CN119783060A - A method for predicting errors of digital temperature and humidity instruments in workshops based on machine learning - Google Patents

Abstract

The invention discloses a machine learning-based workshop temperature and humidity digital instrument error prediction method, and belongs to the technical field of industrial metering. The method accurately predicts the measurement error of the measured table by introducing data smoothing, peak value extraction, filtering and outlier processing and establishing a multiple linear regression model, dynamically adjusts the metering calibration period according to the prediction error, ensures the accuracy of the measurement data and realizes error prediction. The method effectively solves the metering problem caused by small data quantity, unaligned data time sequence, poor data continuity, unclear performance index and poor abnormal response on site, combines a time sequence and a self-adaptive adjustment algorithm, and improves the reliability of model prediction. The technical method can reduce the human error of on-site measurement, the algorithm responds in real time, the labor is saved, the construction of a digital workshop is promoted, and the mathematical foundation support of metering is ensured.

Description

Machine learning-based workshop temperature and humidity digital instrument error prediction method

Technical Field

The invention belongs to the technical field of industrial metering, and particularly relates to a workshop temperature and humidity digital instrument error prediction method based on machine learning.

Background

In the production process of a production type enterprise workshop, temperature and humidity are key factors influencing product quality and production efficiency. At present, temperature and humidity monitoring in workshops mainly depends on traditional digital instruments for measurement, but the instruments have the problems of insufficient precision, slow response speed, incapability of real-time early warning and the like, and the change of workshop environments is difficult to reflect in time, so that the stability of product quality is affected. Therefore, the temperature and humidity digital instrument needs to be measured, and the problems of small data quantity, misalignment of data time sequence, poor data continuity, unclear performance index and poor abnormal response exist in the measurement process of the digital instrument.

Therefore, the data of the instrument are aligned and analyzed by adopting a proper algorithm, which is very important for metering of the instrument.

Disclosure of Invention

In order to solve the defects in the prior art, the invention provides a workshop temperature and humidity digital instrument error prediction method based on machine learning, which is realized by introducing an advanced machine learning algorithm, and intelligent analysis and error prediction are carried out on the temperature and humidity digital instrument data, so that the precision and response speed of temperature and humidity measurement are improved, and real-time monitoring and adjustment of the temperature and humidity digital instrument are realized. .

In order to achieve the above purpose, the present invention adopts the following technical scheme:

A workshop temperature and humidity digital instrument error prediction method based on machine learning adopts a temperature and humidity sensor, and comprises the following steps:

Step1, acquiring temperature and humidity data through a temperature and humidity sensor, and storing the acquired temperature and humidity data in a local database;

Step 2, preprocessing the acquired temperature and humidity data and aligning the data;

step 3, carrying out error prediction on the acquired temperature and humidity data;

and 4, adjusting and determining the metering period.

Preferably, in step 2, the method specifically comprises the following steps:

step 2.1, eliminating random noise in the original data by using a moving average method or an exponential smoothing method to enable the data to be smoother, wherein the calculation of the moving average method is shown as a formula (1):

Where k is the window radius, t is the time index, Is the smoothed data of the sensor at time t, D _sensor (t+i) is the smoothed data of the sensor at time t+i, and i is the cyclic variable.

Step 2.2, extracting peak value, and obtaining local maximum value by using local extremum detection, wherein the local maximum value is shown in a formula (2):

D_t＝X_t-X_t-1(2);

X _t is a local maximum when D _t-1 >0 and D _t is less than or equal to 0;

step 2.3, filtering and outlier processing;

Step 2.4, aligning time series mean values;

Data of different time sequences are aligned using dynamic time warping.

Preferably, in step 2.3, the method specifically comprises the following steps:

step 2.3.1, filtering treatment;

and filtering the smoothed data by using a low-pass filter, wherein the transfer function of the filter is as follows:

Wherein τ is a time constant and s is a Laplacian transform variable;

step 2.3.2, processing abnormal values;

and (3) cleaning the abnormal value of the acquired data by adopting a 3 sigma method, and if a certain data point meets the following conditions:

If it is ThenIs an outlier;

where μ is the mean of the data and σ is the standard deviation.

Preferably, in step 3, the method specifically includes the following steps:

step 3.1, resampling the data, unifying the time scale of the data, and processing the data into time sequence data;

resampling includes downsampling and upsampling, wherein,

Downsampling, namely selecting data according to a fixed step length s when the data sampling frequency is too high:

X^′＝{X₁,X_1+s,X_1+2s,...,X_n}

upsampling, namely increasing data points by using an interpolation method when the data sampling frequency is too low;

Step 3.2, processing the missing value in the time sequence data;

the missing value is processed by a linear interpolation method, and the linear interpolation calculation formula is as follows:

Where (x ₁,y₁),(x₂,y₂) is the coordinates of two known points, (x, y) is the coordinates of the point to be interpolated, where x is the x coordinate value to be interpolated;

Step 3.3, constructing a multiple linear regression prediction model, and constructing a relation model between the metering error and various influencing factors;

Model form:

wherein β ₀ is the intercept, β ₁ is the regression coefficient, ε t) is the error term; In order for the data set of the table to be tested, Is a standard table dataset;

The decision coefficient R ² of the regression model is:

Wherein, The average value of the measured table data is used for evaluating the fitting degree of the model;

the actual observed value at time t _i, i.e. the output value of the sensor under test at that time;

A standard or predicted value at time t _i;

β ₀, an intercept term representing the offset of the regression model, i.e., the predicted value of the dependent variable when the independent variable is zero;

beta ₁ slope, the strength and direction of the linear relationship between the independent and dependent variables;

m is the total number of samples.

Preferably, in step 4, the method specifically includes the following steps:

Step 4.1, predicting future metering errors according to a multiple linear regression model;

And 4.2, dynamically adjusting the calibration period by using an adaptive adjustment algorithm according to the predicted metering error.

The invention has the beneficial technical effects that:

The invention provides a workshop temperature and humidity digital instrument error prediction method based on machine learning, which effectively solves the metering problems caused by small data quantity, unaligned data time sequence, poor data continuity, unclear performance index and poor abnormal response on site, combines a time sequence and a self-adaptive adjustment algorithm, and improves the reliability of model prediction. The technical method can reduce the human error of on-site measurement, the algorithm responds in real time, the labor is saved, the construction of a digital workshop is promoted, and the mathematical foundation support of metering is ensured.

Drawings

FIG. 1 is a flow chart of the algorithm of the method of the present invention.

FIG. 2 is a schematic diagram of the error prediction result of the method of the present invention.

Detailed Description

The invention is described in further detail below with reference to the attached drawings and detailed description:

In the embodiment, temperature and humidity sensor data of a workshop are selected as a measured meter data set D _sensor, and customized high-precision temperature and humidity probe data are selected as a standard data set D _standard. The two sets of data were subjected to a comparative analysis of the metrology data by the following procedure.

The machine learning-based workshop temperature and humidity digital instrument error prediction method has a flow shown in fig. 1 and comprises the following steps:

and step 1, acquiring temperature and humidity data in a workshop, namely, additionally arranging a high-precision temperature and humidity sensor near the original temperature and humidity instrument to acquire the data, and storing the acquired temperature and humidity data in a local database.

And 2, carrying out data preprocessing and data alignment on the acquired temperature and humidity data. The method specifically comprises the following steps:

And 2.1, eliminating random noise in the original data by using a moving average method or an exponential smoothing method, so that the data is smoother, and the subsequent analysis is convenient. The moving average method has the following calculation formula:

Where k is the window radius, t is the time index, Is the smoothed data of the sensor at time t.

And 2.2, extracting peak values, and obtaining local maximum values by using local extremum detection. The formula is as follows:

D_t＝X_t-X_t-1

X _t is a local maximum when D _t-1 >0 and D _t is less than or equal to 0.

Step 2.3, filtering and outlier processing, namely eliminating high-frequency noise and outlier data points and improving data quality, wherein the method specifically comprises the following steps of:

step 2.3.1 Filter treatment

And filtering the smoothed data by using a low-pass filter, wherein the transfer function of the filter is as follows:

where τ is the time constant and s is the Laplace transform variable.

Taking a Kalman filter (KALMAN FILTER) as an example:

a) State prediction:

b) Error covariance prediction:

P_t|t-1＝AP_t-1|t-1A^T+Q

c) Kalman gain calculation:

K_t＝P_t|t-1H^T(HP_t|t-1H^T+R)^-1

d) And (5) updating the state:

e) Error covariance update:

P_t|t＝(I-K_tH)P_t|t-1

Wherein A is a state transition matrix, B is a control matrix, u _t-1 is a control vector, Q is a process noise covariance, H is an observation matrix, R is a measurement noise covariance, and Z _t is an observation value.

Step 2.3.2, outlier processing:

and (3) cleaning the abnormal value of the acquired data by adopting a 3 sigma method, and if a certain data point meets the following conditions:

If it is ThenIs an outlier

Where μ is the mean of the data and σ is the standard deviation.

Step 2.4 time series mean alignment

Data of different time sequences are aligned for comparison and analysis using dynamic time warping.

The measured table dataset and the standard dataset are aligned on a time axis and unified to be the same time stamp T= { T ₁,t₂,...,t_n }.

Calculating the average value of the corresponding time points to obtain an aligned time sequence:

where N is the number of valid data at time point t _i.

And 3, carrying out error prediction on the acquired data. The method specifically comprises the following steps:

And 3.1, resampling the data, unifying the time scale of the data, eliminating the influence caused by different sampling frequencies, and processing the data into time sequence data.

Resampling:

Downsampling, namely selecting data according to a fixed step length s when the data sampling frequency is too high:

X^′＝{X₁,X_1+s,X_1+2s,...,X_n}

Upsampling, when the data sampling frequency is too low, the data points are added using interpolation methods, such as linear interpolation or spline interpolation.

And 3.2, processing the missing value in the time sequence data.

And (3) applying a linear interpolation method to the acquired temperature and humidity data, and processing the missing value. The linear interpolation calculation method is as follows.

Where (x ₁,y₁),(x₂,y₂) is the coordinates of two known points and (x, y) is the coordinates of the point to be interpolated, where x is the x coordinate value to be interpolated.

And 3.3, constructing a multiple linear regression prediction model, and constructing a relation model between the metering error and various influencing factors.

Model form:

Wherein β ₀ is the intercept, β ₁ is the regression coefficient, and e (t) is the error term; In order for the data set of the table to be tested, Is a standard table dataset.

The decision coefficient R ² of the regression model is:

Wherein, The average value of the measured table data is used for evaluating the fitting degree of the model;

For this is the actual observation at time t _i, i.e. the output value of the sensor under test at that time;

This is the standard or predicted value at time t _i;

Beta ₀ is an intercept term representing the offset of the regression model, i.e., the predicted value of the dependent variable when the independent variable is zero;

beta ₁ is the slope, which represents the strength and direction of the linear relationship between the independent and dependent variables;

m is the total number of samples.

And carrying out 10 tests, and carrying out model evaluation on the error prediction model to obtain an average R ² =0.98 of the model, wherein the model fitting effect is good. The error prediction result is shown in fig. 2.

And 4, metering cycle adjustment and determination. The method specifically comprises the following steps:

And 4.1, predicting future metering errors according to the multiple linear regression model.

Substituting the influence factors of the future time into a regression model:

Wherein,

H is the prediction step size.

Predicted metrology error at future time t+h;

Intercept term in regression model;

the degree of influence of independent variable pressure, temperature and humidity on metering errors;

P _t+h,T_t+h,H_t+h pressure, temperature, humidity values at future time t+h.

And 4.2, dynamically adjusting the calibration period by using an adaptive adjustment algorithm according to the predicted metering error, so as to ensure the metering accuracy.

Adjustment principle:

(1) An allowable maximum error threshold E _max is defined.

(2) When (when)Extending the calibration period T;

(3) When (when) At this time, the calibration period T is shortened.

And (3) adjusting a formula:

wherein the function is adjusted The definition is as follows:

k is an adjusting coefficient which is 0<k is less than or equal to 1;

t _new a new calibration period;

t _current current calibration period;

And (3) an adjustment function, wherein the adjustment factor of the new calibration period is calculated according to the predicted metering error.

Predicted metering errors or actual metering errors;

e _max maximum error threshold.

In the embodiment, the technical method provided by the invention is adopted to accurately predict the measurement error of the measured table by introducing data smoothing, peak value extraction, filtering and outlier processing and establishing a multiple linear regression model. The metering calibration period is dynamically adjusted according to the prediction error, so that the accuracy of measured data is ensured, and error prediction is realized.

According to the method, the machine learning algorithm is introduced to conduct intelligent analysis and error prediction on the temperature and humidity digital instrument data, the precision and response speed of temperature and humidity measurement are improved, the error analysis and prediction on the temperature and humidity digital instrument are achieved, and the problems of small data quantity, misalignment of data time sequence, poor data continuity, unclear performance indexes and poor abnormal response are solved. The invention solves the defects in the prior art, and therefore, the invention has value of being applied to remote measurement of temperature and humidity digital instrument data.

It should be understood that the above description is not intended to limit the invention to the particular embodiments disclosed, but to limit the invention to the particular embodiments disclosed, and that the invention is not limited to the particular embodiments disclosed, but is intended to cover modifications, adaptations, additions and alternatives falling within the spirit and scope of the invention.

Claims (5)

1. A method for predicting errors of digital temperature and humidity instruments in workshops based on machine learning, characterized in that a temperature and humidity sensor is used, and the method comprises the following steps: Step 1: Collect temperature and humidity data through the temperature and humidity sensor, and store the collected temperature and humidity data in the local database; Step 2: Preprocess and align the collected temperature and humidity data; Step 3: Predict the error of the collected temperature and humidity data; Step 4: Adjustment and determination of measurement cycle. 2. The method for predicting the error of a digital temperature and humidity meter based on machine learning according to claim 1 is characterized in that: in step 2, the method specifically comprises the following steps: Step 2.1: Use the moving average method or exponential smoothing method to eliminate random noise in the original data and make the data smoother; the moving average method calculation is shown in formula (1): Where k is the window radius, t is the time index, is the smoothed data of the sensor at time t; D _sensor (t+i) is the data of the sensor at time t+i; i is the loop variable; Step 2.2: Peak extraction, using local extreme value detection to obtain the local maximum value; as shown in formula (2): D _t =X _t -X _t-1 (2); When D _t-1 >0 and D _t ≤0, X _t is a local maximum; Step 2.3: Filtering and outlier processing; Step 2.4: Time series mean alignment; Use dynamic time warping to align data from different time series. 3. The method for predicting the error of a workshop temperature and humidity digital instrument based on machine learning according to claim 2 is characterized in that: in step 2.3, the method specifically comprises the following steps: Step 2.3.1: Filtering; The smoothed data is filtered using a low-pass filter, and the filter transfer function is: Among them, τ is the time constant and s is the Laplace transform variable; Step 2.3.2: Outlier processing; The 3σ method is used to clean the collected data for outliers. If a data point satisfies: like but is an outlier; Among them, μ is the mean of the data and σ is the standard deviation. 4. The method for predicting the error of a workshop temperature and humidity digital instrument based on machine learning according to claim 1 is characterized in that: in step 3, the following steps are specifically included: Step 3.1: Resample the data, unify the time scale of the data, and process it into time series data; Resampling includes downsampling and upsampling; among them, Downsampling: When the data sampling frequency is too high, select data according to a fixed step size s: X ^′ ={X ₁ ,X _1+s ,X _1+2s ,...,X _n } Upsampling: When the data sampling frequency is too low, use interpolation to increase the data points; Step 3.2: Handle missing values in time series data; Missing values are processed by linear interpolation; the linear interpolation calculation formula is as follows: Among them, (x ₁ ,y ₁ ), (x ₂ ,y ₂ ) are the coordinates of two known points, (x,y) are the coordinates of the point to be interpolated, and x is the x-coordinate value to be interpolated; Step 3.3: Construct a multivariate linear regression prediction model to establish a relationship model between measurement error and multiple influencing factors; Model form: Among them, β ₀ is the intercept, β ₁ is the regression coefficient, ∈(t) is the error term; is the data set of the table under test, It is a standard table data set; The coefficient of determination ^R2 of the regression model is: in, The average value of the data in the tested table is used to evaluate the degree of model fit; The actual observed value at time _ti , that is, the output value of the measured sensor at that time; The standard value or predicted value at time _ti ; β ₀ : intercept term, which represents the offset of the regression model, that is, the predicted value of the dependent variable when the independent variable is zero; β ₁ : slope, which indicates the strength and direction of the linear relationship between the independent variable and the dependent variable; m: the total number of samples. 5. The method for predicting the error of a digital temperature and humidity meter in a workshop based on machine learning according to claim 1 is characterized in that: in step 4, the following steps are specifically included: Step 4.1: Predict future measurement errors based on the multivariate linear regression model; Step 4.2: Based on the predicted metrology error, the calibration period is dynamically adjusted using an adaptive adjustment algorithm.