A refrigerator load temperature regulation system

Abstract

The application relates to the field of refrigerator temperature regulation, and specifically relates to a refrigerator load temperature regulation system. Load data is obtained from a historical temperature curve, an STL time sequence decomposition method is adopted to extract periodic characteristics and trend characteristics, and a load mode benchmark sequence is generated. From real-time compressor energy consumption data, a CUSUM variable point detection algorithm is adopted to segment the sequence, a plurality of independent energy consumption sequences are generated, multi-level data fusion and dynamic analysis are carried out, the accuracy and stability of refrigerator load temperature regulation are significantly improved, and the energy consumption abnormality and temperature fluctuation risk are effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of refrigerator temperature control, specifically a refrigerator load temperature control system. Background Technology

[0002] A complex technical challenge exists in refrigerator load temperature control systems. 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 periodicity and trend characteristics. Then, using methods such as Fourier transform and exponential smoothing, a load pattern baseline sequence is generated. Finally, the system needs to fuse this baseline sequence with real-time data to generate a composite temperature sequence.

[0003] The system faces the challenge of accurately detecting temperature deviations. It needs to compare synthetic temperature sequences with real-time temperature curves and use a dynamic time warping algorithm to identify the starting point of the deviation. Simultaneously, the system also needs to process compressor energy consumption data, using the CUSUM algorithm and an adaptive time window mechanism to generate continuous energy consumption sequences.

[0004] Another key challenge for the system is determining the consistency between temperature and energy consumption trends. This requires calculating correlation coefficients and, if necessary, performing multivariate linear regression to correct for trends. Finally, the system needs to quantify abnormal trends, generate indicator sequences, and compare them with preset thresholds to identify outliers. This process involves the collaborative work of multiple complex algorithms, requiring a balance between real-time data processing and accurate anomaly detection. Summary of the Invention

[0005] The purpose of this invention is to provide a refrigerator load temperature control system to solve the technical problems mentioned in the background art.

[0006] Based on the above ideas, the present invention provides the following technical solution:

[0007] Load data for the past 24 hours was obtained from historical temperature curves. The STL time series decomposition method was used to extract periodic and trend features to generate a load pattern baseline sequence.

[0008] Based on the load pattern baseline sequence and combined with real-time humidity fluctuation data, a composite temperature sequence is generated through linear weighted fusion. The weights are calculated based on the physical influence coefficient of humidity on temperature, as shown in the formula:

[0009] St = wh*Ht + (1 - wh)*Lt,

[0010] St represents the synthesized temperature sequence, Ht represents the humidity fluctuation data, Lt represents the load pattern baseline sequence, and wh represents the humidity weight;

[0011] The synthesized temperature sequence is compared with the real-time temperature curve, and the mean square error (MSE) is calculated using the following formula:

[0012] MSE = Σ(St - Rt)^2 / n,

[0013] St is the synthesized temperature sequence, Rt is the real-time temperature curve, and n is the sequence length. If MSE exceeds the preset threshold Tmse, Tmse is calculated based on the variance of historical data, and a temperature deviation sequence is generated.

[0014] For the temperature deviation sequence, a dynamic time warping algorithm is used to calculate the alignment path between the synthesized temperature sequence and the real-time temperature curve. The formula is as follows:

[0015] DTW(S, R) = min(Σd(Si, Rj)),

[0016] d represents the Euclidean distance, and Si and Rj are points in the synthesized temperature sequence and the real-time temperature curve, respectively, generating a sequence of starting points for temperature deviation.

[0017] From real-time compressor energy consumption data, the CUSUM change point detection algorithm is used to segment the sequence, with the following formula:

[0018] Sk = max(0, S{k-1} + (Ek - μ0)),

[0019] Sk is the cumulative sum, Ek is the energy consumption data, and μ0 is the average energy consumption, generating multiple independent energy consumption sequences;

[0020] For multiple independent energy consumption sequences, an adaptive time window sliding mechanism is adopted. The window size is calculated based on the energy consumption fluctuation frequency Fe, and the formula is as follows:

[0021] W = 1 / Fe,

[0022] The bridged continuous energy consumption sequence is generated by weighted averaging, with the weighting formula being:

[0023] wi = 1 / σi^2,

[0024] wi is the weight within the window, and σi is the local energy consumption variance;

[0025] The Pearson correlation coefficient was calculated between the bridged continuous energy consumption sequence and the synthesized temperature sequence, using the following formula:

[0026] ρ = cov(Ec, St) / (σEc*σSt),

[0027] E_c is the continuous energy consumption sequence after bridging, St is the synthesized temperature sequence, cov is the covariance, σ is the standard deviation. If ρ is lower than the preset threshold Tρ, which is calculated based on historical data, then an inconsistent trend sequence is generated.

[0028] For inconsistent trend sequences, multivariate linear regression is used to fuse humidity fluctuation data and multiple independent energy consumption sequences. The formula is as follows:

[0029] Tr = β0 + β1*Ht + β2*Em,

[0030] Tr is the corrected trend sequence, Ht is the humidity fluctuation data, Em is the multi-segment independent energy consumption sequence, and β is the regression coefficient, which generates the corrected trend sequence.

[0031] Based on the corrected trend sequence, the Z-score algorithm is used to calculate the rate of temperature increase and duration, as follows:

[0032] Z = (Tr - μT) / σT,

[0033] Tr is the corrected trend sequence, μT and σT are the mean and standard deviation, respectively, to generate an abnormal trend quantitative index sequence, which is compared with the preset threshold Tz. Tz is calculated based on the distribution of historical data to generate an abnormal point sequence of the refrigerator load temperature control system.

[0034] The technical solution of the present invention may include the following beneficial effects:

[0035] This invention discloses a method for detecting abnormal load temperature control in refrigerators based on multi-source data fusion. Addressing the issue of inconsistent load temperature fluctuations and energy consumption trends during refrigerator operation, this method integrates historical temperature, real-time humidity, and compressor energy consumption data to construct an intelligent control system. First, this invention extracts a load pattern baseline sequence through STL decomposition, generates a synthetic temperature sequence by combining humidity weights, and calculates the mean square error by comparing it with the real-time temperature curve to identify temperature deviations. Then, a dynamic time warping algorithm is used to generate an alignment path to accurately locate the starting point of the deviation. For energy consumption data, this invention uses CUSUM variable point detection to segment the sequence, combines it with an adaptive time window sliding mechanism to generate a continuous energy consumption sequence, and analyzes its correlation with the synthetic temperature sequence using Pearson correlation coefficient. For sequences with inconsistent trends, this invention fuses humidity and energy consumption data through multivariate linear regression to generate a corrected trend sequence, and uses the Z-score algorithm to quantify abnormal trends, generating a sequence of control anomaly points. Through multi-level data fusion and dynamic analysis, this invention significantly improves the accuracy and stability of refrigerator load temperature control, effectively reducing the risks of abnormal energy consumption and temperature fluctuations. Attached Figure Description

[0036] Figure 1 This is a flowchart of a refrigerator load temperature control system according to the present invention.

[0037] Figure 2 This is a schematic diagram of a refrigerator load temperature control system according to the present invention.

[0038] Figure 3 This is another schematic diagram of a refrigerator load temperature control system according to the present invention. Detailed Implementation

[0039] Example

[0040] like Figure 1-3 This embodiment of a refrigerator load temperature control method may specifically include:

[0041] S101. Obtain the load data of the past 24 hours from the historical temperature curve, and use the STL time series decomposition method to extract periodic and trend features to generate a load pattern benchmark sequence.

[0042] Load data for the past 24 hours from historical temperature curves is acquired, and an initial load dataset is obtained using a pre-defined time window division method. The initial load dataset is then processed using the STL time series decomposition method to separate periodic, trend, and residual components, yielding the decomposed time series features. For these decomposed features, the periodic and trend components are extracted, and Fourier transform analysis is used to analyze the frequency characteristics of the periodic components, identifying the periodicity. Based on the periodicity and trend components, a moving average method is used to smooth the trend components, generating a smoothed trend series. If the fluctuation amplitude of the smoothed trend series exceeds a preset threshold, a second smoothing process is performed using exponential smoothing to obtain a stable trend series. By combining the periodic features and the stable trend series, linear interpolation is used to fill in missing data, generating a complete load pattern baseline series.

[0043] S102. Based on the load pattern baseline sequence and combined with real-time humidity fluctuation data, a composite temperature sequence is generated through linear weighted fusion. The weights are calculated based on the physical influence coefficient of humidity on temperature, using the following formula:

[0044] St = wh*Ht + (1 - wh)*Lt,

[0045] St represents the synthesized temperature sequence, Ht represents the humidity fluctuation data, Lt represents the load pattern baseline sequence, and wh represents the humidity weight;

[0046] Real-time humidity fluctuation data and load pattern baseline sequences are acquired, and an initial fusion dataset is obtained through a preset time window segmentation method. Humidity weighting coefficients are calculated using a lookup method based on a humidity physical influence coefficient table to obtain weight values. For the initial synthesized temperature sequence, if the sequence fluctuation amplitude exceeds a preset threshold, a moving average method is used for smoothing to obtain a smoothed temperature sequence. Based on the smoothed temperature sequence, missing data portions are filled in using a linear interpolation method to obtain a complete temperature sequence. For the complete temperature sequence, Fourier transform analysis is used to determine its frequency characteristics and periodicity. Based on the periodicity and the complete temperature sequence, an exponential smoothing method is used to perform a secondary processing on the sequence to obtain a stable synthesized temperature sequence.

[0047] For example, in practical power load analysis scenarios, acquiring real-time humidity fluctuation data and load pattern baseline sequences are crucial steps in optimizing temperature load forecasting. Humidity fluctuation data is typically collected hourly by sensors; for instance, the humidity value of a certain area over the past 24 hours may fluctuate between 60% and 80%. The load pattern baseline sequence is derived from the decomposition of historical temperature curves, reflecting the periodicity and trend characteristics of the load. Suppose that the load baseline sequence for a certain day shows a peak of 5000 kW at noon and a trough of 3000 kW at 4 AM. A preset time window can be selected with a 1-hour unit, aligning the humidity data and load sequence by time to form an initial fused dataset.

[0048] Specifically, the humidity physical influence coefficient table is constructed based on the law of environmental impact on load. High humidity may increase the operating load of equipment such as air conditioners. Assuming the lookup table shows that a weighting coefficient of 0.6 corresponds to 70% humidity, while a weighting coefficient of 0.7 corresponds to 80% humidity, the composite temperature series is calculated using a linear weighted fusion formula. For example, at a certain moment, if the humidity data is 75%, the load baseline value is 4500 kW, and the weighting coefficient is 0.65, then the composite value reflects the comprehensive impact of humidity on the load.

[0049] If the initial synthesized temperature sequence fluctuates significantly, for example, if a segment of the sequence jumps from 4500 kW to 5500 kW within one hour, exceeding a preset threshold by 10%, a moving average method is used for smoothing. A 3-hour window is used to calculate the hourly average, generating a smoothed temperature sequence to reduce the impact of short-term fluctuations.

[0050] For example, to address the problem of missing data in smooth temperature sequences, if data for a certain hour is lost due to sensor malfunction, a linear interpolation method can be used to estimate the missing point as 4550 kW based on the 4500 kW and 4600 kW of the two hours before and after, thus obtaining the complete temperature sequence.

[0051] Specifically, the Fourier transform is used to analyze the frequency characteristics of a complete temperature series. Assuming the analysis reveals a 24-hour diurnal cycle in the series, it indicates that the load is significantly affected by diurnal variations. This periodicity can be used to predict future load patterns.

[0052] Exponential smoothing performs a secondary processing on the complete temperature series. Assuming a series exhibits significant fluctuations during peak periods, exponential smoothing assigns higher weights to recent data, for example, a smoothing factor of 0.8, generating a stable, composite temperature series. This series better reflects long-term trends, reduces short-term noise interference, and provides a reliable basis for load forecasting.

[0053] Understandably, each step of the above method is closely linked to the relationship between humidity and load, ensuring that the generated stable synthetic temperature sequence can accurately reflect the impact of environmental factors on the load, thus providing support for power dispatch and energy management.

[0054] S103. Compare the synthesized temperature sequence with the real-time temperature curve and calculate the mean square error (MSE). The formula is MSE = Σ(S_t - R_t)^2 / n, where S_t is the synthesized temperature sequence, R_t is the real-time temperature curve, and n is the sequence length. If the MSE exceeds the preset threshold T_mse, which is calculated based on the variance of historical data, a temperature deviation sequence is generated.

[0055] Acquire the synthesized temperature sequence and real-time temperature curve, and align the timestamps of the two sequences using a time alignment method to obtain the aligned temperature dataset. The mean square error formula is then used.

[0056] MSE = Σ(St - Rt)^2 / n,

[0057] St is the synthesized temperature sequence, Rt is the real-time temperature curve, and n is the sequence length. If MSE exceeds the preset threshold Tmse, and Tmse is calculated based on the variance of historical data, then a temperature deviation sequence is generated. Then, the difference between the synthesized temperature sequence and the real-time temperature curve is calculated to generate a temperature deviation sequence.

[0058] Based on the temperature deviation sequence, a local weighted regression method is used to smooth the deviation sequence, resulting in a smoothed deviation sequence. For the smoothed deviation sequence, an adaptive threshold segmentation method is used to extract salient deviation regions, resulting in a salient deviation dataset. Based on the salient deviation dataset, a K-means clustering algorithm is used to classify the deviation data, yielding deviation classification results. Using these classification results, statistical features of each deviation class are extracted to generate a deviation feature dataset.

[0059] For example, after obtaining the synthetic temperature sequence and real-time temperature profiles, time alignment is a crucial step in ensuring the accuracy of subsequent analyses. Time alignment methods can be based on linear interpolation of timestamps or nearest-neighbor matching.

[0060] If the timestamp discrepancy between the two sequences is small, nearest neighbor matching can be used, directly selecting the points closest in time for pairing. In practical scenarios, such as temperature monitoring of industrial equipment, this method ensures the consistency of data points over time, facilitating subsequent deviation analysis. When calculating the mean squared error (MSE), the MSE formula is used to quantify the difference between the synthesized temperature sequence and the real-time temperature curve.

[0061] In factory cooling system monitoring, the synthesized temperature sequence may be generated based on humidity and load patterns, while the real-time temperature curve comes from sensor data. If the MSE value is 5.2, while the threshold Tmse calculated based on historical data variance is 4.0, it indicates a large deviation, requiring further processing. This method of quantifying deviation can intuitively reflect the accuracy of the synthesized sequence, providing a basis for subsequent deviation correction. For cases with large deviations, generating a temperature deviation sequence is a crucial step.

[0062] By subtracting the values ​​point by point, the difference between the synthesized temperature sequence and the real-time temperature curve is obtained, forming a deviation sequence. If the synthesized temperature at a certain time point is 25.0°C and the real-time temperature is 26.5°C, then the deviation is -1.5°C. This deviation sequence clearly shows the trend of difference between the two sequences.

[0063] In industrial settings, deviation sequences can be used to identify abnormal temperature fluctuations during equipment operation. Locally weighted regression methods can effectively reduce the impact of noise when smoothing deviation sequences.

[0064] For example, when processing cooling system deviation sequences, a sliding window, such as five time points, can be set, and a smoothed value can be generated for each point using weighted regression. Data closer to the center point within the window is given higher weight, while edge points are given lower weight. This method can preserve the trend characteristics of the deviation sequence while reducing the interference of random fluctuations, making it suitable for temperature monitoring scenarios requiring high stability. For the extraction of significant deviation regions, an adaptive threshold segmentation method can dynamically determine the deviation threshold.

[0065] Based on the statistical distribution of the deviation sequence, a threshold is set as the mean plus twice the standard deviation, and regions exceeding the threshold are extracted. Assuming the mean of the deviation sequence is 0.5°C and the standard deviation is 0.3°C, the threshold is 1.1°C; points exceeding this value are marked as significant deviations. This method can effectively locate abnormal temperature areas in practical applications, such as overheated equipment. In deviation classification, the K-means clustering algorithm can divide the deviation data into different categories.

[0066] The deviation sequence includes positive deviations, negative deviations, and near-zero deviations. By setting K=3, the clustering algorithm can classify the data into three categories: excessively high temperatures, excessively low temperatures, and normal range. The classification results can help analyze the patterns of deviations; for example, positive deviations may be related to excessive equipment load.

[0067] When extracting statistical features for deviation classification, the mean, variance, and range of each class can be calculated. These features reflect the concentration and magnitude of the deviations. In industrial scenarios, this feature extraction method can be used to optimize temperature control strategies and improve system stability.

[0068] S104. For the temperature deviation sequence, the dynamic time warping algorithm is used to calculate the alignment path between the synthesized temperature sequence and the real-time temperature curve. The formula is:

[0069] DTW(S, R) = min(Σd(Si, Rj)),

[0070] d represents the Euclidean distance, and Si and Rj are points in the synthesized temperature sequence and the real-time temperature curve, respectively, generating a sequence of starting points for temperature deviation.

[0071] Acquire synthetic and real-time temperature sequences, and extract temperature sequence points from them. If the number of temperature sequence points is inconsistent, the shorter sequence is padded using linear interpolation to obtain a sequence point set of uniform length. A dynamic time warping algorithm is used to calculate the Euclidean distance matrix between the padded synthetic and real-time temperature sequences, obtaining the distance matrix. Using dynamic programming, the optimal matching path is calculated based on the distance matrix to determine the sequence alignment path. Based on the sequence alignment path, corresponding point pairs between the synthetic and real-time temperature sequences are extracted to generate a temperature deviation sequence. For the temperature deviation sequence, the absolute value of the deviation for each corresponding point pair is calculated to obtain the starting point sequence of the deviation sequence. If the absolute value of the deviation of the starting point sequence exceeds a preset threshold, the deviation sequence is smoothed to obtain a smoothed deviation sequence. The final temperature deviation starting point sequence is determined using the smoothed deviation sequence.

[0072] S105. From the real-time compressor energy consumption data, the CUSUM change point detection algorithm is used to segment the sequence. The formula is:

[0073] Sk = max(0, S{k-1} + (Ek - μ0)),

[0074] Sk is the cumulative sum, Ek is the energy consumption data, and μ0 is the average energy consumption, generating multiple independent energy consumption sequences;

[0075] Real-time energy consumption data of the compressor is acquired by using a sensor acquisition system to extract continuous energy consumption sequences from the equipment, resulting in a real-time energy consumption sequence. The mean of the real-time energy consumption sequence is calculated using statistical methods to generate an energy consumption average and determine an initial reference baseline. The real-time energy consumption sequence is processed using the CUSUM algorithm to calculate the cumulative sum sequence, obtaining basic data for change point detection. If the cumulative sum Sk exceeds a preset threshold, a change point is identified, its location is recorded, and a change point sequence is generated. The real-time energy consumption sequence is segmented based on the change point sequence, generating multiple independent energy consumption sequences, resulting in segmented energy consumption segments. The K-means clustering algorithm is used to classify the segmented energy consumption segments, determining the operating mode of each segment based on energy consumption characteristics, and generating mode classification results. The energy consumption change trend of each segment is analyzed based on the mode classification results to generate energy consumption change characteristics and determine the operating status description.

[0076] S106. For multiple independent energy consumption sequences, an adaptive time window sliding mechanism is adopted. The window size is calculated based on the energy consumption fluctuation frequency Fe, and the formula is:

[0077] W = 1 / F_e,

[0078] The continuous energy consumption sequence after bridging is generated by weighted averaging. The weight formula is wi = 1 / σi^2, where wi is the weight within the window and σi is the local energy consumption variance.

[0079] Initial energy consumption data is determined by collecting multiple independent energy consumption sequences. If the energy consumption sequence contains missing values, linear interpolation is used to fill in the missing values, resulting in a complete energy consumption sequence. Based on the complete energy consumption sequence, the energy consumption fluctuation frequency is calculated using the Fast Fourier Transform algorithm to determine the size of the time window. A sliding mechanism is used to move the data along the complete energy consumption sequence in units of time window W, acquiring local energy consumption data and calculating the local variance σi to obtain the weights of the data points.

[0080] w_i=1 / σ_i^2,

[0081] A weighted average method is used to generate a bridged continuous energy consumption sequence based on the data point weights wi for local energy consumption data within each time window. If outliers exist in the continuous energy consumption sequence, a median filtering algorithm is used for smoothing to obtain a smoothed energy consumption sequence. Based on the smoothed energy consumption sequence, an adaptive mechanism is used to dynamically adjust the time window size W, recalculate the local variance and weights, and generate an optimized continuous energy consumption sequence.

[0082] S107. Calculate the Pearson correlation coefficient between the bridged continuous energy consumption sequence and the synthesized temperature sequence. The formula is:

[0083] ρ = cov(Ec, St) / (σEc*σSt),

[0084] Ec is the continuous energy consumption sequence after bridging, St is the synthesized temperature sequence, cov is the covariance, and σ is the standard deviation. If ρ is lower than the preset threshold Tρ, which is calculated based on historical data, then an inconsistent trend sequence is generated.

[0085] After bridging, the continuous energy consumption sequence Ec and the synthesized temperature sequence St are obtained and standardized to obtain the standardized energy consumption sequence Ec' and the standardized temperature sequence St'. The covariance cov(Ec', St') is calculated using the standardized energy consumption sequence Ec' and the standardized temperature sequence St', yielding the covariance value. The standard deviation σEc' is calculated based on the standardized energy consumption sequence Ec', and the standard deviation σSt' is calculated based on the standardized temperature sequence St', yielding two standard deviation values. The Pearson correlation coefficient ρ is calculated by dividing the covariance value cov(Ec', St') by the product of the standard deviations σEc' and σSt', yielding the correlation coefficient value. A preset threshold Tρ is obtained from historical data statistics. If the correlation coefficient ρ is lower than Tρ, the sequence trend is determined to be inconsistent, and a trend inconsistency sequence identifier is generated. For the trend inconsistency sequence identifier, the corresponding time period data in Ec and St are extracted to generate a trend inconsistency sequence dataset. By using a dataset of inconsistent trend sequences, time series analysis methods are employed to determine the specific time points and magnitudes of change in the inconsistent sequences, thus obtaining the final analysis results.

[0086] S108. For inconsistent trend sequences, multivariate linear regression is used to fuse humidity fluctuation data and multiple independent energy consumption sequences. The formula is:

[0087] Tr = β0 + β1* H_t + β2* Em,

[0088] Tr is the corrected trend sequence, Ht is the humidity fluctuation data, Em is the multi-segment independent energy consumption sequence, and β is the regression coefficient, which generates the corrected trend sequence.

[0089] Humidity fluctuation data and multiple energy consumption sequences were acquired, and standardized data sequences were obtained through data preprocessing. Principal component analysis was used to extract key features from the standardized data sequences, resulting in eigenvectors. A multivariate linear regression model was constructed based on the eigenvectors to determine the regression coefficients. If the significance of the regression coefficients exceeded a preset threshold, a revised trend sequence was calculated using the linear regression formula to obtain a preliminary trend sequence. Residual analysis was performed on the preliminary trend sequence to obtain the residual distribution characteristics. The regression model parameters were adjusted based on the residual distribution characteristics to obtain an optimized trend sequence. By comparing the optimized trend sequence with the original sequence, trend consistency was assessed, resulting in the final revised trend sequence.

[0090] S109. Based on the corrected trend sequence, the Z-fraction algorithm is used to calculate the rate of temperature increase and duration. The formula is:

[0091] Z = (Tr - μT) / σT,

[0092] Tr is the corrected trend sequence, μT and σT are the mean and standard deviation, respectively, to generate an abnormal trend quantitative index sequence, which is compared with the preset threshold Tz. Tz is calculated based on the distribution of historical data to generate an abnormal point sequence of the refrigerator load temperature control system.

[0093] The corrected temperature trend sequence Tr is obtained by collecting real-time temperature data from the refrigerator load temperature control system sensors and smoothing it using a sliding window. The Z-score algorithm is then used to calculate the corrected temperature trend sequence Tr, generating a sequence of abnormal trend quantification indicators for temperature change rate and duration. Distribution features are extracted from historical temperature data, and a preset threshold Tz is calculated using Gaussian distribution fitting to determine the benchmark for the abnormal trend quantification indicators. If any Z-value in the abnormal trend quantification indicator sequence is greater than the preset threshold Tz, that point is marked as an anomaly, generating a preliminary anomaly point sequence. Time series analysis is used to cluster the preliminary anomaly point sequence, merging temporally consecutive anomalies to obtain an optimized anomaly point sequence. Based on the optimized anomaly point sequence, a decision tree algorithm is used to classify the load temperature control status corresponding to the anomalies and determine whether the anomalies trigger an alarm in the control system. Through correlation analysis between the anomaly point sequence and the control status, the operating parameters of the refrigerator load temperature control system are updated, generating the final control command sequence.

[0094] For example, if a sensor in a refrigerator compartment records a temperature sequence of 4.2, 4.5, 4.3, 4.7, and 4.1 degrees Celsius, and the average is taken through a sliding window, the result is 4.36 degrees Celsius, which is used as a point in the Tr curve. This smoothing process helps generate a stable trend sequence, facilitating subsequent analysis.

[0095] Specifically, the Z-score algorithm is used to quantify anomalies in the rate and duration of temperature changes. Assuming the mean of the T_r sequence is 4.5 degrees Celsius and the standard deviation is 0.3 degrees Celsius, if a point in the T_r sequence has a value of 5.1 degrees Celsius, then the Z-value is (5.1 - 4.5) / 0.3 = 2. This Z-value reflects the degree to which the temperature deviates from the normal range.

[0096] It should be noted that the Z-score algorithm, through standardized data, can intuitively compare the degree of anomaly at different time points, providing a quantitative basis for anomaly detection.

[0097] The distribution characteristics of historical temperature data can be extracted by fitting a Gaussian distribution. Assuming that historical data shows that the temperature in the cold storage room 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 outliers.

[0098] For example, points with a Z-value of 2.5 are marked as outliers, generating a preliminary outlier sequence. This method ensures the reliability of anomaly detection through statistical properties.

[0099] For example, clustering in time series analysis can employ a time proximity-based clustering method. Assuming the initial outlier sequence contains time points t1, t2, and t3, where t1 and t2 are 1 minute apart and t3 is 10 minutes apart, after clustering, t1 and t2 are merged into a single outlier event. This merging reduces scattered outliers, improves the coherence of the anomaly trend, and facilitates subsequent analysis.

[0100] Specifically, decision tree algorithms can be used to classify the load temperature control status corresponding to outliers. Suppose an outlier occurs when the freezer temperature rises rapidly to -15 degrees Celsius, the decision tree determines whether to trigger high-power operation of the compressor based on features such as the rate of temperature change and duration.

[0101] If the rate of change exceeds 0.5 degrees Celsius per minute and persists for 3 minutes, it is classified as a state requiring an alarm. This classification method can accurately identify control needs.

[0102] The correlation analysis between anomaly sequences and control states can be achieved through statistical correlation analysis. Assuming anomalies are concentrated during the nighttime operation of the refrigerator compartment, analysis reveals a correlation with increased load. The nighttime compressor operating frequency can be adjusted, for example, from 50% to 60%. The final control command sequence might include increasing cooling power or adjusting fan speed. This correlation analysis ensures a high degree of match between control commands and actual anomalies.

[0103] The above description of the embodiments is only for the purpose of helping to understand the technical solutions and core ideas of this application; those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A refrigerator load temperature control system, characterized in that, Includes the following steps: The load data for the past 24 hours is obtained from the historical temperature curve, and the initial load dataset is obtained by using the time window division method. The initial load dataset is processed using the STL time series decomposition method to separate the periodic component, trend component, and residual component, thereby obtaining the decomposed time series features. Extract the periodic and trend components from the decomposed time series features, and use Fourier transform to analyze the frequency characteristics of the periodic components to determine the periodic features. The trend components are smoothed using a moving average method to obtain a smoothed trend sequence; If the fluctuation range of the smoothed trend sequence exceeds a preset threshold, then the exponential smoothing method is used for secondary smoothing to obtain a stable trend sequence. By using linear interpolation, missing data portions in the stable trend sequence and the periodic features are filled in to generate a load pattern baseline sequence; Real-time humidity fluctuation data is obtained, and the humidity weighting coefficient is calculated by using the humidity physical influence coefficient table and a lookup method to generate weight values. A linear weighted fusion formula is used to fuse the load pattern baseline sequence and the real-time humidity fluctuation data to generate an initial synthetic temperature sequence. The initial synthesized temperature sequence is smoothed, and the missing data is filled in using a linear interpolation method to obtain a complete temperature sequence. The frequency characteristics of the complete temperature sequence were analyzed using Fourier transform to determine its periodicity. The complete temperature sequence was processed twice using an exponential smoothing method to obtain a stable synthetic temperature sequence. Obtain the real-time temperature curve, and align the timestamps of the stable synthetic temperature sequence and the real-time temperature curve using a time alignment method to obtain the aligned temperature dataset. The deviation of the aligned temperature dataset is calculated using the mean square error formula to obtain the mean square error value. If the mean square error value exceeds a preset threshold calculated based on the variance of historical data, a temperature deviation sequence is generated. A dynamic time warping algorithm is used to calculate the alignment path between the stable synthesized temperature sequence and the real-time temperature curve, generating a temperature deviation starting point sequence.

2. The refrigerator load temperature control system according to claim 1, characterized in that, The process employs a linear weighted fusion formula to fuse the load pattern baseline sequence and the real-time humidity fluctuation data to generate an initial synthetic temperature sequence, including: The load pattern baseline sequence and the real-time humidity fluctuation data are processed using a preset time window division method to obtain an initial fused dataset. Based on the table of physical influence coefficients of humidity, the humidity weight coefficient is calculated using a lookup method to generate the weight value; The initial synthetic temperature sequence is generated by calculating the load pattern baseline sequence and the real-time humidity fluctuation data in the initial fusion dataset using a linear weighted fusion formula. If the fluctuation range of the initial synthesized temperature sequence exceeds a preset threshold, a moving average method is used for smoothing to obtain a smoothed temperature sequence.

3. A refrigerator load temperature control system according to claim 2, characterized in that, The step of analyzing the frequency characteristics of the complete temperature sequence using Fourier transform to determine periodic features includes: Perform a fast Fourier transform on the complete temperature sequence to generate frequency spectrum data; Extract the main frequency components from the frequency spectrum data to determine the periodicity characteristics of the complete temperature sequence; Based on the periodicity characteristics, the period length and amplitude characteristics of the complete temperature sequence are calculated to generate a periodicity characteristic dataset.

4. A refrigerator load temperature control system according to claim 1, characterized in that, The process employs a dynamic time warping algorithm to calculate the alignment path between the stable synthesized temperature sequence and the real-time temperature curve, generating a temperature deviation starting point sequence, including: Extract the respective temperature sequence points from the stable synthesized temperature sequence and the real-time temperature curve; If the number of temperature sequence points is inconsistent, the shorter sequence is padded by linear interpolation to obtain a sequence point set of uniform length; The Euclidean distance matrix of the sequence point set is calculated using the dynamic time warping algorithm, and a distance matrix is ​​generated. The optimal matching path is calculated based on the distance matrix using dynamic programming to determine the sequence alignment path. Based on the sequence alignment path, the corresponding point pairs between the stable synthesized temperature sequence and the real-time temperature curve are extracted to generate a temperature deviation starting point sequence.

5. A refrigerator load temperature control system according to claim 4, characterized in that, The step of calculating the deviation of the aligned temperature dataset using the mean square error formula to obtain the mean square error value includes: The mean square error value is obtained by calculating the deviation between the stable synthesized temperature sequence and the real-time temperature curve in the aligned temperature dataset using the mean square error formula. If the mean square error value exceeds a preset threshold calculated based on the variance of historical data, then the difference between the stable synthesized temperature sequence and the real-time temperature curve is calculated to generate a temperature deviation sequence. The temperature deviation sequence is smoothed using a local weighted regression method to obtain a smoothed deviation sequence.

6. A refrigerator load temperature control system according to claim 5, characterized in that, The step of smoothing the temperature deviation sequence using a local weighted regression method to obtain a smoothed deviation sequence includes: The weighted regression coefficients of the temperature deviation sequence are calculated using a local weighted regression method to generate a smoothed deviation sequence. For the smoothed deviation sequence, an adaptive threshold segmentation method is used to extract the significant deviation regions and generate a significant deviation dataset. Based on the significant deviation dataset, the K-means clustering algorithm is used to classify the deviation data in the significant deviation dataset and generate deviation classification results.

7. A refrigerator load temperature control system according to claim 6, characterized in that, The process of acquiring real-time humidity fluctuation data involves using a humidity physical influence coefficient table, calculating humidity weighting coefficients through a lookup method, and generating weight values, including: The real-time humidity fluctuation data is extracted from the sensor acquisition system to generate a humidity data sequence; Based on the table of physical influence coefficients of humidity, a lookup method is used to match the humidity values ​​of the humidity data sequence and generate the corresponding humidity weight coefficients. The weight values ​​of the real-time humidity fluctuation data are calculated using the humidity weighting coefficient to generate a weighted dataset.

8. A refrigerator load temperature control system according to claim 7, characterized in that, The initial load dataset is processed using the STL time series decomposition method to separate the periodic component, trend component, and residual component, obtaining the decomposed time series features, including: The initial load dataset is decomposed using the 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 a periodic feature dataset. For the aforementioned trend components, a smoothed trend sequence is generated using a moving average method.

9. A refrigerator load temperature control system according to claim 8, characterized in that, The step of classifying the deviation data in the significant deviation dataset using the K-means clustering algorithm to generate deviation classification results includes: Using the K-means clustering algorithm, based on the feature vectors of the significant deviation dataset, the deviation data in the significant deviation dataset are classified to generate deviation classification results. Based on the deviation classification results, the statistical features of each type of deviation are extracted to generate a deviation feature dataset; Based on the deviation feature dataset, the distribution characteristics of each type of deviation are determined, and a deviation distribution feature set is generated.

Images