Denoising method for heat flow signal measurement of refrigerating device box body

Abstract

The invention discloses a denoising method for heat flow signal measurement of a box body of a refrigerating device. The denoising method comprises the following steps: S1, obtaining a to-be-denoised steady-state heat flow signal of the refrigerating device; s2, constructing a digital filtering algorithm heat flow denoising model; s3, constructing a repeatability steady-state evaluation index and a fidelity steady-state evaluation index; s4, constructing a repeatability dynamic evaluation index and a fidelity dynamic evaluation index; s5, selecting an optimal filtering model and working parameters thereof; s6, determining the capacity of a heat flow signal sample; and S7, determining the sampling period of the heat flow signal sample. According to the method, a multi-dimensional quantitative evaluation index based on the repeatability and fidelity of the denoising effect is provided, the repeatability and the fidelity are considered, the optimal working parameters of the filters can be determined, and the denoising effect of each filter is compared by adopting the intersection region rule; and a filter with large denoising amount and relatively high signal characteristic fidelity level and working parameters of the filter are easy to select.

Description

Technical Field

[0001] The invention relates to the technical field of heat leakage load testing of refrigeration devices, and in particular to a denoising method for measuring heat flow signals of a refrigeration device box. Background Art

[0002] In engineering, the reverse heat balance method is usually used to measure the heat leakage load of the refrigeration device. That is, in a constant temperature environment, the compressor of the refrigeration device is turned off, and an electric heater is built into the refrigeration box. The power of the electric heater is adjusted to make the temperature inside the refrigeration box higher than the temperature of the external environment. The reverse experimental system uses the temperature balance inside and outside the box as the basis for thermal steady-state judgment. It has low construction cost and easy operation, and has a wide range of applications. However, the temperature balance-thermal steady-state method is only suitable for occasions with large heat flux density values. Due to the large thermal resistance of the refrigeration box wall and the small heat flux density value (20W / m 2 When the measurement noise is caused by the insufficient precision of the measuring instrument or the unstable flow field on both sides of the wall, it will interfere with the measurement accuracy of the heat flux signal.

[0003] In response to this type of interference, digital filters are usually used to denoise the collected signals. There are many types of digital filters. For a set of time series noisy signals, there is no systematic discussion on how to select the appropriate type of filter, determine the optimal working parameters of the filter, and the optimal sampling parameters of the sample data. Quantitative evaluation of the denoising effect of heat flow signals is a relatively important task. The root mean square error of the exact solution and the estimated value is the most common evaluation indicator. This evaluation method only gives the denoising results and denoising effects under different conditions. It is difficult to determine the optimal working parameters of the filter based on the existing evaluation indicators. Summary of the invention

[0004] The purpose of the present invention is to solve the above-mentioned background technical problems and to provide a denoising method for measuring heat flow signals of a refrigeration device cabinet.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A denoising method for measuring heat flow signals of a refrigeration device cabinet comprises:

[0007] S1. After the air temperature inside the refrigeration box and the air temperature outside the refrigeration box are stable, the heat flux density sensor is posted on the wall of the refrigeration box, a DC voltage signal is output, and the heat flux density value is calculated;

[0008] S2. Constructing a digital filtering algorithm heat flow denoising model, wherein the digital filtering algorithm heat flow denoising model includes a rolling average model, a moving average model, a median model, a clipping model, an SG model, a Gaussian model and a Kalman model;

[0009] S3. constructing a repeatability steady-state evaluation index and a fidelity steady-state evaluation index, wherein the repeatability steady-state evaluation index includes a standard deviation and a mean absolute deviation, and the fidelity steady-state evaluation index includes a signal-to-noise ratio and a correlation coefficient;

[0010] S4. Constructing a dynamic evaluation index of repeatability and a dynamic evaluation index of fidelity, wherein the dynamic evaluation index of repeatability includes a running standard deviation and a running mean absolute deviation, and the dynamic evaluation index of fidelity includes a running signal-to-noise ratio, a running correlation coefficient, and a running coefficient of variation;

[0011] S5. Set the window width range, limit value range, polynomial highest order range, distribution parameter range, process noise covariance range and measurement noise covariance range of each model, calculate the heat flux value and steady-state evaluation index after denoising, and use the intersection area rule to determine the local optimal working parameters;

[0012] S6. Under the premise that the sample starting point is fixed, the sample end point is rolled forward, and the running signal-to-noise ratio, running standard deviation, running mean absolute deviation, running correlation coefficient and four running coefficients of variation are calculated. The running coefficient of variation curves under different sample end points are drawn. The horizontal axis value corresponding to the last intersection point of each curve with the coefficient of variation of 2% is regarded as the sample-independent capacity;

[0013] S7, adjust the sampling period of the heat flux signal, calculate the heat flux density and dynamic operation index after denoising, repeat step S6, obtain the sample independence capacity under different sampling periods, and calculate the sample independence duration under different sampling periods.

[0014] As a further solution of the present invention: in step S1, the operating standard deviation is used to determine the stability of the air temperature inside the refrigeration box and the air temperature outside.

[0015] As a further solution of the present invention: the air temperature inside the refrigeration box is recorded as T in , the external air temperature of the refrigeration box is recorded as T out , the running standard deviation is determined as follows: When both are less than 0.08, it indicates that the temperature inside and outside the refrigeration box has reached a stable state.

[0016] As a further solution of the present invention: in step S2, the heat flow denoising model of the digital filtering algorithm is as follows:

[0017] The rolling average model is:

[0018]

[0019] The moving average model is:

[0020] v = (n-1) / 2;

[0021] The median model is:

[0022] q fk =median n {q k-v ,...,q k-1 ,q k ,q k+1 ,...,q k+v},(k=1,2,...v=(n-1) / 2);

[0023] The clipping model is:

[0024]

[0025] The SG model is:

[0026]

[0027] The Gaussian model is:

[0028]

[0029] The Kalman model is:

[0030]

[0031] and

[0032] As a further solution of the present invention: in the step S3, the calculation formulas of the repeatability steady-state evaluation index and the fidelity steady-state evaluation index are as follows:

[0033] The signal-to-noise ratio is:

[0034] The standard deviation is:

[0035] The mean absolute deviation is:

[0036] The correlation coefficient is:

[0037] As a further solution of the present invention: in the step S4, the calculation formulas of the repeatability dynamic evaluation index and the fidelity dynamic evaluation index are as follows:

[0038] The operating signal-to-noise ratio is:

[0039] The running standard deviation is:

[0040] The running mean absolute deviation is:

[0041] The operational correlation coefficient is:

[0042]

[0043] The running coefficient of variation is:

[0044] As a further solution of the present invention: in step S5, the window width is set in the range of 5, 11, 51, 101, 201, 401, 801, the limit value is set in the range of 2, 1.5, 1, 0.5, 0.2, 0.1, 0.05, the highest order of the polynomial is set in the range of 51, 31, 15, 9, 5, 3, 1, the distribution parameter is set in the range of 0.5, 3, 7, 11, 17, 21, 25, the process noise covariance is set in the range of 10 -3 , 10 -4 , 10 -5 , 10 - 6, 10 -7 , 10 -8 , 10 -9 , the measurement noise covariance range is set to 10 -7 , 10 -6 , 10 -5 , 10 -4 , 10 -3 , 10 -2 , 10 -1 .

[0045] As a further solution of the present invention: in step S5, the step of using the intersection area rule to determine the local optimal working parameters includes:

[0046] S51, locating the intersection working parameter point of the standard deviation, mean absolute deviation and correlation coefficient curves;

[0047] S52. Based on the principle of repeatability priority, the point with the smallest standard deviation in the intersection area is taken as the optimal working parameter point.

[0048] As a further solution of the present invention: in the step S6, the four operating coefficients of variation are respectively the operating signal-to-noise ratio coefficient of variation, the operating standard deviation coefficient of variation, the operating mean absolute deviation coefficient of variation and the operating correlation coefficient coefficient of variation, the operating coefficients of variation are four curves, the 2% coefficient of variation is a horizontal straight line, and when drawn on the same coordinate axis, the four curves and the horizontal straight line respectively have multiple intersection points, and within the sample capacity range, the horizontal coordinate value of the last intersection point, that is, the maximum horizontal coordinate value, is the sample independence capacity.

[0049] Beneficial effects of the present invention:

[0050] (1) In the present invention, a multi-dimensional quantitative evaluation index based on the repeatability and fidelity of the denoising effect is proposed, which takes both repeatability and fidelity into consideration and can determine the optimal working parameters of the filter.

[0051] (2) In the present invention, the intersection area rule is used to compare the denoising effects of various filters, which makes it easy to select filters and their operating parameters with large denoising amount but high signal feature fidelity level. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The present invention will be further described below in conjunction with the accompanying drawings.

[0053] Figure 1 It is a flow chart of a denoising method for measuring heat flow signals of a refrigeration device box according to the present invention;

[0054] Figure 2 This is a diagram showing the trend of the air temperature change inside and outside the box and the stability analysis in the experiment of the present invention;

[0055] Figure 3 It is a graph of the change trend and stability analysis of the time series heat flow signal to be denoised in the experiment of the present invention;

[0056] Figure 4 Schematic diagram of the time series denoising curve of the median filter under different WWs in the present invention;

[0057] Figure 5 is a schematic diagram of a time series denoising curve of a moving average filter under different WWs in the present invention;

[0058] Figure 6 It is a schematic diagram of the timing denoising curve of the SG filter under different WWs in the present invention;

[0059] Figure 7 is a schematic diagram of a time series denoising curve of a Gaussian filter under different WWs in the present invention;

[0060] Figure 8 It is a schematic diagram of the variation trend of the evaluation index of the median filter according to the present invention with WW;

[0061] Fig. 9 It is a schematic diagram of the changing trend of the evaluation index of the moving average filter according to the present invention with WW;

[0062] Fig.10 It is a schematic diagram of the variation trend of the evaluation index of the SG filter according to the present invention with WW;

[0063] Fig.11 It is a schematic diagram of the variation trend of the evaluation index of the Gaussian filter with WW in the present invention;

[0064] Fig.12 is a schematic diagram of a timing denoising curve of the SG filter at different highest orders in the present invention;

[0065] Fig.13 It is a schematic diagram of the variation trend of the evaluation index of the SG filter according to the highest order in the present invention;

[0066] Fig.14 is a schematic diagram of a time series denoising curve of a limiting filter under different limiting values ​​in the present invention;

[0067] Fig.15 It is a schematic diagram of the variation trend of the evaluation index of the limiting filter according to the limit value in the present invention;

[0068] Fig.16 is a schematic diagram of a time series denoising curve of a Gaussian filter under different distribution parameters in the present invention;

[0069] Fig.17 It is a schematic diagram of the variation trend of the evaluation index of the Gaussian filter with the distribution parameter in the present invention;

[0070] Fig.18 is a schematic diagram of the Gaussian filter distribution parameter rule in the present invention;

[0071] Fig.19 is a schematic diagram of an actual denoising template of a Gaussian filter in the present invention;

[0072] Fig. 20 is a schematic diagram of a time series denoising curve of a Kalman filter under different process noise covariances in the present invention;

[0073] Fig.21 It is a schematic diagram of the variation trend of the evaluation index of the Kalman filter with the process noise covariance in the present invention;

[0074] Fig. 22 is a schematic diagram of a time series denoising curve of a Kalman filter under different measurement noise covariances in the present invention;

[0075] Fig.23It is a schematic diagram of the variation trend of the evaluation index of the Kalman filter according to the measurement noise covariance in the present invention;

[0076] Fig.24 It is a schematic diagram of comparing denoising effect evaluation indicators under different filtering algorithms in the present invention. DETAILED DESCRIPTION

[0077] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0078] See also Figure 1 As shown, the present invention is a denoising method for measuring heat flow signals of a refrigeration device cabinet, comprising:

[0079] S1. Obtain the steady-state heat flow signal to be denoised of the refrigeration device.

[0080] Specifically, the air temperature inside the refrigeration box is T in , outside air temperature T out After stabilization, the heat flux density sensor is posted on the wall of the refrigeration box, and a DC voltage signal E is output. The heat flux density value q is calculated using the formula q=E / S, where S is the sensitivity of the heat flux density sensor.

[0081] The operating standard deviation is used to determine the stability of the air temperature inside the refrigeration box and the air temperature outside. The operating standard deviation is determined as follows: When both are less than 0.08, it indicates that the temperature inside and outside the refrigeration box has reached a stable state.

[0082] S2. Construct a digital filtering algorithm heat flow denoising model.

[0083] Among them, the digital filtering algorithm heat flow denoising model includes rolling average model, moving average model, median model, limit model, SG model, Gaussian model and Kalman model.

[0084] Specifically, the rolling average model is:

[0085] The moving average model is: v = (n-1) / 2.

[0086] The median model is: q fk =median n {q k-v ,...,q k-1 ,qk ,q k+1 ,...,q k+v},(k=1,2,...v=(n-1) / 2).

[0087] The clipping model is:

[0088] The SG model is:

[0089] The Gaussian model is:

[0090]

[0091] The Kalman model is:

[0092] and

[0093] S3. Construct repeatability steady-state evaluation indicators and fidelity steady-state evaluation indicators.

[0094] Among them, the repeatability steady-state evaluation indicators include standard deviation and mean absolute deviation, and the fidelity steady-state evaluation indicators include signal-to-noise ratio and correlation coefficient.

[0095] Specifically, the signal-to-noise ratio is:

[0096] The standard deviation is:

[0097] The mean absolute deviation is:

[0098] The correlation coefficient is:

[0099] S4. Construct dynamic evaluation indicators of repeatability and fidelity.

[0100] Among them, the dynamic evaluation indicators of repeatability include the operating standard deviation and the operating mean absolute deviation, and the dynamic evaluation indicators of fidelity include the operating signal-to-noise ratio, the operating correlation coefficient and the operating coefficient of variation.

[0101] Specifically, the operating signal-to-noise ratio is:

[0102] The running standard deviation is:

[0103] The running mean absolute deviation is:

[0104] The running correlation coefficient is:

[0105] The running coefficient of variation is:

[0106] S5. Select the optimal filtering model and its operating parameters.

[0107] Specifically, set the window width WW range of the median model, set the window width WW range of the moving average model, set the window width WW range of the median model, set the limit value L range of the limit model, set the window width WW range of the SG model and the highest order ρ range of the polynomial, set the window width WW range of the Gaussian model and the distribution parameter σ range, set the process noise covariance Q of the Kalman model k Range and measurement noise covariance R k Range, calculate the denoised heat flux value q fk And steady-state evaluation indicators SNR, SD, MAD, CC, plot SNR, SD, MAD, CC with WW, L, ρ, σ, Q k , R k The local optimal working parameters are determined by the intersection area rule based on the change curve of

[0108] The window width WW is set to 5, 11, 51, 101, 201, 401, 801, the limit value L is set to 2, 1.5, 1, 0.5, 0.2, 0.1, 0.05, the highest order of the polynomial ρ is set to 51, 31, 15, 9, 5, 3, 1, the distribution parameter σ is set to 0.5, 3, 7, 11, 17, 21, 25, and the process noise covariance Q is set to k Range 10 -3 , 10 -4 , 10 -5 , 10 - 6, 10 -7 , 10 -8 , 10 -9 , set the measurement noise covariance R k Range 10 -7 , 10 -6 , 10 -5 , 10 -4 , 10 -3 , 10 -2 , 10 -1 .

[0109] Furthermore, the step of using the intersection area rule to determine the local optimal working parameters includes:

[0110] S51, first locate the intersection working parameter point of the standard deviation, mean absolute deviation and correlation coefficient curves.

[0111] S52. Based on the principle of repeatability priority, the point with the smallest standard deviation in the intersection area is taken as the optimal working parameter point.

[0112] S6. Clarify the sample capacity of heat flow signal.

[0113] Specifically, under the premise that the sample starting point is fixed, the sample end point rolls forward to calculate the running signal-to-noise ratio RSNR, running standard deviation RSD, running mean absolute deviation RMAD, running correlation coefficient RCC and four running coefficients of variation (the four running coefficients of variation are respectively the running signal-to-noise ratio coefficient of variation Rc v-SNR , coefficient of variation of operating standard deviation Rc v-SD , running mean absolute deviation coefficient of variation Rc v-MAD and the coefficient of variation of the running correlation coefficient Rc v-CC ), draw the running coefficient of variation curves at different sample endpoints, and compare the coefficient of variation c v = 2% of the last intersection point corresponding to the horizontal axis value is regarded as the sample-independent capacity l d .

[0114] Among them, the operating coefficient of variation Rc v-SNR , Rc v-SD , Rc v-MAD and Rc v-CC There are four curves, c v =2% is a horizontal straight line. When drawn on the same coordinate axis, the four curves and a horizontal straight line have multiple intersection points. Within the sample capacity range, the horizontal coordinate value of the last intersection point, that is, the maximum horizontal coordinate value, is the sample irrelevant capacity l d .

[0115] S7. Define the sampling period of heat flow signal samples.

[0116] Adjust the sampling period Δt of the heat flux signal, obtain the heat flux density samples with different sampling periods, input the denoising model and the dynamic operation index calculation model, calculate the heat flux density and dynamic operation index after denoising, repeat step S6, and obtain the sample-independent capacity l under different sampling periods d , calculate the sample independence duration t under different sampling periods t , which is the sampling period Δt and the sample-independent capacity l d The product of t =t t × d .

[0117] During the research, the inventors found that the prior art did not systematically discuss how to select the appropriate type of filter, determine the optimal working parameters of the filter and the optimal sampling parameters of the sample data. For example, the arithmetic mean method is used to smooth the observation curve. This method is too simple and rough, which will cover up the real fluctuation of the signal. In addition, only a single measurement was performed, and the repeatability of the measurement results was not discussed. Repeated measurements under the same measurement conditions cannot determine whether the same measurement results can be obtained.

[0118] The observed heat flux signal is smoothed using a rolling average filter, and the running average standard deviation is introduced to judge the stability. However, if the fidelity of the denoised signal is not judged, it is easy to cause the generation of distorted signals and the loss of important signal features (the smoothing effect of the rolling average filter can easily mask important features in the signal, especially in applications that need to monitor mutations, outliers or instantaneous changes. Smoothing can also cause these key changes to be ignored, thereby missing the judgment of signal abnormalities. For example, in heat flux signals, some special thermal peaks or load mutations are more important, but this information is easily lost due to excessive smoothing), which in turn affects the accuracy of stability judgments. In other fields, in order to obtain more accurate heat flux signals, existing technologies focus on the design of advanced heat flux density sensors and the improvement of filtering algorithms, such as designing a new sensor to measure less than 0.2W / cm 2 The small convective heat flux of the sensor is measured, the measurement performance of bimetallic and semiconductor heat flux sensors is compared, and the inverse problem of nonlinear heat conduction-radiation heat transfer is solved by using Kalman filter. In the past, there was a lack of discussion on how to select a suitable filter for a given heat flux signal and determine the sampling parameters of the measurement signal.

[0119] The root mean square error between the exact solution and the estimated value is a common evaluation index for quantitative evaluation of the denoising effect of heat flow signals. The root mean square error is sensitive to the degree of deviation between the two sets of signals. In the field of image denoising, indicators such as two-dimensional correlation coefficient, structural similarity, peak signal-to-noise ratio and contrast-to-noise ratio are often used to evaluate the denoising effect. Among them, the signal-to-noise ratio represents the amount of denoising, and the correlation coefficient represents the degree of proximity between the original observed data and the filtered data. In the denoising process, how to determine the appropriate denoising amount to ensure that the noise information is filtered out instead of the real heat flow information For quasi-steady-state heat flow signals, the larger the denoising amount, the smoother the curve, the higher the repeatability level, and the lower the fidelity level. Conversely, the smaller the denoising amount, the greater the volatility of the curve, the lower the repeatability level, and the higher the fidelity level. Commonly used evaluation indicators only give the denoising results and denoising effects under different conditions.

[0120] In this application, the denoising method for measuring the heat flow signal of the refrigeration device cabinet constructs a heat flow denoising model of a digital filtering algorithm, proposes a filter and its working parameters that are compatible with repeatability and fidelity to determine the measured heat flow signal, and constructs a coefficient of variation to determine the local optimal sampling size and sampling period of the online sample data. The proposed multidimensional quantitative evaluation index based on the repeatability and fidelity of the denoising effect can take into account both repeatability and fidelity, and facilitates the determination of the optimal working parameters of the filter.

[0121] It can be understood that the intersection area rule adopted by the present invention compares the denoising effects of various filters, making it easy to select filters and their operating parameters with large denoising amount but high signal feature fidelity level.

[0122] In the specific experiment, for the selection of sample signals, the internal temperature of the refrigeration box needs to be set to 45°C and the external ambient temperature of the box to 25°C, which is used to simulate the actual refrigeration process where the internal temperature difference of 5°C and the ambient temperature is maintained at 20°C. A set of 50,000 time series steady-state heat flow signals measured by the CHC30 sensor is used as samples, with a range of 5W·m -2 ~15W·m -2 The sampling period is 3.308s, and it takes about 48h to collect 50,000 heat flow signals. In fact, including the collection of the unsteady heat flow signal at the beginning of the observation, a total of 52,500 heat flow signals were collected.

[0123] See also Figure 2 and Figure 3 As shown in the figure, it shows the change trend of the air temperature inside and outside the box, the heat flow signal and the stability analysis results, where T i -RSD, T e -RSD and q-RSD represent the operating standard deviations of the internal and external air temperatures and heat flow signals, respectively, that is, from the moment of collection, they are continuously sliding forward for calculation and update. It can reflect two aspects of information: one is that the operating standard deviation of the temperature signal is as low as about 0.070, while the operating standard deviation of the heat flow signal is as high as over 1.000, indicating that the temperature signal has a small degree of discreteness and a high level of repeatability, while the heat flow signal has a large degree of discreteness and a low level of repeatability, and it is difficult to obtain a definite steady-state value, and denoising is urgently needed; second, starting from 2.297h (i.e., the 2501st observation point), the operating standard deviations of the temperature signal and the heat flow signal have changed little, and it can be determined that the box is in a thermally stable state. The sample data from 1st to 50000th are used in this invention to explore the influence of filtering algorithms and data characteristics on the denoising effect. The sample fluctuation range is 5 to 15 W·m -2 The arithmetic mean is 6.994 W·m -2 , based on the arithmetic mean, the volatility is -28.51% to 20.44%.

[0124] For the selection of working parameters of the filtering algorithm and their influence on the denoising effect. First, the working parameters are selected, and six digital filters are implemented on the MATLAB software platform. The window width is an important working parameter of the median, moving average, SG and Gaussian filters, which determines the length of the observation point data involved in the filtering calculation each time; the highest order (p) of the SG filter polynomial will affect the fitting coefficient of the polynomial, the limit value (L) is the only denoising basis of the limiting filter, the distribution value (σ) determines the denoising template of the Gaussian filter, and the process noise covariance (Q k ) and the measurement noise covariance (R k ) affects the dynamic response speed and convergence stability of the Kalman filter. By analyzing the influence of the above working parameters on the denoising effect, the local optimal working point of each digital filter within the set working parameter range is found. The following table shows the preset working parameters of the six filters. In order to avoid missing the optimal working point, each working parameter is set in a wide range:

[0125] Table 1 Preset values ​​of filter operating parameters

[0126]

[0127] Regarding the effect of window width (WW), the highest order of the SG filter and the distribution parameter of the Gaussian filter are both kept at 3.

[0128] See also Figure 4-Figure 7 As shown in the figure, RAW is the original observed signal curve. It can be seen from the figure that the larger the window width (WW), the shorter the denoised heat flow signal curve. This is because as WW increases, the effective starting point of the filter calculation moves forward and the effective end point moves backward. Qualitatively, for the median, moving average and SG filters, the denoising effect is better as WW increases. However, the Gaussian filter shows the opposite trend. The observed signal point to be denoised is the center point of the time signal sequence within the WW, and its denoising effect is related to other signal points before and after it. The smaller the WW, the less the median and moving average filters are affected by nearby observed signal points, so the fidelity is also better. For the SG filter, all WW values ​​are greater than their highest order values. In essence, when WW increases, more overlapping observed signal points will participate in multiple fittings, resulting in a smoother denoised heat flow signal curve. For the Gaussian filter, WW is also the length of the denoising template, and the weight coefficients in the denoising template obey the Gaussian distribution, that is, the normal distribution. The closer to the signal point to be denoised, the greater the weight of the observed signal point, that is, the greater the impact on the denoising result. Figure 7 It shows that the more original observation signal points are involved in the weight calculation, the worse the denoising effect.

[0129] See also Figure 8-Figure 10As shown in the figure, the changing trends of the evaluation index curves are very similar, indicating that the median, mean and polynomial fitting values ​​of the observed heat flow signals within the WW are close. When WW increases from 5 to 801, SNR, SD and CC fluctuate between 25db~15db, 1~0.15 and 0.9~0.6, respectively, and show a downward trend. On the contrary, MAD increases from 0.3 to 0.9. Overall, the above trends show that with the improvement of noise removal and repeatability, the fidelity will decrease. However, the changing trend of the evaluation index curve is not strictly monotonic. Within a certain WW, there may be an extreme value, and the extreme point is the potential optimal working point. Due to q k -q fk The terms are located in the denominator and numerator respectively, and the SNR and MAD curves are almost symmetrical.

[0130] See also Fig.11 As shown in Figure 1, the evaluation index curve of the Gaussian filter has a different trend. The SNR increases from 3.3dB to 22.7dB, while the MAD decreases from 2.8 to 0.38. The less noise is removed, the better the fidelity. CC remains at around 0.9 and does not change with WW.

[0131] It can be understood that the WW independence values ​​of the median, moving average, SG, and Gaussian filters are 401, 401, 401, and 51, respectively. The difficulty lies in how to determine the local optimal WW value by using multiple evaluation indicators. As mentioned earlier, SNR and MAD have the same important term q k -q fk , so MAD is retained to determine the local optimal WW value, and SNR is only used to evaluate the denoising amount. The intersection area rule can be used to determine the local optimal WW value: first, locate the intersection point of the SD, MAD and CC curves; then, based on the principle of repeatability priority, the point with the smallest SD in the intersection area is taken as the optimal WW value. According to this rule, the optimal WW values ​​of the median, moving average and SG filters are 51, 51 and 101 respectively. For the Gaussian filter, 11 should be the optimal WW. Since MAD is smaller when other indicators do not change, 101 is selected as the final optimal WW. In addition, the WWs of 51 and 101 are not large. When used for actual heat flow signal measurement, a large amount of data does not need to be collected, which can save collection time.

[0132] See also Figure 8-Figure 10 As shown in the figure, when WW is 801, the SNR of the arithmetic mean method is close to that of the median, moving average, and SG filters. It can be inferred that the denoising amount of the arithmetic mean method is similar to that of the above three filters. On the other hand, the difference between SD and MAD indicates that the shape of the removed noise will also be different. Fig.11As shown in the figure, when WW is 5 to 11, the SNR of the arithmetic mean method is close to that of the Gaussian filter, and its lower SD and MAD values ​​indicate that the arithmetic mean method has repeatability and fidelity advantages compared with the above four filters under maximum or minimum WW conditions. However, compared with the optimal WW condition, the denoising amount of the four filters is less than that of the arithmetic mean method, the repeatability level is reduced, the fidelity level is improved, and the authenticity of the denoised heat flow signal is greatly improved.

[0133] See also Figure 12-13 As shown, regarding the influence of the highest order (ρ), Fig.12 It is shown in Figure 2 that the denoising effect increases as ρ decreases. The polynomial with a large ρ value can better fit the changing trend of the 101 observation points, and the denoised signal is closer to the original observation signal. Fig.13 In the figure, when ρ decreases to 3, SNR, MAD, SD, and CC change dramatically. The larger ρ is, the closer the denoised heat flow curve is to the original observed curve, but it does not mean that the two curves can completely match when ρ reaches the maximum value. WW is the number of points of polynomial interpolation. When ρ increases to a value close to the number of interpolation points, polynomial oscillation may occur. Fitting the cubic polynomial q using the least squares method i =a 0 +a 1 i+a 2 i 2 +a 3 i 3 , the third-order SG filter retains the low-frequency components of the original observation signal and removes the high-frequency components, that is, the third-order heat flow signal ( Fig.12 The brown curve in the figure is the original observed heat flow signal ( Fig.12 The black curve in Fig. 3 is selected as the optimal ρ of the SG filter polynomial.

[0134] See also Fig.12 As shown in the figure, the SNR of the arithmetic average method is close to that of the first-order SG filter. The lower SD and MAD of the arithmetic average method mean that it has repeatability and fidelity advantages compared with the first-order SG filter. As ρ increases, the repeatability level decreases and the fidelity level increases. At the optimal operating parameter point, the MAD of the SG filter is 33.4% lower than that of the arithmetic average method.

[0135] See also Figure 14-15 As shown, regarding the influence of the limit value (L), Fig.14In the figure, when L is set to 1.0, an obvious local straight line appears. The limiting filter can filter out noise with higher peak values, but it cannot filter out periodic noise. For example, when broilers flap their wings, pulsating noise often appears in the broiler weighing system. Under such working conditions, the limiting filter algorithm shows good denoising effect. Obviously, the original heat flow signal fluctuates periodically. If L is not set reasonably, denoising failure will occur, such as the L-1.0 curve. Fig.15 In the example, the mutation value of L is 1. When L is less than 1, except for SD, MAD, CC and SNR oscillate periodically. The distribution frequency of the sample data is shown in the following table:

[0136] Table 2 Distribution frequency of sample data

[0137]

[0138] From Table 2, we can see The frequency distributed in [-1,1] is as high as more than 60%, that is, when L is less than 1, the limiting filter has a better denoising effect, and 0.05 is selected as the optimal L.

[0139] See also Figure 16-Figure 19 As shown in Figure 3, the distribution parameter, namely the standard deviation (σ), has a great influence on the denoising effect, and the distribution parameter directly controls the shape of the Gaussian filter denoising template. Fig.16 As σ increases, G(q k ) The curve becomes shorter and wider. When σ is smaller, it is closer to The larger the weight coefficient of the observation point, the greater the noise reduction factor. When σ is large, the denoising template is approximately a straight line, that is, all observation points within the window width have the same weight. At this time, the denoising effect of the Gaussian filter is similar to that of the moving average filter. Fig.18 The distribution parameter rule of Fig.19 The actual Gaussian denoising template for the study case is given. k ) increases with the increase of σ. 11 is the critical σ value. When σ is greater than 11, it is difficult for the denoising template to present a normal distribution. Fig.17 All evaluation indicators in the are approximately linearly varied, and there is a strict contradiction between the repeatability and fidelity of the Gaussian filter. The middle point, i.e., 11, is regarded as the local optimal σ value.

[0140] The size of the σ value plays a key role in the retention and removal of mutation points. Choosing a suitable σ value can retain the information of the mutation points while suppressing noise. Fig.17In the example above, the SNR decreases strictly as σ increases. The SNR of the arithmetic mean method is close to that of the Gaussian filter with σ of 25, and although the two methods achieve similar amounts of denoising, the MAD and SD are different, which can be explained by the difference in the shape of the denoised signal curve. When σ is 11, the amount of denoising is reduced, resulting in a higher level of fidelity.

[0141] See also Figure 20-23 As shown, for the process noise covariance (Q k ) and the measurement noise covariance (R k ), the process noise covariance (Q k ) and the measurement noise covariance (Q k ) determines the covariance of the kth prior heat flow estimate and the Kalman gain matrix (K k ). The predicted value based on the kth heat flow signal and measured values The optimal estimate of the kth heat flow signal The kth denoising result can be recursively calculated. In practical engineering, since it is difficult to determine the values ​​of process noise and measurement noise, Q k and R k is usually unknown. In the embodiment, Q k and R k The optimal estimate is crucial. Since there is no clear selection principle in existing research, trial and error is usually used for calculation and analysis. k Initially fixed at 1×10 -3 , Q k Set to 1×10 -9 ~1×10 -3 . After finding the local optimal Q k After R k Set to 1×10 -7 ~1×10 -1 .

[0142] Q k and R k To K k are positively correlated and negatively correlated, respectively. k Reduce or R k As it increases, the kth prior heat flow estimate The weight of the prior estimate increases, that is, in the optimal estimate of the denoising result, the effective information of the prior estimate accounts for a larger proportion. Close to Fig.21 and Fig.23 The curve shows that as The increase in weight leads to a greater amount of denoising, and accordingly, a higher level of repeatability and a lower level of fidelity. Therefore, in engineering applications, when the measurement conditions are poor, it is recommended to select a larger R k / Q k , in order to reduce the oscillation of the recursive result. Finally, select Q k =1×10 -7 and R k =1×10 -3 As the local optimal working point of the Kalman filter suitable for the heat flow signal measured in this embodiment. Fig.21 and Fig.23 In, R k / Q k The range of is 1~1×106, which can explain that the changing trends of the evaluation indicators in the two figures are basically the same. Fig.23 In, when Q k Set to 1×10 -3 , 1×10 -2 and 1×10 -1 When Q is , the SNR of the Kalman filter remains basically unchanged and is close to the SNR of the arithmetic mean method. The MADs are slightly lower than the MADs of the arithmetic mean method, and the SDs show the opposite trend. k When it is higher, the denoising effect of the Kalman filter is close to the arithmetic mean method.

[0143] For the comparative analysis of the denoising effects of different filtering algorithms, combined with the above analysis results, it can be inferred that the filtered noise includes the denoising amount and the denoising shape. SNR can only reflect the size of the denoising amount, not the denoising shape. Under the premise of the same denoising amount, the difference in denoising shape can be reflected by the three indicators of SD, MAD and CC. When using different filtering algorithms, due to the different denoising characteristics, the same SNR corresponds to different SD, MAD and CC combinations. In order to prevent some mutation points of the original observed heat flow signal from being deleted, multi-dimensional indicators are used to evaluate the repeatability and fidelity levels.

[0144] The optimal working parameters and denoising effects of different filters are shown in the following table:

[0145] Table 3 Optimal working parameters and denoising effects of different filters

[0146]

[0147] Except for the algorithm averaging method, the denoising result is the last value in each group of denoised heat flow time series signals.

[0148] See also Fig.24As shown in Figure 1, the order of noise removal is as follows: Clipping filter, Kalman filter, moving average filter, SG filter, median filter, and Gaussian filter. SNR and SD strictly increase in this order, and MAD strictly decreases in this order. However, CC interrupts this order: Kalman and moving average filters are slightly higher than clipping and SG filters, respectively. Clipping and moving average filters have both repeatability and fidelity advantages compared to Kalman and SG filters.

[0149] The denoising results of the median, moving average and Kalman filters are close to the arithmetic mean, the denoising results of the SG and Gaussian filters are close to the original observed values, and the denoising result of the limiting filter is the smallest. -2 For steady-state heat flow signals, the following filter selection rules are proposed:

[0150] (1) If you want to ensure high repeatability and fidelity at the same time, moving average filter and SG filter are recommended;

[0151] (2) If repeatability is a priority, the Kalman filter is recommended;

[0152] (3) If fidelity is a priority, the median filter is recommended;

[0153] (4) Gaussian filters are suitable for situations where high fidelity is required. Generally speaking, they are more suitable for studying the changes in the waveform of heat flow signals rather than just taking the final value;

[0154] (5) The limiting filter is not suitable for denoising the measured heat flow signal. The limiting filter has a good limiting effect on pulse noise. Obviously, the noise in this embodiment is periodic, not pulsating.

[0155] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of application of the present invention should still fall within the scope of the claims of the present invention.

Claims

1. A denoising method for measuring heat flow signals of a refrigeration device cabinet, characterized in that: include: S1. After the air temperature inside the refrigeration box and the air temperature outside the refrigeration box are stable, the heat flux density sensor is posted on the wall of the refrigeration box, a DC voltage signal is output, and the heat flux density value is calculated; S2. Constructing a digital filtering algorithm heat flow denoising model, wherein the digital filtering algorithm heat flow denoising model includes a rolling average model, a moving average model, a median model, a clipping model, an SG model, a Gaussian model and a Kalman model; S3. constructing a repeatability steady-state evaluation index and a fidelity steady-state evaluation index, wherein the repeatability steady-state evaluation index includes a standard deviation and a mean absolute deviation, and the fidelity steady-state evaluation index includes a signal-to-noise ratio and a correlation coefficient; S4. Constructing a dynamic evaluation index of repeatability and a dynamic evaluation index of fidelity, wherein the dynamic evaluation index of repeatability includes a running standard deviation and a running mean absolute deviation, and the dynamic evaluation index of fidelity includes a running signal-to-noise ratio, a running correlation coefficient, and a running coefficient of variation; S5. Set the window width range, limit value range, polynomial highest order range, distribution parameter range, process noise covariance range and measurement noise covariance range of each model, calculate the heat flux value and steady-state evaluation index after denoising, and use the intersection area rule to determine the local optimal working parameters; S6. Under the premise that the sample starting point is fixed, the sample end point is rolled forward, and the running signal-to-noise ratio, running standard deviation, running mean absolute deviation, running correlation coefficient and four running coefficients of variation are calculated. The running coefficient of variation curves under different sample end points are drawn. The horizontal axis value corresponding to the last intersection point of each curve with the coefficient of variation of 2% is regarded as the sample-independent capacity; S7, adjust the sampling period of the heat flux signal, calculate the heat flux density and dynamic operation index after denoising, repeat step S6, obtain the sample independence capacity under different sampling periods, and calculate the sample independence duration under different sampling periods.

2. A denoising method for measuring heat flow signals of a refrigeration device cabinet according to claim 1, characterized in that: In step S1, the operating standard deviation is used to determine the stability of the air temperature inside the refrigeration box and the air temperature outside the refrigeration box.

3. A denoising method for measuring heat flow signals of a refrigeration device cabinet according to claim 2, characterized in that: The air temperature inside the refrigeration box is recorded as T in , the external air temperature of the refrigeration box is recorded as T out , the running standard deviation is determined as follows: When both are less than 0.08, it indicates that the temperature inside and outside the refrigeration box has reached a stable state.

4. A denoising method for measuring heat flow signals of a refrigeration device cabinet according to claim 1, characterized in that: In step S2, the heat flow denoising model of the digital filtering algorithm is as follows: The rolling average model is: The moving average model is: The median model is: q fk =median n {q k-v ,...,q k-1 ,q k ,q k+1 ,...,q k+v },(k=1,2,...v=(n-1) / 2); The clipping model is: The SG model is: The Gaussian model is: The Kalman model is:

5. The method for denoising a heat flow signal of a refrigeration device cabinet according to claim 1, characterized in that: In step S3, the calculation formulas of the repeatability steady-state evaluation index and the fidelity steady-state evaluation index are as follows: The signal-to-noise ratio is: The standard deviation is: The mean absolute deviation is: The correlation coefficient is:

6. A denoising method for measuring heat flow signals of a refrigeration device cabinet according to claim 1, characterized in that: In step S4, the calculation formulas of the repeatability dynamic evaluation index and the fidelity dynamic evaluation index are as follows: The operating signal-to-noise ratio is: The running standard deviation is: The running mean absolute deviation is: The operational correlation coefficient is: The running coefficient of variation is:

7. A denoising method for measuring heat flow signals of a refrigeration device cabinet according to claim 1, characterized in that: In step S5, the window width is set in the range of 5, 11, 51, 101, 201, 401, 801, the limit value is set in the range of 2, 1.5, 1, 0.5, 0.2, 0.1, 0.05, the highest order of the polynomial is set in the range of 51, 31, 15, 9, 5, 3, 1, the distribution parameter is set in the range of 0.5, 3, 7, 11, 17, 21, 25, the process noise covariance is set in the range of 10 -3 , 10 -4 , 10 -5 , 10 - 6, 10 -7 , 10 -8 , 10 -9 , the measurement noise covariance range is set to 10 -7 , 10 -6 , 10 -5 , 10 -4 , 10 -3 , 10 -2 , 10 -1 .

8. The method for denoising a heat flow signal of a refrigeration device cabinet according to claim 1, characterized in that: In step S5, the step of using the intersection area rule to determine the local optimal working parameters includes: S51, locating the intersection working parameter point of the standard deviation, mean absolute deviation and correlation coefficient curves; S52. Based on the principle of repeatability priority, the point with the smallest standard deviation in the intersection area is taken as the optimal working parameter point.

9. A denoising method for measuring heat flow signals of a refrigeration device cabinet according to claim 1, characterized in that: In step S6, the four operating coefficients of variation are respectively the operating signal-to-noise ratio coefficient of variation, the operating standard deviation coefficient of variation, the operating mean absolute deviation coefficient of variation and the operating correlation coefficient coefficient of variation. The operating coefficients of variation are four curves, and the 2% coefficient of variation is a horizontal straight line. When drawn on the same coordinate axis, the four curves and the horizontal straight line have multiple intersection points respectively. Within the sample capacity range, the horizontal coordinate value of the last intersection point, that is, the maximum horizontal coordinate value, is the sample independence capacity.

Images