CN111458672A - Multi-range current measurement calibration system based on machine learning - Google Patents
Multi-range current measurement calibration system based on machine learning Download PDFInfo
- Publication number
- CN111458672A CN111458672A CN202010346676.1A CN202010346676A CN111458672A CN 111458672 A CN111458672 A CN 111458672A CN 202010346676 A CN202010346676 A CN 202010346676A CN 111458672 A CN111458672 A CN 111458672A
- Authority
- CN
- China
- Prior art keywords
- current
- range
- module
- terminal
- measuring
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R35/00—Testing or calibrating of apparatus covered by the other groups of this subclass
- G01R35/005—Calibrating; Standards or reference devices, e.g. voltage or resistance standards, "golden" references
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Abstract
The invention relates to a multi-range current measurement calibration system based on machine learning, which comprises a multi-range current measurement terminal, a load simulation terminal and an upper computer, wherein a three-range current measurement circuit of the current measurement terminal is connected with the load simulation terminal, supplies power to the load simulation terminal, simultaneously acquires analog current signals generated by the load simulation terminal through small, medium and large three-range circuits, and sends the analog current signals to the upper computer after processing; the load simulation terminal is used for generating simulation current signals with different sizes; the coordination module of the upper computer sends different current actual values to the load simulation terminal, and the load simulation terminal generates corresponding simulation current signals; the receiving and transmitting module receives current signals of each range sent by the current measuring terminal, the current signals are stored in a database after being processed, the sample generating module forms an initial calibration data set according to the current actual value and three corresponding range current data, and the initial calibration data set is input into the machine learning module to be constructed and trained to obtain a calibration model. The system is beneficial to improving the measurement accuracy.
Description
Technical Field
The invention relates to the technical field of embedded equipment current measurement, in particular to a multi-range current measurement calibration system based on machine learning.
Background
With the continuous development of the internet of things technology, embedded systems have been widely deployed in large quantities. For battery-powered embedded systems, energy consumption has become a quality attribute of major concern in development, operation, and maintenance. In addition, as reported in literature, 80% of energy consumption of the embedded system is closely related to software operation activities. Therefore, the design and implementation of the energy consumption measurement system during the operation of the embedded system software have important significance and value for analyzing, evaluating and optimizing the energy efficiency of the embedded system.
The energy consumption E of the embedded system software during operation can be calculated by the formula (1).
Pj=Uj×Ij(1)
In formula (1), T0And TnThe start time and the end time of the software operation, and TjAnd Tj+1Respectively representing sampling points of power measurement at the jth moment and the (j +1) moment in software operation; pjAnd Pj+1The measured instantaneous power for the j-th and (j +1) -th sample points, respectively. By accumulating the area of the trapezoid formed by two adjacent sampling time points and the corresponding instantaneous power, the energy consumption of the embedded system software during operation can be approximately calculated, as shown in fig. 1. Instantaneous power PjBy means of the voltage U at the j-th sampling pointjAnd current IjAnd (4) calculating. U shapejCan be measured directly, and IjOnly by indirect measurement. Depending on the operating state of the embedded system (e.g. standby, data acquisition and communication, etc.), the software runs on a current IjCan vary from tens of microamperes to nearly a thousand milliamperes, and can vary up to frequencies on the order of M Hz. Thus, the embedded system software runtime workerThe working current has the characteristics of large variation range and high variation frequency.
At present, dynamic current measurement methods based on inductance, coulometer and resistance are available. The inductance-based method has a problem of poor resistance to electromagnetic noise, and the coulometer-based method is difficult to capture the current change of high frequency in real time. The method based on the resistance only is more suitable for measuring the current when the embedded system software operates, and the working principle of the method is shown in figure 2. In fig. 2, a sampling resistor R is connected in series with a system under test and a power supply; the voltage generated by the working current of the system to be tested on R is amplified by an amplifier and then is transmitted to an A/D converter for digital processing.
In fig. 2, the supply voltage is divided over the system under test and the resistor R. On one hand, the distributed voltage on the R is increased along with the increase of the working current of the system to be tested, and the value of the R cannot be too large in order to ensure the normal working voltage of the system to be tested. On the other hand, the smaller the value of R, the lower the voltage drop across R, which requires a larger multiple of amplifiers or higher resolution a/D converters to meet the wide range current measurement requirements. However, increasing the amplification factor of the amplifier leads to a decrease in measurement accuracy due to the amplification of noise, and increasing the resolution of the a/D converter leads to an increase in cost. The two factors are comprehensively considered, and the existing measuring method based on the resistance adopts a multi-range current measuring idea: the current values in different ranges are divided into different ranges, different R values or different amplification factors are set for the different ranges, and the measurement range of the current is effectively enlarged while the normal work of a system to be measured can be guaranteed through automatic switching of the ranges. Figure 3 shows a schematic of a three-range current measurement. 0 to i in FIG. 3max1、0~imax2And 0 to imax3The three ranges correspond to small, medium and large ranges, and the set R values become smaller (the small range set R value is the largest). In fig. 3, there are two range switching points imax1And imax2And defining boundary points for switching the small current measuring circuit to the medium current measuring circuit and the medium current measuring circuit to the large current measuring circuit respectively. The existing resistance-based method can better meet the operation requirement of embedded system software through automatic range switchingA wide range of current measurement needs. There are still some problems:
(1) automatic range switching requires some auxiliary circuits and corresponding interrupt handlers, thereby introducing certain complexity in the dynamic current measurement system;
(2) measurement noise may cause dead range switching. E.g. due to the presence of noise in fig. 2, rng1The actual value of the range of current values may be greater than imax1In this scenario, the range switching should be performed but no switching is performed; and rng2The actual value for the range of current values may be less than imax1In this scenario, the range switching should not be performed but performed. Similarly, rng3And rng4Range (i)max2Neighborhood range) failed range switching may also occur.
(3) The switching of the measuring range has certain time delay, and the current value in the switching process is difficult to be accurately given, so that certain measuring error is introduced.
Disclosure of Invention
The invention aims to provide a multi-range current measurement calibration system based on machine learning, which is beneficial to improving the measurement accuracy.
In order to achieve the purpose, the invention adopts the technical scheme that: a multi-range current measurement calibration system based on machine learning is characterized by comprising a multi-range current measurement terminal, a load simulation terminal and an upper computer;
the current measuring terminal comprises a three-range current measuring circuit and a current signal acquisition and processing system, the three-range current measuring circuit comprises a circuit for measuring current in small, medium and large ranges at the same time, the three-range current measuring circuit is connected with a load simulation terminal, used for supplying power to the load simulation terminal and simultaneously collecting the simulation current signals generated by the load simulation terminal through circuits with small, medium and large measuring ranges, the current signal acquisition and processing system comprises three ADC modules, a timing module, a DMA transmission module and a communication module, the ADC module is connected with a three-range current measuring circuit, the timing module generates an interrupt signal and simultaneously starts the three ADC modules, and each ADC module respectively converts three analog current signals acquired by the small, medium and large ranges into digital current signals and then sends the digital current signals to the upper computer through the DMA transmission module and the communication module;
the load simulation terminal is used for generating a resistance value with a set size through the variable resistor and generating simulation current signals with different sizes under the power supply voltage of the three-range current measuring circuit;
the upper computer comprises a coordination module, a transceiving module, a filtering module, a storage module, a database, a sample generation module and a machine learning module; the coordination module sends different current actual values to a load simulation terminal, the load simulation terminal calculates a resistance value to be generated according to the power supply voltage and the current actual values, controls the variable resistor to generate a corresponding resistance value, and then outputs a simulation current signal under the power supply voltage; the receiving and transmitting module receives current signals of each range sent by the current measuring terminal, noise is filtered by the filtering module, current data of small, medium and large ranges are stored in the database through the storage module, and the sample generating module forms an initial calibration data set according to the actual current value sent by the cooperation module and the current data of the three ranges corresponding to the actual current value; the machine learning module builds and trains a calibration model based on the initial calibration data set.
Further, the small, medium and large three measuring ranges are [0,25mA ], (25mA,250mA ] and (250mA,2500mA ].
Further, the method for generating the initial calibration data set comprises the following steps:
the coordination module sends a load definition (y, t) comprising an actual current value y and a duration time t to a load simulation terminal, the load simulation terminal generates a simulation current signal corresponding to the actual current value y within the duration time t, the current measurement terminal respectively carries out current measurement through circuits with small, medium and large ranges within the duration time t, a group of readings are respectively obtained in each range, the readings are sent to an upper computer after being processed by a current signal acquisition and processing system, and the readings corresponding to each range are stored in a database after being processed by the upper computer; the sample generation moduleTaking the average value of each group of readings to obtain readings x corresponding to small, medium and large measuring ranges1、x2And x3Further, a sample point data (x) is generated1,x2,x3,y);
The coordination module respectively generates a set number of current actual values in the small, medium and large measuring range, and the method is repeated to construct the data sets DS with the small, medium and large measuring ranges1、DS2And DS3。
Further, the machine learning module constructs a calibration model as follows:
1) from a three-range data set DS1、DS2And DS3Randomly extracting a set number of samples as test samples to form a corresponding test setAndfurther generating training data sets for small, medium and large three-range calibrationAndand testing the data setAndtraining data setAndas shown in equation (2), test data setAndis as defined in formula (3):
2) constructing a calibration model of each measuring range: based on a machine learning algorithm, respectively constructing an optimal calibration model with small, medium and large measuring rangesAnd
3) generating a synthetic data set: optimal calibration model according to each measuring rangeAndand training data set Andgenerating a synthesized training set DSTrThe specific definition is shown as a formula (4); optimal calibration model according to each measuring rangeAndand testing the data setAndgenerating a synthesized test set DSTsSpecifically defined as shown in formula (5):
in the formulae (4) and (5), DSTrAnd DSTsFor each sample point in (x)1,x2,x3,a1,a2,a3Y) represents wherein x1、x2、x3Three input variables correspond to readings of three ranges, a1、a2、a3Three input variables are respectively composed of Andcalculating to obtain a variable y which is an actual current value;
4) constructing a synthetic calibration model: from the resultant data set DSTrAnd DSTsBased on a machine learning algorithm, a linear or nonlinear relation between the measuring point in the range transition neighborhood and the optimal calibration model of each range is determined, and then an optimal synthetic calibration model is constructed, so that the error between the current value output by the synthetic calibration model and the current actual value is minimum.
Further, the optimal calibration model and the optimal synthetic calibration model of each measurement range ARE constructed by adopting the average relative error ARE defined by the formula (6) as a performance evaluation index:
wherein, yiThe value of the output variable, i.e. the current actual value,the method comprises the steps that a predicted value of current obtained according to input of the ith sample in a test set is shown, n represents the size of the sample of the test set, and the smaller the value of ARE is, the better the calibration model is;
the machine learning algorithm adopted for constructing the optimal calibration model and the optimal synthetic calibration model of each range is linear regression L R or support vector regression SVR.
Compared with the prior art, the invention has the following beneficial effects: aiming at the problem that the current is difficult to accurately measure due to the dynamic change in a wide range when the software of the embedded equipment runs, the multi-range current measurement and calibration system based on machine learning is provided, the system adopts three ranges of small, medium and large to simultaneously measure the current, a calibration model is established and calibration training is carried out to obtain an optimal calibration model. The calibration model obtained by the system is used for current measurement, range switching is not needed, and compared with a range switching method, the method can effectively reduce measurement errors and improve measurement accuracy.
Drawings
FIG. 1 is a schematic diagram of approximate calculation of energy consumption during operation of embedded system software in the prior art.
Fig. 2 is a schematic diagram of a prior art resistance-based current measurement.
Fig. 3 is a schematic diagram of a three-range current measurement in the prior art.
Fig. 4 is a schematic diagram of the overall structure of the system according to the embodiment of the invention.
FIG. 5 is a schematic diagram of a three-range current measurement circuit in an embodiment of the invention.
Fig. 6 is a schematic diagram of a current signal collecting and processing system according to an embodiment of the present invention.
Fig. 7 is a flow chart of the multi-range current measurement calibration based on machine learning in the embodiment of the present invention.
Detailed Description
The invention is described in further detail below with reference to the figures and the embodiments.
The invention provides a multi-range current measurement calibration system based on machine learning, which comprises a multi-range current measurement terminal, a load simulation terminal and an upper computer, wherein the multi-range current measurement terminal, the load simulation terminal and the upper computer are shown in figure 4.
The current measuring terminal comprises a three-range current measuring circuit and a current signal acquisition and processing system. As shown in fig. 6, the three-range current measuring circuit includes a circuit for measuring current in small, medium, and large ranges at the same time, and is connected to the load simulation terminal, and is configured to supply power to the load simulation terminal and collect the simulation current signals generated by the load simulation terminal through the small, medium, and large circuits.
The embedded system software can run in different states of standby, data acquisition, communication and the like to generate working current in different ranges. By referring to the current ranges of the embedded software under different communication protocols when receiving and sending data and considering a certain margin, in the embodiment, the ranges corresponding to the small, medium and large measuring ranges are set to be [0,25mA ], (0,250mA ] and (0,2500mA ].
Fig. 5 shows a circuit for simultaneously measuring the current in small, medium and large three measuring ranges. In particular, in a wide-range measuring circuit, the amplifier 1 and the resistor R1Converting the current load of 0-2500mA into voltage of 0-2.5V; in a medium-range measuring circuit, an amplifier 2 and a resistor R2Converting the 0-250mA current load into 0-2.5V voltage; in a small-scale measuring circuit, an amplifier 3 and a resistor R2The 0-25mA current load is converted into a voltage of 0-2.5V. The output of each amplifier is connected to a corresponding a/D converter on the microcontroller for subsequent processing. To increase the voltage distributed by the load to be measured, the resistor R2Is measured andthe measuring circuit with small measuring range is shared. In addition, the voltage-stabilized power supply in FIG. 5 can provide a stable 0-10V selectable power supply for the load to be tested.
As shown in fig. 6, the current signal collecting and processing system includes three ADC modules, a TIM timing module, a DMA transmission module, and a USB communication module. In the embodiment, the current signal acquisition and processing system adopts a high-performance singlechip STM32F407 of Italian corporation as a microcontroller, and simultaneously acquires and converts current signals of small, medium and large ranges through 3 built-in 12-bit A/D converters. The resolution of each range is 7uA, 700uA and 7000uA respectively. In order to transmit the digital signals of the three paths of currents to a transceiver module of an upper computer in real time through a DMA channel and a USB2.0, the sampling frequency is set to be 1 MHz. The sampling frequency is higher than that of some existing resistance-based measuring methods, and the high-frequency change characteristic of the current during the operation of the embedded system software can be better met.
The working flow of the current signal acquisition and processing system is as follows: the TIM timing module generates an interrupt signal and simultaneously starts three ADC modules, the ADC modules are connected with a three-range current measuring circuit, each ADC module respectively converts three analog current signals acquired by small, medium and large ranges into digital current signals, then the DMA transmission module and the USB communication module are started, and the current signals of the three ranges are transmitted to the transceiver module of the upper computer.
The load simulation terminal is used for generating a resistance value with a set size through the variable resistor and generating simulation current signals with different sizes under the power supply voltage of the three-range current measuring circuit.
As shown in fig. 4, the upper computer includes a coordination module, a transceiver module, a filtering module, a storage module, a database, a sample generation module, and a machine learning module; the coordination module sends different current actual values to a load simulation terminal, the load simulation terminal calculates a resistance value to be generated according to the power supply voltage and the current actual values, controls a variable resistor to generate a corresponding resistance value through an internal electronic switch, and then outputs a simulation current signal under the power supply voltage; the receiving and transmitting module receives current signals of each range sent by the current measuring terminal, noise is filtered by the filtering module, current data of small, medium and large ranges are stored in the database through the storage module, and the sample generating module forms an initial calibration data set according to the actual current value sent by the cooperation module and the current data of the three ranges corresponding to the actual current value; the machine learning module builds and trains a calibration model based on the initial calibration data set.
The generation method of the initial calibration data set comprises the following steps:
the coordination module sends a load definition (y, t) comprising an actual current value y and a duration time t to a load simulation terminal, the load simulation terminal generates a simulation current signal corresponding to the actual current value y within the duration time t, the current measurement terminal respectively carries out current measurement through circuits with small, medium and large ranges within the duration time t, a group of readings are respectively obtained in each range, the readings are sent to an upper computer after being processed by a current signal acquisition and processing system, and the readings corresponding to each range are stored in a database after being processed by the upper computer; the sample generation module respectively averages the readings of each group to obtain readings x corresponding to small, medium and large measuring ranges1、x2And x3Further, a sample point data (x) is generated1,x2,x3,y);
The coordination module respectively generates a set number of current actual values in the small, medium and large measuring range, and the method is repeated to construct the data sets DS with the small, medium and large measuring ranges1、DS2And DS3。
Input variable x of initial calibration data set1、x2And x3The readings of the small measuring range, the medium measuring range and the large measuring range are respectively expressed, and the output variable y represents the actual value of the current. The number of samples for machine learning should not be less than(nvar is the number of input variables), therefore, in this embodiment, co-acquisitionAnd (4) sampling. Consider thatSample was measured over the entire measurement range 0,2500mA]Uniformly distributed in the [0,25mA ]]、(25,250mA]And (250,2500 mA)]In each case 60 samples are taken, forming three data sets DS1、DS2And DS3. From DS1∪DS2∪DS3An initial calibration data set is constructed and table 1 illustrates an empty initial calibration data set. Table 1 one sample per row.
TABLE 1 empty initial calibration data set
In order to generate the initial calibration data set, the actual current values of the last column in each row in table 1 are first generated; then defining 180 loads needing simulation according to the actual current value y of each row and with t (10 times of sampling frequency of a measuring terminal) as duration; then, for a load definition (y, t), the load generation, current measurement and filtering process of the calibrated system can obtain three range readings x corresponding to the small, medium and large ranges of y1、x2And x3So that each row in table 1 can be determined.
The algorithm of table 2 can generate the actual values of the current for each of the three intervals of table 1. The algorithm of table 2 takes into account the resolution, number of samples, and factors such as being as random and uniform as possible within the sampling range.
TABLE 2 sampling algorithm for actual values of sample currents
Based on the 180 simulation load definitions given above, the workflow of the multi-range current measurement calibration system of the present invention includes three steps of current load generation, initial calibration data set reconstruction, and calibration model construction.
1. Current load generation
As shown in fig. 4, the load simulation terminal is supplied by a current measurement terminal, which implements an electronic switch-controlled variable resistor and is responsible for generating the current load. Specifically, after the load simulation terminal receives load definitions such as current size and duration sent by the upper computer coordination module, resistance is calculated according to power supply voltage and current size required to be generated, then a specified resistance is generated by controlling an internal electronic switch, and a start-stop signal is sent to the current measurement terminal through a serial port according to the duration.
2. Initial calibration data set reconstruction
The initial calibration data set reconstruction process is illustrated below with the mth row example of table 1. The load simulation terminal generates a current load with the value y in the mth row of table 1 for the duration t. And the upper computer stores three groups of readings with different measuring ranges sampled and obtained in the time t in a database. The sample generation module first performs median filtering on each group of data, and then averages each group of readings to obtain (x)1,x2,x3) And fills in the corresponding column in row m of table 1. Similarly, the load simulation terminal sequentially generates 180 current loads corresponding to table 1, i.e. an initial calibration data set for machine learning can be reconstructed.
3. Calibration model construction
Establishing and outputting a calibration model f based on an initial data set through two-stage machine learning*(x1,x2,x3). The construction of the calibration model will be explained in detail below.
FIG. 7 is a multi-range current measurement calibration process based on machine learning, namely a process of constructing a calibration model by a machine learning module. The process mainly comprises two stages, wherein the first stage comprises two steps of generating each range data set and constructing each range calibration model.
1) Generating range datasets
TABLE 1 three intervals [0,25mA]、(25,250mA]And (250,2500 mA)]The 60 samples collected form three data sets DS1、DS2And DS3。
From a three-range data set DS1、DS2And DS3Randomly extracting a set number of samples (10% in the present embodiment) as test samples to form a corresponding test setAndfurther generating training data sets for small, medium and large three-range calibrationAndand testing the data setAndtraining data setAndas shown in equation (2), test data setAndis as defined in formula (3):
2) constructing calibration model of each measuring range
The linear and nonlinear noise of each range measurement is reduced by selecting a proper machine learning algorithm. Respectively constructing based on the selected machine learning algorithmOptimal calibration model for small, medium and large measuring rangesAnd
the second stage learns and synthesizes the calibration model on the basis of the optimal calibration model of each range, namely the reading x passing through three ranges1,x2,x3And optimal calibration model of each rangeAndobtaining an optimal synthetic model to output a current indication with a minimum errorThe second phase includes two steps of generating a synthetic dataset and constructing a synthetic calibration model.
1) Generating a synthetic data set
Optimal calibration model according to each measuring rangeAndand training data setAndgenerating a synthesized training set DSTrThe specific definition is shown as a formula (4); optimal calibration model according to each measuring rangeAndand testing the data setAndgenerating a synthesized test set DSTsSpecifically defined as shown in formula (5):
in the formulae (4) and (5), DSTrAnd DSTsFor each sample point in (x)1,x2,x3,a1,a2,a3Y) represents wherein x1、x2、x3Three input variables correspond to readings of three ranges, a1、a2、a3Three input variables are respectively composed of Andcalculating to obtain a variable y which is an actual current value;
2) building a synthetic calibration model
From the resultant data set DSTrAnd DSTsSelecting proper machine learning algorithm to determine the linear or nonlinear relation between the measuring point in the range transition neighborhood and the optimal calibration model of each range, and further constructing an optimal synthetic calibration model to ensure that the current value output by the synthetic calibration model is mistaken for the actual current valueThe difference is minimal.
The Average Relative Error (ARE) defined by formula (6) is used as a performance evaluation index for constructing the optimal calibration model and the optimal synthetic calibration model of each range:
wherein, yiThe value of the output variable, i.e. the current actual value,the method comprises the steps that a predicted value of current obtained according to input of the ith sample in a test set is shown, n represents the size of the sample of the test set, and the smaller the value of ARE is, the better the calibration model is;
in this embodiment, the machine learning algorithm used to construct the optimal calibration model and the optimal synthetic calibration model for each range is linear regression L R or support vector regression SVR, the main reason is that the measurement of the dynamic current contains both linear and nonlinear measurement errors, and this result holds for the measurement of each range.
Through experimental research, the average relative error (1.42%, 0.44% and 0.97%) of small, medium and large ranges and the average relative error (0.71%, 0.12% and 0.26%) in the transition neighborhood range of each range are superior to the range switching method when the calibration model obtained by the system calibration is used for current measurement.
The above are preferred embodiments of the present invention, and all changes made according to the technical scheme of the present invention that produce functional effects do not exceed the scope of the technical scheme of the present invention belong to the protection scope of the present invention.
Claims (5)
1. A multi-range current measurement calibration system based on machine learning is characterized by comprising a multi-range current measurement terminal, a load simulation terminal and an upper computer;
the current measuring terminal comprises a three-range current measuring circuit and a current signal acquisition and processing system, the three-range current measuring circuit comprises a circuit for measuring current in small, medium and large ranges at the same time, the three-range current measuring circuit is connected with a load simulation terminal, used for supplying power to the load simulation terminal and simultaneously collecting the simulation current signals generated by the load simulation terminal through circuits with small, medium and large measuring ranges, the current signal acquisition and processing system comprises three ADC modules, a timing module, a DMA transmission module and a communication module, the ADC module is connected with a three-range current measuring circuit, the timing module generates an interrupt signal and simultaneously starts the three ADC modules, and each ADC module respectively converts three analog current signals acquired by the small, medium and large ranges into digital current signals and then sends the digital current signals to the upper computer through the DMA transmission module and the communication module;
the load simulation terminal is used for generating a resistance value with a set size through the variable resistor and generating simulation current signals with different sizes under the power supply voltage of the three-range current measuring circuit;
the upper computer comprises a coordination module, a transceiving module, a filtering module, a storage module, a database, a sample generation module and a machine learning module; the coordination module sends different current actual values to a load simulation terminal, the load simulation terminal calculates a resistance value to be generated according to the power supply voltage and the current actual values, controls the variable resistor to generate a corresponding resistance value, and then outputs a simulation current signal under the power supply voltage; the receiving and transmitting module receives current signals of each range sent by the current measuring terminal, noise is filtered by the filtering module, current data of small, medium and large ranges are stored in the database through the storage module, and the sample generating module forms an initial calibration data set according to the actual current value sent by the cooperation module and the current data of the three ranges corresponding to the actual current value; the machine learning module builds and trains a calibration model based on the initial calibration data set.
2. The machine-learning based multi-range current measurement calibration system of claim 1 wherein the small, medium and large three measurement ranges are [0,25mA ], (25mA,250mA ], and (250mA,2500mA ].
3. The machine-learning based multi-range current measurement calibration system of claim 1, wherein the initial calibration data set is generated by:
the coordination module sends a load definition (y, t) comprising an actual current value y and a duration time t to a load simulation terminal, the load simulation terminal generates a simulation current signal corresponding to the actual current value y within the duration time t, the current measurement terminal respectively carries out current measurement through circuits with small, medium and large ranges within the duration time t, a group of readings are respectively obtained in each range, the readings are sent to an upper computer after being processed by a current signal acquisition and processing system, and the readings corresponding to each range are stored in a database after being processed by the upper computer; the sample generation module respectively averages the readings of each group to obtain readings x corresponding to small, medium and large measuring ranges1、x2And x3Further, a sample point data (x) is generated1,x2,x3,y);
The coordination module respectively generates a set number of current actual values in the small, medium and large measuring range, and the method is repeated to construct the data sets DS with the small, medium and large measuring ranges1、DS2And DS3。
4. The machine-learning based multi-range current measurement calibration system of claim 3, wherein the machine learning module constructs the calibration model by:
1) from a three-range data set DS1、DS2And DS3Randomly extracting a set number of samples as test samples to form a corresponding test setAndfurther generating training data sets for small, medium and large three-range calibrationAndand testing the data setAndtraining data setAndas shown in equation (2), test data setAndis as defined in formula (3):
2) constructing a calibration model of each measuring range: based on machine learning algorithm, constructs separatelyOptimum calibration model for establishing small, medium and large measuring rangesAnd
3) generating a synthetic data set: optimal calibration model according to each measuring rangeAndand training data set Andgenerating a synthesized training set DSTrThe specific definition is shown as a formula (4); optimal calibration model according to each measuring rangeAndand testing the data setAndgenerating a synthesized test set DSTsSpecifically defined as shown in formula (5):
in the formulae (4) and (5), DSTrAnd DSTsFor each sample point in (x)1,x2,x3,a1,a2,a3Y) represents wherein x1、x2、x3Three input variables correspond to readings of three ranges, a1、a2、a3Three input variables are respectively composed of Andcalculating to obtain a variable y which is an actual current value;
4) constructing a synthetic calibration model: from the resultant data set DSTrAnd DSTsBased on a machine learning algorithm, a linear or nonlinear relation between the measuring point in the range transition neighborhood and the optimal calibration model of each range is determined, and then an optimal synthetic calibration model is constructed, so that the error between the current value output by the synthetic calibration model and the current actual value is minimum.
5. The machine-learning-based multi-range current measurement calibration system of claim 4, wherein the optimal calibration model and the optimal synthetic calibration model for each range ARE constructed using the average relative error ARE defined by equation (6) as a performance evaluation index:
wherein, yiThe value of the output variable, i.e. the current actual value,the method comprises the steps that a predicted value of current obtained according to input of the ith sample in a test set is shown, n represents the size of the sample of the test set, and the smaller the value of ARE is, the better the calibration model is;
the machine learning algorithm adopted for constructing the optimal calibration model and the optimal synthetic calibration model of each range is linear regression L R or support vector regression SVR.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010346676.1A CN111458672B (en) | 2020-04-28 | 2020-04-28 | Multi-range current measurement calibration system based on machine learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010346676.1A CN111458672B (en) | 2020-04-28 | 2020-04-28 | Multi-range current measurement calibration system based on machine learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111458672A true CN111458672A (en) | 2020-07-28 |
CN111458672B CN111458672B (en) | 2022-08-05 |
Family
ID=71682647
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010346676.1A Active CN111458672B (en) | 2020-04-28 | 2020-04-28 | Multi-range current measurement calibration system based on machine learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111458672B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113433501A (en) * | 2021-06-24 | 2021-09-24 | 紫光展讯通信(惠州)有限公司 | Current calibration method, system, medium and calibration board |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130088213A1 (en) * | 2011-10-05 | 2013-04-11 | Veris Industries, Llc | Current switch with automatic calibration |
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 |
CN103926546A (en) * | 2013-01-16 | 2014-07-16 | 湖北迅迪科技有限公司 | Portable DC large-current field calibrator |
CN104535954A (en) * | 2015-01-12 | 2015-04-22 | 长沙天恒测控技术有限公司 | Calibration device of active leakage current tester |
CN106249038A (en) * | 2016-07-19 | 2016-12-21 | 北京机械设备研究所 | A kind of General portable far-end high-power rudder electromechanics pressure, current measurement and calibrating installation |
-
2020
- 2020-04-28 CN CN202010346676.1A patent/CN111458672B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130088213A1 (en) * | 2011-10-05 | 2013-04-11 | Veris Industries, Llc | Current switch with automatic calibration |
CN103926546A (en) * | 2013-01-16 | 2014-07-16 | 湖北迅迪科技有限公司 | Portable DC large-current field calibrator |
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 |
CN104535954A (en) * | 2015-01-12 | 2015-04-22 | 长沙天恒测控技术有限公司 | Calibration device of active leakage current tester |
CN106249038A (en) * | 2016-07-19 | 2016-12-21 | 北京机械设备研究所 | A kind of General portable far-end high-power rudder electromechanics pressure, current measurement and calibrating installation |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113433501A (en) * | 2021-06-24 | 2021-09-24 | 紫光展讯通信(惠州)有限公司 | Current calibration method, system, medium and calibration board |
Also Published As
Publication number | Publication date |
---|---|
CN111458672B (en) | 2022-08-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106324538B (en) | A kind of shelf depreciation automated calibration system | |
CN102375106A (en) | Device for testing harmonic influence of electronic mutual inductor | |
CN111458672B (en) | Multi-range current measurement calibration system based on machine learning | |
CN110672996A (en) | Urban cable multidimensional data integrated live detection device | |
CN110542793A (en) | passive device intelligent equivalent circuit model, parameter measuring device and working method | |
CN113395189A (en) | Vehicle-mounted Ethernet SQI signal quality testing method and system | |
CN111308232B (en) | System and method for measuring stray parameters of current loop of high-power current conversion module | |
CN207366731U (en) | A kind of intelligent electric energy meter detection device | |
CN202093095U (en) | Direct-current resistance tester for transformer | |
CN206292324U (en) | A kind of frequency-characteristic measuring-testing instrument | |
CN113884818A (en) | LSTM-based power distribution network fault traveling wave arrival time accurate estimation method | |
CN111398726B (en) | Method, system and equipment for testing charging and discharging response time of energy storage frequency modulation system | |
CN113189513A (en) | Ripple-based redundant power supply current sharing state identification method | |
CN111488266A (en) | Embedded software energy consumption measuring system based on multiple ranges | |
CN110703117A (en) | Portable storage battery performance testing device | |
CN203838301U (en) | Automatic testing device for pre-amplifier | |
CN207366633U (en) | plasma probe measurement and control system | |
CN108982967A (en) | A kind of network of ship parameter real time monitoring system and method | |
CN209841927U (en) | Meter source integrated synchronous dynamic distortion power source with trigger | |
CN113702676B (en) | Broadband arbitrary waveform generation device and method based on CML interface | |
CN214201716U (en) | Online vehicle-mounted hydrogen fuel cell impedance spectrum measuring device | |
CN101915864B (en) | Vector oscilloscope device | |
CN102495335B (en) | Sensitivity detection method of travelling wave distance measuring device and system thereof | |
CN110568233A (en) | meter-source integrated synchronous dynamic distortion power source with trigger and implementation method | |
CN201724764U (en) | Multi-channel ground stress data collection instrument |
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 |