Refrigerator load temperature regulation and control system

Abstract

The invention relates to the field of refrigerator temperature regulation and control, in particular to a refrigerator load temperature regulation and control system, which comprises the following steps: acquiring load data from a historical temperature curve, extracting periodic characteristics and trend characteristics by adopting an STL time sequence decomposition method, and generating a load mode reference sequence; from real-time compressor energy consumption data, a CUSUM change point detection algorithm is adopted to segment the sequence and generate a multi-segment independent energy consumption sequence, and through multi-level data fusion and dynamic analysis, the accuracy and stability of refrigerator load temperature regulation and control are remarkably improved, and the risk of energy consumption abnormity and temperature fluctuation is effectively reduced.

Description

TECHNICAL FIELD

[0001] The application relates to the field of refrigerator temperature regulation, and in particular to a refrigerator load temperature regulation system. BACKGROUND

[0002] In the refrigerator load temperature regulation system, there is a complex technical problem. The system needs to process historical temperature curves, real-time humidity fluctuations, and compressor energy consumption data to identify potential anomalies. First, the system performs time series decomposition on the load data to extract periodic and trend features. Then, through Fourier transform and exponential smoothing methods, a load pattern benchmark sequence is generated. The system also needs to fuse this benchmark sequence with real-time data to generate a synthetic temperature sequence.

[0003] The challenge faced by the system is how to accurately detect temperature deviations. It needs to compare the synthetic temperature sequence with the real-time temperature curve, and use the dynamic time warping algorithm to identify the deviation starting point. At the same time, the system also needs to process the compressor energy consumption data, and use the CUSUM algorithm and adaptive time window mechanism to generate a continuous energy consumption sequence.

[0004] Another key issue of the system is how to judge the consistency of temperature and energy consumption trends. It needs to calculate the correlation coefficient and perform multivariate linear regression when necessary to correct the trend. Finally, the system needs to quantify abnormal trends, generate an index sequence, and compare it with the preset threshold to identify abnormal points. This process involves the coordinated work of multiple complex algorithms, and needs to strike a balance between real-time data processing and accurate anomaly detection. SUMMARY

[0005] The purpose of the present application is to provide a refrigerator load temperature regulation system to solve the technical problems raised in the background.

[0006] Based on the above idea, the present application provides the following technical solutions: Obtain the past 24-hour load data from the historical temperature curve, use the STL time series decomposition method to extract periodic and trend features, and generate a load pattern benchmark sequence; According to the load pattern benchmark sequence, combine the real-time humidity fluctuation data, and generate a synthetic temperature sequence through linear weighted fusion, wherein the weight is calculated based on the physical impact coefficient of humidity on temperature, and the formula is: St= wh*Ht + (1 - wh)*Lt, St is the synthetic temperature sequence, Ht is the humidity fluctuation data, Lt is the load pattern benchmark sequence, and wh is the humidity weight; Compare the synthetic temperature sequence with the real-time temperature curve, calculate the mean square error MSE, and the formula is: MSE = Σ(St - Rt)^2 / n, St is a synthetic temperature sequence, Rt is a real-time temperature curve, n is a sequence length, if MSE exceeds a preset threshold Tmse, Tmse is calculated based on historical data variance, then a temperature deviation sequence is generated; For the temperature deviation sequence, a dynamic time warping algorithm is used to calculate the alignment path of the synthetic temperature sequence and the real-time temperature curve, the formula is: DTW(S, R) = min(Σd(Si, Rj)), d is the Euclidean distance, Si and Rj are points of the synthetic temperature sequence and the real-time temperature curve respectively, and the temperature deviation starting point sequence is generated; From the real-time compressor energy consumption data, a CUSUM change point detection algorithm is used to segment the sequence, the formula is: Sk = max(0, S{k-1} + (Ek - μ0)), Sk is the cumulative sum, Ek is the energy consumption data, μ0 is the energy consumption mean, and multiple independent energy consumption sequences are generated; For multiple independent energy consumption sequences, an adaptive time window sliding mechanism is used, the window size is calculated based on the energy consumption fluctuation frequency Fe, the formula is: W = 1 / Fe, The bridged continuous energy consumption sequence is generated by weighted average, the weight formula is wi = 1 / σi^2, wi is the weight in the window, σi is the local energy consumption variance; The Pearson correlation coefficient is calculated between the bridged continuous energy consumption sequence and the synthetic temperature sequence, the formula is ρ = cov(Ec, St) / (σEc*σSt), E_c is the bridged continuous energy consumption sequence, St is the synthetic temperature sequence, cov is the covariance, σ is the standard deviation, if ρ is lower than the preset threshold Tρ, Tρ is calculated based on historical data statistics, then the trend inconsistent sequence is generated; For the trend inconsistent sequence, the humidity fluctuation data and multiple independent energy consumption sequences are fused by multivariate linear regression, the formula is: Tr = β0 + β1*Ht + β2*Em, Tr is the modified trend sequence, Ht is the humidity fluctuation data, Em is the multiple independent energy consumption sequence, β is the regression coefficient, and the modified trend sequence is generated; According to the modified trend sequence, the Z-score algorithm is used to calculate the temperature rise rate and duration, the formula is: Z = (Tr - μT) / σT, Tr is the corrected trend sequence, μT and σT are the mean and standard deviation respectively, the abnormal trend quantitative index sequence is generated, compared with the preset threshold Tz, Tz is calculated based on the historical data distribution, and the refrigerator load temperature regulation system abnormal point sequence is generated.

[0007] The technical scheme of the present application can include the following beneficial effects: The present application discloses a refrigerator load temperature regulation anomaly detection method based on multi-source data fusion, aiming at the business scene problem of inconsistent load temperature fluctuation and energy consumption trend in the operation of the refrigerator, an intelligent regulation system is constructed by integrating historical temperature, real-time humidity and compressor energy consumption data. First of all, the present application extracts the load mode reference sequence by STL decomposition, generates the synthetic temperature sequence combined with the humidity weight, and calculates the mean square error by comparing with the real-time temperature curve to identify the temperature deviation. Then, the aligned path is generated by the dynamic time warping algorithm to accurately locate the deviation starting point. For energy consumption data, the present application uses CUSUM variable point detection to segment the sequence, combines the adaptive time window sliding mechanism to generate continuous energy consumption sequence, and analyzes the correlation between it and the synthetic temperature sequence by Pearson correlation coefficient. For inconsistent trend sequence, the present application fuses humidity and energy consumption data by multivariate linear regression to generate the corrected trend sequence, and uses Z-score algorithm to quantify the abnormal trend to generate the regulation abnormal point sequence. Through multi-level data fusion and dynamic analysis, the present application significantly improves the accuracy and stability of the refrigerator load temperature regulation, effectively reduces the energy consumption anomaly and temperature fluctuation risk. BRIEF DESCRIPTION OF DRAWINGS

[0008] Fig. 1 A flowchart of a refrigerator load temperature regulation system of the present application.

[0009] Fig. 2 A schematic diagram of a refrigerator load temperature regulation system of the present application.

[0010] Fig. 3 Another schematic diagram of a refrigerator load temperature regulation system of the present application. DETAILED DESCRIPTION EMBODIMENT

[0011] As Figs. 1-3 , the refrigerator load temperature regulation method of the present embodiment can specifically include: S101, obtain the past 24 hours load data from the historical temperature curve, use the STL time series decomposition method to extract the periodic feature and trend feature, and generate the load mode reference sequence.

[0012] The past 24 hours of load data in the historical temperature curve is obtained, and a preset time window division method is used to obtain an initial load data set. The initial load data set is processed by the STL time series decomposition method to separate the periodic component, the trend component and the residual component, and the decomposed time series features are obtained. For the decomposed time series features, the periodic component and the trend component are extracted, the frequency characteristics of the periodic component are analyzed by Fourier transform, and the periodic characteristics are determined. According to the periodic characteristics and the trend component, the moving average method is used to smooth the trend component to generate a smoothed trend sequence. If the fluctuation amplitude of the smoothed trend sequence exceeds the preset threshold, the sequence is subjected to secondary smoothing, and the exponential smoothing method is used to obtain a stable trend sequence. By combining the periodic characteristics and the stable trend sequence, the linear interpolation method is used to fill in the missing data to generate a complete load mode reference sequence.

[0013] In S102, a synthetic temperature sequence is generated by linear weighted fusion according to the load mode reference sequence and real-time humidity fluctuation data, wherein the weight is calculated based on the physical impact coefficient of humidity on temperature, and the formula is: St= wh*Ht + (1 - wh)*Lt, St is the synthetic temperature sequence, Ht is the humidity fluctuation data, Lt is the load mode reference sequence, and wh is the humidity weight; The real-time humidity fluctuation data and the load mode reference sequence are obtained, and a preset time window division method is used to obtain an initial fusion data set. The humidity weight coefficient is calculated by looking up the humidity physical impact coefficient table to obtain the weight value. For the initial synthetic temperature sequence, if the sequence fluctuation amplitude exceeds the preset threshold, the moving average method is used for smoothing to obtain a smoothed temperature sequence. According to the smoothed temperature sequence, the linear interpolation method is used to fill in the missing data in the sequence to obtain a complete temperature sequence. For the complete temperature sequence, the frequency characteristics are analyzed by Fourier transform to determine the periodic characteristics. According to the periodic characteristics and the complete temperature sequence, the exponential smoothing method is used for secondary processing of the sequence to obtain a stable synthetic temperature sequence.

[0014] For example, in the actual power load analysis scene, obtaining real-time humidity fluctuation data and load mode reference sequence is an important step in optimizing temperature load prediction. The humidity fluctuation data is usually collected by sensors every hour, for example, the humidity value of a certain area in the past 24 hours may fluctuate between 60% and 80%. The load mode reference sequence is obtained based on the decomposition of the historical temperature curve, reflecting the periodicity and trend characteristics of the load. Assuming that the load reference sequence on a certain day shows a peak at noon 12:00 with a value of 5000 kW and a trough at 4:00 in the morning with a value of 3000 kW. The preset time window can be selected as 1 hour, and the humidity data and the load sequence are aligned by time to form an initial fusion data set.

[0015] Specifically, the construction of the humidity physical impact coefficient table is based on the influence law of the environment on the load. When the humidity is high, the running load of equipment such as air conditioners may increase. Assuming that the lookup table shows that when the humidity is 70%, the corresponding weight coefficient is 0.6, and when the humidity is 80%, the weight coefficient is 0.7. The synthesized temperature sequence is calculated by a linear weighted fusion formula, for example, at a certain time, the humidity data is 75%, the load reference value is 4500 kilowatts, and the weight coefficient is 0.65, then the synthesized value reflects the comprehensive influence of humidity on the load.

[0016] If the fluctuation amplitude of the initial synthesized temperature sequence is large, for example, a certain sequence jumps from 4500 kilowatts to 5500 kilowatts within 1 hour, exceeding the preset threshold of 10%, then the moving average method is used for smoothing processing. Take a 3-hour window, calculate the average value of each hour, generate a smoothed temperature sequence, and reduce the influence of short-term fluctuations.

[0017] For example, for the data missing problem of the smoothed temperature sequence, assuming that the data of a certain hour is missing due to sensor failure, the missing point can be estimated to be 4550 kilowatts based on the linear interpolation method based on the 4500 kilowatts and 4600 kilowatts of the previous and next two hours, and the complete temperature sequence is obtained.

[0018] Specifically, the Fourier transform is used to analyze the frequency characteristics of the complete temperature sequence. Assuming that the analysis finds that the sequence has a 24-hour daily cycle characteristic, indicating that the load is significantly affected by the day-night change. This periodic characteristic can be used to predict future load patterns.

[0019] The exponential smoothing method performs secondary processing on the complete temperature sequence. Assuming that a certain sequence fluctuates greatly during the peak period, by exponential smoothing, the recent data is given a higher weight, for example, the smoothing factor is 0.8, and a stable synthesized temperature sequence is generated. This sequence can better reflect the long-term trend and reduce short-term noise interference, providing a reliable basis for load prediction.

[0020] It can be understood that each step of the above method closely combines the relationship between humidity and load, and ensures that the generated stable synthesized temperature sequence can accurately reflect the influence of environmental factors on the load, providing support for power dispatching and energy management.

[0021] S103, compare the synthesized temperature sequence with the real-time temperature curve, calculate the mean square error MSE, the formula is MSE = Σ(S_t - R_t)^2 / n, S_t is the synthesized temperature sequence, R_t is the real-time temperature curve, n is the sequence length, if MSE exceeds the preset threshold T_mse, T_mse is calculated based on the variance of historical data, then generate a temperature deviation sequence.

[0022] The synthetic temperature sequence and the real-time temperature curve are obtained, the timestamps of the two sequences are aligned through a time alignment method, and the aligned temperature dataset is obtained. Through the mean square error formula: MSE = Σ(St - Rt)^2 / n, St is the synthetic temperature sequence, Rt is the real-time temperature curve, and n is the sequence length. If the MSE exceeds the preset threshold Tmse, Tmse is calculated based on the variance of historical data, then the temperature deviation sequence is generated; the synthetic temperature sequence and the real-time temperature curve are calculated by difference, and the temperature deviation sequence is generated.

[0023] According to the temperature deviation sequence, a local weighted regression method is used to smooth the deviation sequence to obtain a smoothed deviation sequence. For the smoothed deviation sequence, an adaptive threshold segmentation method is used to extract the significant deviation region to obtain a significant deviation dataset. According to the significant deviation dataset, a K-means clustering algorithm is used to classify the deviation data to obtain a deviation classification result. Through the deviation classification result, the statistical features of each class of deviation are extracted to generate a deviation feature dataset.

[0024] For example, after obtaining the synthetic temperature sequence and the real-time temperature curve, time alignment is a key step to ensure the accuracy of subsequent analysis. The time alignment method can be based on linear interpolation or nearest neighbor matching of timestamps.

[0025] If the timestamps of the two sequences have small deviations, nearest neighbor matching can be used to directly select the closest points in time for pairing. This method can ensure the consistency of data points in the time dimension in practical scenarios, such as industrial equipment temperature monitoring, which is convenient for subsequent deviation analysis. In calculating the mean square error, the MSE formula is used to quantify the difference between the synthetic temperature sequence and the real-time temperature curve.

[0026] In the monitoring of the cooling system of a factory, the synthetic temperature sequence may be generated based on humidity and load patterns, while the real-time temperature curve comes from sensor collection. If the MSE value is 5.2 and the threshold Tmse calculated based on the variance of historical data is 4.0, it indicates that the deviation is large and needs to be further processed. This way of quantifying the deviation can intuitively reflect the accuracy of the synthetic sequence and provide a basis for subsequent deviation correction. For the case of large deviation, generating the temperature deviation sequence is a key step.

[0027] By point-by-point subtraction, the difference between the synthetic temperature sequence and the real-time temperature curve is obtained to form the deviation sequence. Assuming that the synthetic temperature at a certain time point is 25.0°C and the real-time temperature is 26.5°C, the deviation is -1.5°C. This deviation sequence can clearly show the difference trend of the two sequences.

[0028] In industrial scenarios, the deviation sequence can be used to identify abnormal temperature fluctuations in device operation. In smoothing the deviation sequence, the local weighted regression method can effectively reduce the influence of noise.

[0029] For example, when processing the cooling system deviation sequence, a sliding window can be set, such as 5 time points, and a smooth value is generated for each point by weighted regression. The data close to the center point in the window is given a higher weight, and the edge point has a lower weight. This method can retain the trend characteristics of the deviation sequence, while reducing the interference of random fluctuations, and is suitable for temperature monitoring scenes that require high stability. For extraction of significant deviation regions, an adaptive threshold segmentation method can dynamically determine the deviation threshold.

[0030] Based on the statistical distribution of the deviation sequence, the threshold is set to the mean plus twice the standard deviation, and the region exceeding the threshold is extracted. Assuming that the mean of the deviation sequence is 0.5°C and the standard deviation is 0.3°C, the threshold is 1.1°C, and the points exceeding this value are marked as significant deviations. This method can effectively locate the temperature abnormal region in practical application, such as the overheating point of the device. In deviation classification, the K-means clustering algorithm can classify the deviation data into different categories.

[0031] The deviation sequence contains positive deviation, negative deviation and deviation close to zero. By setting K=3, the clustering algorithm can classify the data into three categories: high temperature, low temperature and normal range. The classification result can help analyze the pattern of deviation, such as positive deviation may be related to high device load.

[0032] When extracting the statistical features of deviation classification, the mean, variance and range of each category can be calculated. These features can reflect the concentration and variation amplitude of the deviation. This feature extraction can be used in industrial scenarios to optimize temperature control strategies and improve system stability.

[0033] S104, for the temperature deviation sequence, a dynamic time warping algorithm is used to calculate the alignment path of the synthesized temperature sequence and the real-time temperature curve, the formula is: DTW(S, R) = min(Σd(Si, Rj)), d is the Euclidean distance, Si and Rj are points of the synthesized temperature sequence and the real-time temperature curve respectively, and the temperature deviation starting point sequence is generated; The synthetic temperature sequence and the real-time temperature sequence are acquired, and respective temperature sequence points are extracted. If the number of temperature sequence points is inconsistent, the shorter sequence is padded through linear interpolation to obtain a sequence point set with uniform length. The Euclidean distance matrix of the padded synthetic temperature sequence and the real-time temperature sequence is calculated by using a dynamic time warping algorithm, and a distance matrix is obtained. The optimal matching path is calculated based on the distance matrix by using a dynamic programming method, and a sequence alignment path is determined. According to the sequence alignment path, corresponding point pairs of the synthetic temperature sequence and the real-time temperature sequence are extracted, and a temperature deviation sequence is generated. For the temperature deviation sequence, the deviation absolute value of each corresponding point pair is calculated, and a starting point sequence of the deviation sequence is obtained. If the deviation absolute value of the starting point sequence exceeds a preset threshold value, the deviation sequence is smoothed to obtain a smoothed deviation sequence. The final temperature deviation starting point sequence is determined through the smoothed deviation sequence.

[0034] In S105, the CUSUM variable point detection algorithm is used to segment the sequence from the real-time compressor energy consumption data, and the formula is: Sk = max(0, S{k-1} + (Ek - μ0)), Sk is the cumulative sum, Ek is the energy consumption data, μ0 is the energy consumption mean, and multiple independent energy consumption sequences are generated; The real-time energy consumption data of the compressor is acquired, and a continuous energy consumption sequence is extracted from the equipment by using a sensor collection system to obtain the real-time energy consumption sequence. The mean value of the real-time energy consumption sequence is calculated by using a statistical method to generate an energy consumption mean, and an initial reference benchmark is determined. The real-time energy consumption sequence is processed by using the CUSUM algorithm to calculate a cumulative sum sequence, and variable point detection basis data are obtained. If the cumulative sum Sk exceeds a preset threshold value, it is judged that a variable point occurs, and the variable point position is recorded to generate a variable point sequence. The real-time energy consumption sequence is segmented according to the variable point sequence to generate multiple independent energy consumption sequences, and segmented energy consumption segments are obtained. The segmented energy consumption segments are classified by using the K-means clustering algorithm, the operating modes of each segment are determined based on the energy consumption characteristics, and a mode classification result is generated. The energy consumption change trend of each segment is analyzed by using the mode classification result, an energy consumption change characteristic is generated, and an operating state description is determined.

[0035] In S106, the adaptive time window sliding mechanism is used for the multiple independent energy consumption sequences, and the window size is calculated based on the energy consumption fluctuation frequency Fe, and the formula is: W = 1 / F_e, The bridged continuous energy consumption sequence is generated by weighted average, and the weight formula is wi = 1 / σi^2, wi is the weight in the window, and σi is the local energy consumption variance.

[0036] The initial energy consumption data is determined by collecting multiple independent energy consumption sequences. If the energy consumption sequence contains missing values, linear interpolation method is used to fill in, to obtain the complete energy consumption sequence. According to the complete energy consumption sequence, the energy consumption fluctuation frequency is calculated by using fast Fourier transform algorithm to determine the size of the time window. Through the sliding mechanism, the local energy consumption data is obtained by moving the time window W on the complete energy consumption sequence, and the local variance σi is calculated to obtain the data point weight: w_i=1 / σ_i^2, The weighted average method is used to generate the continuous energy consumption sequence after bridging based on the data point weight wi for the local energy consumption data in each time window. If there are abnormal values in the continuous energy consumption sequence, the median filtering algorithm is used for smoothing processing to obtain the smoothed energy consumption sequence. According to the smoothed energy consumption sequence, the size of the time window W is dynamically adjusted by using the adaptive mechanism, and the local variance and weight are recalculated to generate the optimized continuous energy consumption sequence.

[0037] S107, the Pearson correlation coefficient calculation is performed on the bridged continuous energy consumption sequence and the synthesized temperature sequence, and the formula is: ρ = cov(Ec, St) / (σEc*σSt), Ec is the bridged continuous energy consumption sequence, St is the synthesized temperature sequence, cov is the covariance, σ is the standard deviation, and if ρ is lower than the preset threshold Tρ, Tρ is calculated based on historical data statistics, then the trend inconsistent sequence is generated.

[0038] The bridged continuous energy consumption sequence Ec and the synthesized temperature sequence St are obtained, and the standardized processing is performed respectively to obtain the standardized energy consumption sequence Ec' and the standardized temperature sequence St'. Through the standardized energy consumption sequence Ec' and the standardized temperature sequence St', the covariance cov(Ec', St') is calculated, and the covariance formula is used to complete the calculation to obtain the covariance value. According to the standardized energy consumption sequence Ec', the standard deviation σEc' is calculated, and according to the standardized temperature sequence St', the standard deviation σSt' is calculated to obtain two standard deviation values. The Pearson correlation coefficient ρ is calculated by using the covariance value cov(Ec', St') divided by the product of the standard deviations σEc' and σSt', and the correlation coefficient value is obtained. The preset threshold Tρ is obtained from the historical data statistics, and if the correlation coefficient ρ is lower than Tρ, it is determined that the sequence trend is inconsistent, and the trend inconsistent sequence identifier is generated. For the trend inconsistent sequence identifier, the corresponding time period data in Ec and St is extracted to generate the trend inconsistent sequence data set. Through the trend inconsistent sequence data set, the time series analysis method is used to determine the specific time point and change amplitude of the inconsistent sequence to obtain the final analysis result.

[0039] S108, for the trend inconsistent sequence, the humidity fluctuation data and the multiple independent energy consumption sequences are fused by using multivariate linear regression, and the formula is: Tr = β0 + β1* H_t + β2* Em, Tr is the corrected trend sequence, Ht is the humidity fluctuation data, Em is the multi-segment independent energy consumption sequence, β is the regression coefficient, and the corrected trend sequence is generated.

[0040] The humidity fluctuation data and the multi-segment energy consumption sequence are obtained, and the standardized data sequence is obtained through data preprocessing. The main characteristics are extracted from the standardized data sequence by principal component analysis to obtain the characteristic vector. The multivariate linear regression model is constructed according to the characteristic vector, and the regression coefficient is determined. If the significance of the regression coefficient is greater than the preset threshold, the corrected trend sequence is calculated by the linear regression formula to obtain the preliminary trend sequence. The residual distribution characteristics are obtained by residual analysis on the preliminary trend sequence. The regression model parameters are adjusted according to the residual distribution characteristics to obtain the optimized trend sequence. The consistency of the optimized trend sequence and the original sequence is judged by comparison, and the final corrected trend sequence is obtained.

[0041] S109, according to the corrected trend sequence, the Z-score algorithm is used to calculate the temperature rise rate and duration, and the formula is: Z = (Tr - μT) / σT, Tr is the corrected trend sequence, μT and σT are the mean and standard deviation respectively, the abnormal trend quantitative index sequence is generated, and the preset threshold Tz is compared, Tz is calculated based on the historical data distribution, and the refrigerator load temperature control system abnormal point sequence is generated.

[0042] The corrected temperature trend sequence Tr is obtained, the real-time temperature data is collected from the refrigerator load temperature control system sensor, and the corrected temperature trend sequence is obtained through sliding window smoothing processing. The Z-score algorithm is used to calculate the corrected temperature trend sequence Tr, and the abnormal trend quantitative index sequence of temperature change rate and duration is generated. The distribution characteristics are extracted from the historical temperature data, the preset threshold Tz is calculated through Gaussian distribution fitting, and the comparison benchmark of the abnormal trend quantitative index is determined. If any Z value in the abnormal trend quantitative index sequence is greater than the preset threshold Tz, the point is marked as an abnormal point, and the preliminary abnormal point sequence is generated. Through time series analysis, the preliminary abnormal point sequence is clustered, and the time-continuous abnormal points are merged to obtain the optimized abnormal point sequence. According to the optimized abnormal point sequence, the decision tree algorithm is used to classify the load temperature control state corresponding to the abnormal point, and to judge whether the abnormal point triggers the control system alarm. Through the association analysis of the abnormal point sequence and the control state, the running parameters of the refrigerator load temperature control system are updated, and the final control instruction sequence is generated.

[0043] For example, the refrigeration chamber sensor records a temperature sequence of 4.2, 4.5, 4.3, 4.7, and 4.1 degrees Celsius, and the sliding window takes the average to obtain 4.36 degrees Celsius as a point of Tr. This smoothing process helps generate a stable trend sequence, facilitating subsequent analysis.

[0044] Specifically, the Z-score algorithm is used to quantify the rate and duration of temperature change anomalies. Assuming that the mean of the T_r sequence is 4.5 degrees Celsius and the standard deviation is 0.3 degrees Celsius, if a point T_r value is 5.1 degrees Celsius, the Z value is (5.1-4.5) / 0.3=2. This Z value reflects the degree of temperature deviation from the normal range.

[0045] It should be noted that the Z-score algorithm can intuitively compare the degree of anomaly at different time points by standardizing the data, providing a quantitative basis for anomaly detection.

[0046] The distribution characteristics of historical temperature data can be extracted by Gaussian distribution fitting. Assuming that historical data shows that the refrigeration chamber temperature follows a Gaussian distribution with a mean of 4.5 degrees Celsius and a standard deviation of 0.3 degrees Celsius, the preset threshold Tz can be set to 2, meaning that points exceeding 2 standard deviations are considered abnormal.

[0047] For example, points with a Z value of 2.5 will be marked as abnormal points, generating a preliminary abnormal point sequence. This method ensures the reliability of anomaly detection through statistical characteristics.

[0048] For example, clustering in time series analysis can use a time proximity-based clustering method. Assuming that the preliminary abnormal point sequence contains time points t1, t2, and t3, where t1 and t2 are separated by 1 minute and t3 is separated by 10 minutes, t1 and t2 are merged into one abnormal event after clustering. This merging reduces scattered abnormal points and improves the coherence of abnormal trends, facilitating subsequent analysis.

[0049] Specifically, the decision tree algorithm can be used to classify the load temperature regulation state corresponding to the abnormal points. Assuming that the abnormal points occur when the freezer temperature rapidly rises to -15 degrees Celsius, the decision tree determines whether to trigger high-power operation of the compressor based on features such as temperature change rate and duration.

[0050] If the change rate exceeds 0.5 degrees Celsius / minute and lasts for 3 minutes, it is classified as a state that requires an alarm. This classification method can accurately identify the regulation needs.

[0051] The association analysis of abnormal point sequence and regulation state can be achieved through statistical correlation analysis. Assuming that the abnormal points are concentrated in the refrigeration chamber night operation period, the analysis finds that they are related to load increase, and the night compressor operation frequency can be adjusted, such as from 50% to 60%. The final regulation instruction sequence may include increasing the refrigeration power or adjusting the fan speed. This association analysis ensures that the regulation instruction is highly matched with the actual anomaly.

[0052] The above descriptions of the embodiments are only used to help understand the technical solutions of the present application and the core ideas thereof; it should be understood by those skilled in the art that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A refrigerator load temperature regulation system, characterized by, The method comprises the following steps: Obtain the load data in the past 24 hours in the historical temperature curve, and obtain an initial load data set by using a time window division method; Process the initial load data set by using an STL time series decomposition method to separate periodic components, trend components, and residual components, and obtain decomposed time series features; Extract the periodic components and trend components from the decomposed time series features, analyze the frequency characteristics of the periodic components by using a Fourier transform, and determine periodic characteristics; Smooth the trend components by using a moving average method to obtain a smooth trend sequence; If the fluctuation amplitude of the smooth trend sequence exceeds a preset threshold, perform secondary smoothing processing on the smooth trend sequence by using an exponential smoothing method to obtain a stable trend sequence; Fill in the missing data in the stable trend sequence and the periodic characteristics by using a linear interpolation method to generate a load mode reference sequence; Obtain real-time humidity fluctuation data, calculate a humidity weight coefficient by using a humidity physical influence coefficient table through a lookup method, and generate a weight value; Fuse the load mode reference sequence and the real-time humidity fluctuation data by using a linear weighted fusion formula to generate an initial synthesized temperature sequence; Smooth the initial synthesized temperature sequence, fill in the missing data by using a linear interpolation method, and obtain a complete temperature sequence; Analyze the frequency characteristics of the complete temperature sequence by using a Fourier transform to determine periodic characteristics; Perform secondary processing on the complete temperature sequence by using an exponential smoothing method to obtain a stable synthesized temperature sequence; Align the timestamps of the stable synthesized temperature sequence and real-time temperature curve by using a time alignment method to obtain an aligned temperature data set; Calculate the deviation of the aligned temperature data set by using a mean square error formula to obtain a mean square error value; If the mean square error value exceeds a preset threshold calculated based on the variance of historical data, generate a temperature deviation sequence; Calculate the alignment path of the stable synthesized temperature sequence and the real-time temperature curve by using a dynamic time warping algorithm to generate a temperature deviation starting point sequence.

2. The refrigerator load temperature regulation system according to claim 1, wherein the linear weighted fusion formula is used to fuse the load mode reference sequence and the real-time humidity fluctuation data to generate an initial synthesized temperature sequence, comprising: processing the load mode reference sequence and the real-time humidity fluctuation data by using a preset time window division method to obtain an initial fusion data set; calculating a humidity weight coefficient by using a lookup method according to a humidity physical influence coefficient table to generate a weight value; calculating the load mode reference sequence and the real-time humidity fluctuation data in the initial fusion data set by using a linear weighted fusion formula to generate an initial synthesized temperature sequence; if the fluctuation amplitude of the initial synthesized temperature sequence exceeds a preset threshold, performing smoothing processing on the initial synthesized temperature sequence by using a moving average method to obtain a smooth temperature sequence.

3. The refrigerator load temperature regulation system according to claim 2, wherein the Fourier transform is used to analyze the frequency characteristics of the complete temperature sequence to determine periodic characteristics, comprising: ​ ​ performing a fast Fourier transform on the complete temperature sequence to generate frequency spectrum data; extracting a main frequency component from the frequency spectrum data to determine a periodicity characteristic of the complete temperature sequence; calculating a period length and an amplitude characteristic of the complete temperature sequence according to the periodicity characteristic to generate a periodicity characteristic data set.

4. The refrigerator load temperature regulation system according to claim 1, wherein the dynamic time warping algorithm is used to calculate an alignment path of the stable synthetic temperature sequence and the real-time temperature curve to generate a temperature deviation starting point sequence, including: extracting temperature sequence points from the stable synthetic temperature sequence and the real-time temperature curve; if the number of temperature sequence points is inconsistent, then the shorter sequence is padded through linear interpolation to obtain a uniform length sequence point set; using the dynamic time warping algorithm to calculate an Euclidean distance matrix of the sequence point set to generate a distance matrix; calculating an optimal matching path based on the distance matrix through a dynamic programming method to determine a sequence alignment path; extracting corresponding point pairs of the stable synthetic temperature sequence and the real-time temperature curve according to the sequence alignment path to generate a temperature deviation starting point sequence.

5. The refrigerator load temperature regulation system according to claim 4, wherein the mean square error formula is used to calculate a deviation of the aligned temperature data set to obtain a mean square error value, including: calculating a deviation of the stable synthetic temperature sequence and the real-time temperature curve in the aligned temperature data set through the mean square error formula to obtain a mean square error value; if the mean square error value exceeds a preset threshold value calculated based on historical data variance, then performing a difference calculation on the stable synthetic temperature sequence and the real-time temperature curve to generate a temperature deviation sequence; using a local weighted regression method to smooth the temperature deviation sequence to obtain a smoothed deviation sequence.

6. The refrigerator load temperature regulation system according to claim 5, wherein the local weighted regression method is used to smooth the temperature deviation sequence to obtain a smoothed deviation sequence, including: calculating a weighted regression coefficient of the temperature deviation sequence through a local weighted regression method to generate a smoothed deviation sequence; using an adaptive threshold segmentation method for the smoothed deviation sequence to extract a significant deviation region to generate a significant deviation data set; using a K-means clustering algorithm for the significant deviation data set to classify the deviation data in the significant deviation data set to generate a deviation classification result.

7. The refrigerator load temperature regulation system according to claim 6, wherein the real-time humidity fluctuation data is obtained using a humidity physical influence coefficient table to calculate a humidity weight coefficient through a lookup method to generate a weight value, including: extracting the real-time humidity fluctuation data from a sensor collection system to generate a humidity data sequence; using a lookup method according to a humidity physical influence coefficient table to match a humidity value of the humidity data sequence to generate a corresponding humidity weight coefficient; calculating a weight value of the real-time humidity fluctuation data through the humidity weight coefficient to generate a weight data set. ​ ​ ​ ​ 8. The refrigerator load temperature regulation system according to claim 7, wherein the initial load dataset is processed by using an STL time series decomposition method to separate periodic components, trend components and residual components, and to obtain decomposed time series features, including: the initial load dataset is decomposed by using an STL time series decomposition method to generate periodic components, trend components and residual components; for the periodic components, Fourier transform is used to extract frequency features and generate periodic feature datasets; for the trend components, a moving average method is used to generate a smooth trend sequence.

9. The refrigerator load temperature regulation system according to claim 8, wherein the significant deviation dataset is classified by using a K-means clustering algorithm to generate deviation classification results, including: the significant deviation dataset is classified by using a K-means clustering algorithm based on feature vectors of the significant deviation dataset to generate deviation classification results; for the deviation classification results, statistical features of each class of deviation are extracted to generate deviation feature datasets; distribution characteristics of each class of deviation are determined based on the deviation feature datasets to generate deviation distribution feature sets. ​ ​

Images