisition system at preset sampling intervals to obtain original data, and counting the number of sampling points as H; taking the original data as a data sequence X, wherein the h-th sampling data is X h Corresponding to a sampling time t h ，h＝1,2,…,H；
S3.2: calculating data X of each sampling point in data sequence X h Original specific gravity p of h ：
Wherein the content of the first and second substances,
representing the f-th sample data x derived from the source signal f The theoretical value of (A);
s3.3: judging H original specific gravities p h Whether the preset linear distribution is met or not, if not, entering a step S3.4, otherwise, entering a step S3.8;
s3.4: calculate each sample point data x h Weight w of h ：
S3.5: calculate each sample point data x h Absolute weight W of h ：
S3.6: calculate each sample point data x h With its next data x h+1 Quantized step λ in between h The specific method comprises the following steps:
1) Let h =1, cumulative error Δ =0;
2) Judging whether H =1 or H = H, if so, stepping the quantization by lambda h =1, go to step 6), otherwise go toStep 3) is carried out;
3) Determine if an absolute weight W h < 1, if yes, go to step 4), otherwise the absolute weight W will be h Rounding as quantization step lambda h Entering step 6);
4) Let Δ = Δ + W h ；
5) Judging whether the accumulated error delta is less than 1, if so, making the quantization step lambda h =0, otherwise let the quantization step by λ h =1, update accumulated error Δ = Δ -1; then entering step 6);
6) Judging whether H is less than H, if so, entering a step 7), and if not, entering a step 8);
7) Let h = h +1, return to step 2);
8) Judging whether to use
If yes, finishing the quantization stepping calculation, otherwise, entering a step 9);
9) Quantizing step lambda from H h Screening out the maximum value
Make->
Finishing the quantitative stepping calculation;
s3.7: for each quantization step λ in turn h Making a judgment if λ h If =0, the sample point data x is discarded h If λ is h =1, then sample point data x h Adding to the optimized data and saving its sampling time t h If λ is h If > 1, the sampling point data x is h Adding to the optimized data while interpolating lambda between the data and the previous optimized data using mean interpolation h -1 optimization datum, while determining the sampling instant of each insertion optimization datum on the basis of the sampling instants of this datum and of the previous optimization datum;
taking the optimized data as a data sequence X, and returning to the step S3.2;
s3.8: and determining the sampling interval of each data in the data to be acquired according to the sampling time of each sampling data in the data sequence X.
4. A method for calibrating a digital acquisition system according to claim 3, characterized in that the absolute weight W in said step 3.6 h And rounding by adopting a banker rounding method.
5. The calibration method for digital acquisition system according to claim 1, wherein the detection method for abnormal data in step S4.1 is: and calculating the similarity between the data value of each sampling point in the data to be calibrated, the reference data and the working operation parameter data and the theoretical true value and the similarity between the data value of each sampling point and the data values of the left and right adjacent sampling points, and marking the data value of each sampling point as abnormal data when any one of the similarities is greater than a corresponding preset threshold.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111639271.8A CN114330553B (en) | 2021-12-29 | 2021-12-29 | Digital acquisition system calibration method based on supervised learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111639271.8A CN114330553B (en) | 2021-12-29 | 2021-12-29 | Digital acquisition system calibration method based on supervised learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114330553A CN114330553A (en) | 2022-04-12 |
| CN114330553B true CN114330553B (en) | 2023-04-07 |
Family
ID=81017388
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111639271.8A Active CN114330553B (en) | 2021-12-29 | 2021-12-29 | Digital acquisition system calibration method based on supervised learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114330553B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116840759A (en) * | 2023-07-04 | 2023-10-03 | 成都泰格微电子研究所有限责任公司 | Rapid calibration system and method suitable for hybrid integrated circuit test system |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103177551A (en) * | 2011-12-21 | 2013-06-26 | 北京普源精电科技有限公司 | Data collection card having self calibration function and data collection apparatus |
| CN103744045A (en) * | 2013-12-17 | 2014-04-23 | 中国电子科技集团公司第二十研究所 | Digital type universal meter automatic calibration system |
| CN106777809A (en) * | 2017-01-13 | 2017-05-31 | 清华大学 | A kind of locomotive traction computation model calibration method based on actual operating data |
| CN107247249A (en) * | 2017-07-12 | 2017-10-13 | 周浩 | A kind of alternating current-direct current ammeter tester calibrating installation and calibration method |
| CN109946636A (en) * | 2019-04-03 | 2019-06-28 | 中国计量大学 | A kind of digital multimeter indicating value automatic calibrating method based on deep learning |
| CN109961186A (en) * | 2019-03-22 | 2019-07-02 | 大唐环境产业集团股份有限公司 | Desulphurization system operating parameter prediction technique based on decision tree and BP neural network |
| CN111537944A (en) * | 2020-05-20 | 2020-08-14 | 南京能瑞自动化设备股份有限公司 | Calibration parameter determination method, microcontroller, metering chip and upper computer |
| CN112528566A (en) * | 2020-12-18 | 2021-03-19 | 江西理工大学南昌校区 | Real-time air quality data calibration method and system based on AdaBoost training model |
| CN113312849A (en) * | 2021-06-15 | 2021-08-27 | 上海耀斑信息科技有限公司 | Automatic updating and calibrating algorithm for water supply network hydraulic model |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7146283B2 (en) * | 2004-08-16 | 2006-12-05 | National Instruments Corporation | Calibrating analog-to-digital systems using a precision reference and a pulse-width modulation circuit to reduce local and large signal nonlinearities |
-
2021
- 2021-12-29 CN CN202111639271.8A patent/CN114330553B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103177551A (en) * | 2011-12-21 | 2013-06-26 | 北京普源精电科技有限公司 | Data collection card having self calibration function and data collection apparatus |
| CN103744045A (en) * | 2013-12-17 | 2014-04-23 | 中国电子科技集团公司第二十研究所 | Digital type universal meter automatic calibration system |
| CN106777809A (en) * | 2017-01-13 | 2017-05-31 | 清华大学 | A kind of locomotive traction computation model calibration method based on actual operating data |
| CN107247249A (en) * | 2017-07-12 | 2017-10-13 | 周浩 | A kind of alternating current-direct current ammeter tester calibrating installation and calibration method |
| CN109961186A (en) * | 2019-03-22 | 2019-07-02 | 大唐环境产业集团股份有限公司 | Desulphurization system operating parameter prediction technique based on decision tree and BP neural network |
| CN109946636A (en) * | 2019-04-03 | 2019-06-28 | 中国计量大学 | A kind of digital multimeter indicating value automatic calibrating method based on deep learning |
| CN111537944A (en) * | 2020-05-20 | 2020-08-14 | 南京能瑞自动化设备股份有限公司 | Calibration parameter determination method, microcontroller, metering chip and upper computer |
| CN112528566A (en) * | 2020-12-18 | 2021-03-19 | 江西理工大学南昌校区 | Real-time air quality data calibration method and system based on AdaBoost training model |
| CN113312849A (en) * | 2021-06-15 | 2021-08-27 | 上海耀斑信息科技有限公司 | Automatic updating and calibrating algorithm for water supply network hydraulic model |
Non-Patent Citations (4)
| Title |
|---|
| Deep Learning for Model Parameter Calibration in Power Systems;Safwan Wshah等;《2020 IEEE International Conference on Power Systems Technology (POWERCON)》;1-6 * |
| Single-parameter-learning-based finite-time tracking control of underactuated MSVs under input saturation;GuibingZhu等;《Control Engineering Practice》;1-11 * |
| 基于FPGA的多通道数据采集系统的设计;王旭东;《中国优秀硕士学位论文全文数据库信息科技辑》;I135-198 * |
| 高平坦度示波器通道设计及校准技术研究;余中豪;《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》;C042-872 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114330553A (en) | 2022-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111079836B (en) | Process data fault classification method based on pseudo label method and weak supervised learning | |
| CN109446189A (en) | A kind of technological parameter outlier detection system and method | |
| CN109085805B (en) | Industrial process fault detection method based on multi-sampling-rate factor analysis model | |
| CN114330553B (en) | Digital acquisition system calibration method based on supervised learning | |
| CN111861272A (en) | Multi-source data-based complex electromechanical system abnormal state detection method | |
| CN111241744A (en) | Low-pressure casting machine time sequence data abnormity detection method based on bidirectional LSTM | |
| CN112284440A (en) | Sensor data deviation self-adaptive correction method | |
| CN115062272A (en) | Water quality monitoring data abnormity identification and early warning method | |
| CN112000081B (en) | Fault monitoring method and system based on multi-block information extraction and Mahalanobis distance | |
| CN114015825A (en) | Method for monitoring abnormal state of blast furnace heat load based on attention mechanism | |
| CN110084301B (en) | Hidden Markov model-based multi-working-condition process working condition identification method | |
| CN113326744B (en) | Method and system for detecting on-orbit state abnormity of spacecraft | |
| CN114216877A (en) | Automatic detection and reconstruction method and system for spectral peak in tea near infrared spectral analysis | |
| CN112215495B (en) | Pollution source contribution calculation method based on long-time and short-time memory neural network | |
| CN106599391B (en) | Association vector machine soft measurement modeling method based on dynamic weighting of triangle angle values | |
| CN111863153A (en) | Method for predicting total amount of suspended solids in wastewater based on data mining | |
| CN113312587B (en) | Sensor acquisition data missing value processing method based on ARIMA prediction and regression prediction | |
| CN115436342A (en) | Method and device for reducing LIBS detection uncertainty among multiple batches of samples | |
| Yu et al. | Design for an SPRT control scheme based on linguistic data | |
| CN109613618B (en) | Long-term drift error compensation method of gravity sensor | |
| CN111781824B (en) | Self-adaptive soft measurement method and system based on vine copula quantile regression | |
| CN112541162A (en) | Time axis self-regularizing method for multi-telemetry parameter joint analysis | |
| CN117688388B (en) | Soft measurement method and system based on data enhancement and prediction combined learning | |
| CN111104985A (en) | Weighting sliding window method for asynchronous track association | |
| CN114386196B (en) | Method for evaluating mechanical property prediction accuracy of plate strip |
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 |