Freezer remote monitoring and fault prediction system

Abstract

The invention relates to the technical field of refrigeration equipment state monitoring and data analysis, and discloses a freezer remote monitoring and fault prediction system, which comprises the steps of intercepting operation data of a freezer in a door closing silence period, calculating an environment dew point temperature sequence and extracting a unit temperature difference load fluctuation ratio sequence of a refrigeration system; based on the time change rate of the environment dew point temperature sequence and the unit temperature difference load fluctuation ratio sequence, an instantaneous cross bispectrum coupling potential energy sequence representing the nonlinear modulation relation of the time change rate and the unit temperature difference load fluctuation ratio sequence is constructed; then, a two-dimensional phase space hysteresis track is constructed with the environment dew point temperature sequence as the horizontal axis and the instantaneous cross bispectrum coupling potential energy sequence as the longitudinal axis, and the cross bispectrum hysteresis entropy of the dew point driving type load volatility is calculated based on the probability density distribution of the track and the directed circulation area; and finally, judging an operation state and outputting a fault prediction result by verifying whether the entropy value is within a historical reference interval.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of refrigeration equipment state monitoring and data analysis, more specifically, it relates to a refrigerator remote monitoring and fault prediction system. BACKGROUND

[0002] In modern retail scenarios such as supermarkets and convenience stores, vertical refrigerated display cabinets with glass doors (commonly known as closed refrigerators) are widely used in order to balance product display and energy efficiency. In order to prevent condensation (fogging) on the surface of the glass door in a high-humidity environment due to the temperature being lower than the ambient dew point, such equipment is usually equipped with an anti-condensation heating system (ASH). Advanced anti-condensation control strategies usually adopt a dew point driven mode, that is, the real-time dew point temperature is calculated according to the temperature and humidity of the store environment, and the start-stop or power size of the heater is dynamically adjusted through pulse width modulation (PWM) and other methods, so as to reduce energy consumption as much as possible while ensuring glass clarity. With the development of Internet of Things technology, remote monitoring of such equipment has become an industry standard, mainly for real-time collection of operating data such as compressor power and cabinet temperature.

[0003] However, existing refrigerator remote monitoring systems mostly rely on static threshold alarms (such as cabinet temperature over-limit alarms) or simple first-order linear regression analysis (such as linear fitting of energy consumption and ambient temperature). Such traditional monitoring dimensions have significant blind spots when facing complex anti-condensation control and thermodynamic response systems. Specifically, because the condensation and evaporation of water film on the glass surface involves latent heat of phase change, and the heating control logic has a dead zone and hysteresis set by humans, the impact of ambient dew point changes on refrigeration system load is not a simple linear superposition, but a nonlinear modulation process with memory effect. When the anti-condensation system has a hidden fault (such as a stuck-open relay contact, a PWM control logic drift, or a change in glass surface thermal damping characteristics), the refrigeration compressor often compensates for the additional heat input by increasing the operating load, so that the cabinet temperature remains within the normal range. At this time, traditional monitoring methods based on temperature thresholds cannot detect abnormalities, resulting in high energy consumption or potential safety hazards (such as slippery floors) in the device under the sick running state. SUMMARY

[0004] The present application provides a refrigerator remote monitoring and fault prediction system, which solves the technical problems raised in the background art.

[0005] The present application provides a refrigerator remote monitoring and fault prediction system, comprising: A data preprocessing module for intercepting operating data of the refrigerator during a closed-door silent period; A dual-channel signal generation module for calculating an ambient dew point temperature sequence based on the operating data and extracting a unit temperature difference load fluctuation rate sequence of the refrigeration system; a data analysis module configured to construct a sequence of instantaneous cross-bispectrum coupling potential representing a nonlinear modulation relationship between a time rate of change of the ambient dew point temperature sequence and the sequence of unit temperature difference load fluctuation rate; a phase space entropy calculation module configured to construct a two-dimensional phase space hysteresis trajectory with the ambient dew point temperature sequence as a horizontal axis and the sequence of instantaneous cross-bispectrum coupling potential as a vertical axis, and calculate a cross-bispectrum hysteresis entropy of the dew point driven load fluctuation rate based on a probability density distribution and a directed flow area of the two-dimensional phase space hysteresis trajectory; a fault prediction output module configured to verify whether the cross-bispectrum hysteresis entropy of the dew point driven load fluctuation rate is within a historical reference interval, and output an ice cabinet running state according to the verification result.

[0006] The present application has the following advantages: the cross-bispectrum hysteresis entropy of the dew point driven load fluctuation rate is constructed to quantitatively analyze the nonlinear dynamic correlation between the anti-condensation control logic and the environmental thermal stress from macroscopic running data; compared with the conventional temperature threshold monitoring, the present application can achieve early identification when the anti-condensation system has been decoupled or failed while the cabinet temperature remains normal, and the cross-bispectrum analysis and phase space hysteresis flow calculation are introduced to effectively eliminate the interference of background noise such as environmental temperature drift, door sealing leakage and conventional compressor start-stop ripple, thereby significantly improving the accuracy and sensitivity of the hidden fault prediction of the ice cabinet under high humidity dynamic conditions. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 is a module diagram of an ice cabinet remote monitoring and fault prediction system of the present application. DETAILED DESCRIPTION

[0008] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is merely meant to provide a better understanding of the subject matter described herein and can be changed in function and arrangement without departing from the scope of the present description. Various processes or components can be omitted, substituted, or added according to desired implementations. Additionally, features described with respect to some examples can be combined in other examples.

[0009] As shown in Figure 1 , an ice cabinet remote monitoring and fault prediction system includes: a data preprocessing module configured to intercept running data of the ice cabinet during a closed-door silent period; a dual-channel signal generation module configured to calculate an ambient dew point temperature sequence based on the running data, and extract a sequence of unit temperature difference load fluctuation rate of the refrigeration system; a data analysis module configured to construct a sequence of instantaneous cross-bispectrum coupling potential representing a nonlinear modulation relationship between the time rate of change of the ambient dew point temperature sequence and the sequence of unit temperature difference load fluctuation rate; a phase space entropy calculation module configured to construct a two-dimensional phase space hysteresis trajectory with the ambient dew point temperature sequence as the horizontal axis and the sequence of instantaneous cross-bispectrum coupling potential as the vertical axis, and to calculate a cross-bispectrum hysteresis entropy of the dew point driven load fluctuation rate based on a probability density distribution and a directed current area of the two-dimensional phase space hysteresis trajectory; a fault prediction output module configured to verify whether the cross-bispectrum hysteresis entropy of the dew point driven load fluctuation rate is within a historical benchmark interval, and to output an operation state of the freezer accordingly.

[0010] It should be noted that the data preprocessing module needs to construct a thermodynamic observation window that can exclude human random interference. First, it involves the acquisition and cleaning of original sensing data. The system acquires a set of original sensing records of the target freezer within a preset historical time window (for example, the past 24 hours) through a cloud interface or an edge computing gateway. The original records contain multi-dimensional data with non-aligned timestamps, including real-time compressor power, cabinet door opening and closing state (usually triggered by a reed switch or a Hall sensor), store ambient light intensity (feedback by a photoresistor or a light sensor), and ambient temperature and humidity, etc. To lock the closed-door silent period, the system performs a filtering logic based on multi-source signal fusion. Specifically, the operation load of the freezer is mainly composed of heat and mass exchange caused by door opening and heat transfer caused by envelope structure penetration. To extract the weak dew point driven load fluctuation (i.e., the correlation between envelope structure penetration and anti-condensation control), the opening interference with a huge energy level and the heat shock caused by high-frequency replenishment during the day must be completely removed from the signal, so the double constraints of door magnet state and ambient light (representing whether the store is open) are needed to lock the pure night static maintenance stage.

[0011] In a preferred embodiment, the system first defines a filtering function for identifying valid time intervals that meet the silent condition from the original time series. Let the time index of the original data set be , the door magnet sensor state signal be (where 0 represents closed and 1 represents open), the ambient light intensity signal be (unit: lux), the preset night light threshold be (for example, 10 lux), and the preset minimum continuous observation duration be (for example, 3 hours). The silent time window set selected by the system satisfies the following logical conditions: After determining the valid time window After that, because the sampling frequency of the original sensor data can be inconsistent (for example, 1 Hz for the power meter and 0.1 Hz for the temperature and humidity meter), it is necessary to perform uniform frequency time series interpolation resampling processing. Linear interpolation algorithm is used to uniformly resample all data to a standard frequency of 1 Hz. For any to-be-aligned time , if it is located between the original sampling points and , the resampled value is calculated by the following formula: . Wherein, represents the interpolation result at the resampling time, which can be the compressor power, the ambient temperature, the ambient humidity or the cabinet temperature; and represent the measured values at the immediately preceding time and the immediately following time in the original data sequence; is the standardized target time stamp, and are the time stamps corresponding to the original data. Through the foregoing steps, the system outputs a time-aligned matrix, which contains synchronized power , ambient temperature , relative humidity and cabinet temperature .

[0012] In some possible embodiments, considering that some old refrigerators or low-cost models can not be equipped with an ambient light sensor, the system can use an alternative solution based on time strategy and load variance joint verification to intercept the closed-door silent period. Specifically, the system no longer relies on the light intensity signal, but uses a pre-set empirical closed store time interval (for example, 02:00 to 05:00). To prevent unintended human activities (such as night inventory or equipment maintenance) from interfering with the data during this period, the system will further calculate the sliding variance of the compressor power sequence after filtering out the time period when the door magnet state shows closed. Only when the macro fluctuation variance of the compressor power is lower than the pre-set steady-state threshold, the period is confirmed as an effective closed-door silent period. Thus, by using the thermodynamic stability characteristics of the system under no human interference, the system can still effectively eliminate abnormal interference in the absence of a light sensor, ensuring the compatibility and coverage of the monitoring system for different hardware configuration models.

[0013] In some possible embodiments, to address the packet loss or communication interruption problem that can occur during data transmission, a data integrity pre-verification step is added before performing interpolation resampling processing. In this implementation, after locking the closed-door silent period, the system scans the time stamp interval in the original data sequence. If the time interval between any two adjacent original data points is found to be greater than the pre-set threshold If the maximum tolerance time limit (e.g. 300 seconds) is exceeded, it is determined that there is a data gap in the time window. The system will give up interpolating the data in the window, directly discard the current window data and try to search for the next silent window that meets the conditions, or report data quality abnormal warning to the server.

[0014] It should be noted that the dew point temperature is a critical amount of water vapor in the air to reach saturation state, and is also the most critical driving variable (i.e. wet stress) in the control logic of the ice cabinet anti-condensation heater (ASH). Since the sensor configuration of the conventional commercial ice cabinet usually only includes a dry bulb temperature sensor and a relative humidity sensor, the system must reconstruct these two basic variables into a dew point temperature that can reflect the condensation risk through the thermodynamic state equation, and the reconstruction process must retain the nonlinear change characteristics, so that it can capture the dynamic response of the control loop to this nonlinear stress through bispectrum analysis. Specifically, the system uses an improved saturated water vapor pressure empirical correlation algorithm based on the Magnus formula, which can maintain very high calculation accuracy in high humidity environments, thereby constructing a high-fidelity wet stress signal source.

[0015] In a preferred embodiment, the system first extracts the in-store ambient temperature data sequence and the in-store ambient relative humidity data sequence from the pre-processed time-aligned operational data sequence.

[0016] To implement the operation based on the natural logarithm and temperature fractional correction term, the system defines an intermediate variable which integrates the logarithmic decay of humidity and the saturation pressure contribution of temperature. Specifically, first calculate the intermediate variable ; then, based on the intermediate variable, calculate the ambient dew point temperature . In the formula, represents the Celsius temperature value (unit: ) at time ; represents the relative humidity percentage value (value range 0-100) at time ; indicates the natural logarithm operation; parameters and are Magnus empirical constants, preferably , . This parameter combination has excellent fitting accuracy in the temperature range of to , which can cover all extreme working conditions of convenience stores.

[0017] In some possible embodiments, considering that different brands or models of ice cabinet controllers may employ different versions of firmware algorithms to determine the risk of condensation, in order to make the calculation benchmark of the monitoring system consistent with the internal logic of the device under test, the calculation module has a parameter self-adaptive configuration function. Specifically, the experience constant and is not a fixed value, but is dynamically loaded according to the ice cabinet device metadata identified by the system. For example, when the system identifies that the monitoring object is an old controller that employs a simplified Tetens formula, the system will automatically call the parameters and to ensure that the calculated dew point sequence is synchronized with the stress signals actually experienced by the device, thereby eliminating false positives caused by algorithm differences and widening the applicability of the scheme to inventory devices.

[0018] In some possible embodiments, to enhance the robustness of the algorithm when the sensor data fluctuates abnormally, the system adds a numerical clamping preprocessing step before performing the natural logarithm operation. Since electromagnetic interference in the actual operating environment may cause the relative humidity reading to jump instantaneously to 0 or a value extremely close to 0, which will cause the operation result to tend towards negative infinity, thereby causing the entire calculation process to crash or produce invalid NaN values. Specifically, the system will first check the value of before calculation. If is detected, it is forced to be assigned a value of 0.1; if is detected, it is forced to be assigned a value of 99.9. Thus, in the event of extreme value drift or sensor failure, the dew point calculation module can still output a continuous and bounded numerical sequence.

[0019] It should be noted that the compressor power of the ice cabinet not only includes the basic refrigeration energy consumption for maintaining a low-temperature environment, but also superimposes the pulse width modulation signal of the anti-condensation heater (ASH) and the dynamic response signal of the control system to environmental perturbations. In order to effectively monitor the health status of the anti-condensation control loop, it is necessary to first normalize the power data with respect to the environmental temperature difference to eliminate the influence of macroscopic thermal load changes, and then separate the low-frequency trend representing the basic refrigeration amount through frequency domain separation technology, thereby extracting the texture signal hidden in the high-frequency band that reflects the control logic action characteristics.

[0020] In a preferred embodiment, the process of extracting the unit temperature difference load fluctuation rate sequence of the refrigeration system needs to follow the operation logic of normalization and statistical filtering. First, the system reads the time-aligned compressor power data sequence , the in-store ambient temperature data sequence , and the cabinet temperature data sequence .

[0021] To eliminate the linear impact of ambient temperature difference on energy consumption benchmark, the system calculates unit temperature difference load at each time point , whose calculation formula is: , wherein is a very small positive number (for example, ) to prevent division by zero error, but in normal freezer operating conditions, and there is a significant temperature difference. Then, to extract the smooth trend component , the system uses an exponential weighted moving average filter (EWMA), whose calculation formula is: , wherein is a smoothing factor (take 0.05, corresponding to a strong time smoothing effect). Then, the high-frequency residual load sequence is obtained by subtraction operation to remove the thermal inertia background. Finally, to quantify the fluctuation intensity of the residual signal, the system uses a sliding time window root mean square (RMS) algorithm to generate a unit temperature difference load fluctuation rate sequence . The specific calculation formula is: . In the formula, represents the width of the sliding window (for example, take 60 sampling points, corresponding to 1 minute of data length), so that the amplitude energy of high-frequency clutter is converted into a continuous fluctuation rate index, and the switching action density of the anti-fog heater is mapped to a numerical size.

[0022] In some possible embodiments, considering the differences in computing power of different embedded processors, if the hardware resources of the target device are limited and it is difficult to support floating point square root operation, the system can use the mean absolute deviation (MAD) algorithm to replace the root mean square algorithm to extract the fluctuation rate. Specifically, the way to obtain the high-frequency residual load sequence remains unchanged, but when calculating the final fluctuation rate sequence , the following simplified algorithm is used: . This embodiment uses absolute value operation to replace square and square root operation, which greatly reduces the computational complexity while still effectively representing the dispersion degree and fluctuation intensity of the load signal on a linear scale, ensuring the algorithm landing on a low-cost hardware platform.

[0023] In some possible embodiments, for some high-end models with variable frequency compressors or complex defrosting logic, a single low-pass filter subtraction may not be able to completely separate the extremely complex background trend. Specifically, the step of extracting the high-frequency residual load sequence is replaced by a digital high-pass filter directly. The system designs a Butterworth high-pass filter with a cutoff frequency of (for example, 0.01 Hz), which directly processes the unit temperature difference load sequence The filter is inputted. The difference equation of the filter directly outputs a residual sequence containing only high-frequency components, automatically filtering out the direct current component and low-frequency thermal drift. Then, the sliding variance is calculated based on the sequence output by the filter. Thus, by using the frequency domain truncation characteristics in signal processing, the interference caused by the slow frequency rise and fall of the variable frequency compressor can be more cleanly removed, so that the extracted volatility signal more purely reflects the transient action of the anti-fogging control loop, and is suitable for precise monitoring scenes with higher requirements for signal purity.

[0024] It should be noted that cross-bispectrum analysis is a high-order statistical tool in the field of signal processing for detecting whether there is quadratic phase coupling (QPC) between two signals. The complex frequency domain operation is projected to the time domain for low-power implementation. Specifically, the anti-fogging control system is a typical nonlinear system, and its response to external wet stress (dew point change) is often not linear, but the faster the stress changes, the more the frequency and amplitude of the controller's action (i.e. load volatility) increases exponentially. Therefore, by calculating the product of the square of the dew point change rate and the load volatility, a proxy index of third-order cumulant is constructed, which can significantly amplify the time when the environment changes dramatically and the system responds dramatically, thereby accurately positioning the effective interval of nonlinear modulation in time sequence and filtering out meaningless background noise.

[0025] In a preferred embodiment, the process of constructing the instantaneous cross-bispectrum coupling potential sequence is performed in the order of time domain difference, nonlinear mapping, and modulation product. First, the system obtains the time-aligned ambient dew point temperature sequence and the unit temperature difference load volatility sequence . To calculate the ambient dew point temperature change rate sequence, the system uses a first-order backward difference algorithm and sets a sampling interval (e.g. 10 seconds, i.e. 10 data points). At time , the ambient dew point temperature change rate is calculated as follows: . Next, to extract the ambient dew point change rate square sequence representing the strength of nonlinear modulation, the system performs a square operation on to obtain . This square operation not only eliminates the directional difference of dew point rise or fall (i.e. focuses on the absolute degree of change), but also simulates the quadratic characteristics of energy transfer in a nonlinear system. Finally, the unit temperature difference load volatility sequence is multiplied point by point with the square sequence to generate the instantaneous cross-bispectrum coupling potential sequence . Its calculation formula is as follows: . In the formula, representative time of the load texture intensity, and are the current time and the dew point temperature value (unit: ) before the time, is the differential step (unit: seconds). The larger the value, the greater the external wet stress impact on the system at that time and the stronger the high-frequency response of the internal control loop, i.e., a significant coupling event occurs.

[0026] In some possible embodiments, in order to solve the inherent response lag problem of the thermodynamic system, a phase synchronization compensation mechanism is introduced when performing the final point-by-point multiplication operation. Since the external dew point changes, the wet air penetrates into the cabinet and triggers the change of the humidity-sensitive characteristics of the glass surface, and then causes the heater to act, which has inherent time delay (TimeLag). Specifically, the dew point change rate square sequence Before being multiplied by the load fluctuation rate sequence , it will be shifted backward along the time axis by a lag step (e.g. seconds). At this time, the calculation logic of the instantaneous cross-bispectrum coupling potential is adjusted as follows: the dew point change impact at time is used to match the load fluctuation response at time . This time-displaced multiplication operation can more accurately capture the causal relationship and prevent signal peak displacement due to slow system response, thereby improving the signal-to-noise ratio of the coupling feature.

[0027] In some possible embodiments, the phase synchronization compensation mechanism, in view of the possible slight jitter noise of the environmental sensor data, uses a Savitzky-Golay smoothing differential filter instead of a simple first-order difference when calculating the environmental dew point temperature change rate. Specifically, the system selects a local window containing data points (e.g. ), and directly solves the derivative of the center point of the window using a polynomial fitting method. This method can effectively suppress high-frequency quantization noise while calculating the change rate, generating a smoother sequence.

[0028] It should be noted that phase space reconstruction technology can map the hidden dynamic evolution laws in a one-dimensional time series to a high-dimensional geometric space for intuitive expression. Specifically, the response of the anti-condensation control loop to changes in the environmental dew point exhibits typical nonlinear hysteresis characteristics, meaning that the response path during the dew point rise phase does not coincide with the recovery path during the dew point fall phase, thus forming a specific closed loop. By mapping the environmental dew point temperature, which serves as the external excitation source, and the cross-bispectral coupling potential energy, which represents the system's internal nonlinear response, to the horizontal and vertical coordinates of the phase plane, the system can transform the abstract control logic state into a geometric trajectory with topological characteristics. The trajectory of a normally functioning system should present a regular ring structure with a certain opening area, while a faulty system will exhibit trajectory collapse, divergence, or disorder.

[0029] In a preferred embodiment, the process of constructing a two-dimensional phase space hysteresis trajectory follows the operational logic of time synchronization mapping and vector space synthesis. The system first reads a length of... Time-aligned sequence: Ambient dew point temperature sequence and instantaneous cross-spectral coupling potential energy sequence To construct a set of trajectories describing dynamic evolution paths. The system is indexed by time. (in Extracting data pairs from the same time point one by one to generate a two-dimensional state vector. Specifically, defining the first state vector in the two-dimensional phase space... State points The coordinates are The horizontal axis coordinate components ordinate components of the vertical axis The resulting phase space hysteresis trajectory It can be represented as a set sequence of ordered coordinate pairs: In the formula, The representative system is Thermodynamic-control coupling state at any given moment. The magnitude of the value represents the intensity of the moisture load applied by the external environment at that time. The magnitude of the value characterizes the severity of the nonlinear adjustment action generated within the system to counteract the wet load. (Arrow) This represents the temporal connectivity of trajectory construction. This temporal connection defines the flow direction of the hysteresis loop (clockwise or counterclockwise), recording not only the position of the state but also the direction of state evolution.

[0030] In some possible embodiments, in order to eliminate the difference in the numerical range of different dimensional physical quantities, prevent the effective features of compressing another coordinate axis due to the too large value of a coordinate axis, and increase a coordinate normalization mapping step before constructing the trajectory. Specifically, the system does not directly use the original physical quantity value as the coordinate, but first performs Min-Max standardization processing on the ambient dew point temperature sequence and the instantaneous cross-bispectrum coupling potential sequence respectively. Let the minimum value of the ambient dew point sequence be , and the maximum value be , then the mapped abscissa ; similarly, the ordinate is processed. After this processing, the constructed two-dimensional phase space hysteresis trajectory is strictly constrained within a unit square region .

[0031] In some possible embodiments, in order to solve the problem that the actual running data may have sporadic outliers, causing the trajectory line to appear sharp broken lines or breaks, the system adopts a spatial domain moving average smoothing strategy after generating the coordinate points and before connecting them into a trajectory. Specifically, the system does not directly connect the original points and , but defines a new smoothed point , whose coordinates are determined by the center of gravity of the points in the neighborhood window of in the original sequence. For example, a three-point moving average formula is used: . The system constructs a connected trajectory based on the smoothed point set . Thus, the high-frequency burrs caused by instantaneous electromagnetic interference of the sensor are filtered out in terms of geometric shape, making the final generated hysteresis loop smoother and better closed.

[0032] It should be noted that by introducing the dual perspectives of information theory and topological geometry, the complex dynamic behavior of the freezer control system is quantized in dimension. Specifically, a healthy anti-dew control system should exhibit ordered complexity in phase space, i.e., its entropy value (representing information quantity) should be at a moderate level, and there must be a significant hysteresis loop area (representing physical memory and phase transition delay). Pure high entropy may mean that the system is completely chaotic and out of control (such as sensor noise), while pure low entropy may mean that the system is dead (such as actuator stuck). Therefore, by constructing a composite index, the area of the hysteresis loop is used as the denominator to correct the Shannon entropy, thereby distinguishing between disordered random noise and ordered control response, and achieving accurate measurement of implicit decoupling faults.

[0033] In a preferred embodiment, the process of calculating the cross-bispectrum hysteresis entropy of the dew point driven load fluctuation rate includes four rigorous mathematical steps of gridding probability statistics, entropy calculation, area integration and weighted fusion. First, the system divides the plane region where the constructed two-dimensional phase space hysteresis trajectory is located into Uniform micro-grids (e.g.) This forms a position index matrix. The system iterates through all trajectories in the trajectory set. The number of two-dimensional coordinate points that fall into the first two-dimensional coordinate point is counted. Line number Column grid cells Number of points inside And calculate the joint probability distribution matrix. Based on this probability matrix, the system uses the Shannon formula to calculate the basic Shannon entropy value. The calculation formula is as follows: ,in To prevent extremely small positive numbers whose logarithms are negative infinity (e.g.) Simultaneously, the system calculates the directional circulation area enclosed by the trajectory using either Green's Theorem or the trapezoidal integral rule. Set the trajectory point sequence as follows: The formula for calculating its area is: This formula accurately calculates the geometric area enclosed by a closed or approximately closed path by summing the magnitudes of the cross products of adjacent vectors, and takes the absolute value to eliminate the influence of the clockwise or counterclockwise direction of the trajectory. Finally, the system calculates the final cross-spectral hysteresis entropy using a weighted correction formula. The formula is: In the formula, The discreteness and richness of the system state were quantified. The regularity and memory depth of the system's response to stimuli were quantified. The preset circulation weighting coefficient (e.g., 2.0) is used to adjust the suppression strength of the area term on the final entropy value, ensuring that normal hysteresis behavior can effectively reduce the entropy value and make the result fall within the judgment interval.

[0034] In some possible implementations, considering that phase space trajectories may appear sparse due to insufficient data within certain short-period observation windows, leading to a large number of zero values ​​in the probability matrix generated by the traditional hard-binning method, and consequently causing instability in entropy calculation, kernel density estimation (KDE) is used instead of the grid counting method to generate the probability distribution. The system uses a Gaussian kernel function to perform smooth convolution on the two-dimensional trajectory points to generate a continuous probability density function. The function is then discretized and sampled to obtain the probability matrix. This is equivalent to applying probability dispersion processing to discrete trajectory points. Even with a small number of data points, it can construct a smooth and statistically significant distribution, thus ensuring the basic entropy value. Robustness in short-window monitoring scenarios.

[0035] In some possible embodiments, for the problem that the trajectory may appear multiple self-intersections under certain working conditions, causing the algebraic area calculated by the Green formula to offset each other, the Convex Hull Area algorithm is used as an alternative solution when calculating the directed loop flow area. Specifically, the system first uses the Graham scan algorithm or the Monotone Chain algorithm to find the smallest convex polygon boundary containing all the trajectory points , and then calculates the geometric area of the convex polygon as . The convex hull area represents the maximum dynamic range of the system in the phase space, which can absolutely and stably measure the tension of the system response, although it ignores the microscopic curling inside the trajectory. When the system fails and the trajectory collapses into a point or a line, the convex hull area will tend to zero, causing the final to rise sharply, thereby acutely triggering an abnormal alarm.

[0036] It should be noted that the numerical size of the cross-bispectrum hysteresis entropy value of the freezer is deeply affected by individual differences such as the aging degree of the equipment hardware, the power configuration of the compressor, and the basic insulation performance of the store. Therefore, the benchmark for fault determination cannot use a fixed hard threshold, but must be based on the historical behavior of the device itself to build a dynamic benchmark. Specifically, the characteristic entropy value of a system in a stable controlled state should obey a certain stable probability distribution (such as a normal distribution). When the entropy value deviates significantly from the interval of this distribution, whether it is because the entropy value is too low (representing a dead state of the system losing response ability) or the entropy value is too high (representing a chaotic state of the system losing control regularity), it means that the coupling relationship between the entity and the control logic has undergone a qualitative change, that is, an implicit failure has occurred. Through bidirectional threshold determination, both control failure and control decoupling typical failure modes can be covered.

[0037] In a preferred embodiment, the fault prediction output module performs the process of state verification based on a sliding window statistical model. The system first retrieves a set of historical entropy values calculated during the normal operation of the target freezer in the past days (for example ) from the database. In order to construct a statistically significant benchmark interval, the system is based on the historical sample set Gaussian distribution under steady state, and calculates the arithmetic mean and the sample standard deviation . The calculation formulas are respectively: and . Based on the above statistical quantities, the system constructs the current effective historical benchmark interval, where the lower threshold , and the upper threshold wherein is a confidence coefficient, preferably corresponding to a confidence interval of 99.7%, meaning that only a very small probability of outliers will be considered abnormal. Subsequently, the system acquires the cross-bispectrum hysteresis entropy of the dew-point driven load fluctuation rate and performs a numerical comparison logic: if then the system determines that the response of the anti-dew control loop of the current freezer to the environmental humidity stress is in a reasonable dynamic balance, and outputs a state of Normal; otherwise, if or then the system determines that a serious decoupling or deadlock has occurred between the control loop and the environmental stress, and outputs a state of Abnormal. By using a judgment method based on adaptive statistical distribution, the drift of the baseline caused by seasonal changes or slow aging of the equipment can be automatically offset, ensuring a high accuracy and low false alarm rate of the fault warning.

[0038] In some possible embodiments, considering that the operation data of some freezers can present a skewed distribution or contain historical dirty data, using mean and variance to construct the interval can not be robust enough, and therefore a non-parametric statistical method based on quantiles is used to define the baseline interval. Specifically, the system sorts the acquired historical entropy value sample set in ascending order, and extracts the 25th percentile (first quartile) and the 75th percentile (third quartile). Then, the interquartile range is calculated. Based on the principle of box-plot, the baseline interval defined by the system is (Q1-1.5*IQR, Q3+1.5*IQR), wherein is an expansion coefficient (for example, 1.5 or 3.0). If the current entropy value falls outside the interval, it is determined to be abnormal. Specifically, the non-parametric statistical method based on quantiles is not sensitive to extreme outliers in the historical data, and even if individual fault data that is not labeled is mixed in the historical records, the baseline interval will not be significantly enlarged, thereby maintaining the system's ability to sensitively capture small fault signs.

[0039] In some possible embodiments, to prevent occasional false alarms caused by single calculation errors, the system introduces a time sequence persistence verification mechanism. Specifically, the fault prediction output module not only compares the current entropy value but also combines a state counter. The system sets a continuous abnormal frame number threshold (for example, 3 frames). When first jumps out of the historical baseline interval, the system does not immediately output the Abnormal state, but marks the device as a suspected abnormality and starts the counter. Only when the current entropy value When the hysteresis entropy values calculated in each observation window (for example, three consecutive night periods) are all outside the reference interval, the system finally confirms the fault and pushes the alarm information to the user. Conversely, if the value of the regression interval appears in the middle, the counter is reset to zero. Thus, by increasing the confirmation logic in the time dimension, false alarms caused by power grid fluctuations, sudden extreme weather, or transient sensor failures are effectively filtered out, significantly improving the usability and user trust of the monitoring system in actual industrial applications.

[0040] The above describes the embodiments of the present embodiment, but the present embodiment is not limited to the specific embodiments described above, which are only illustrative and not restrictive. Those skilled in the art can make many forms under the inspiration of the present embodiment, which are all within the protection of the present embodiment.

Claims

1. A remote monitoring and fault prediction system for freezers, characterized in that, include: The data preprocessing module is used to capture the operating data of the freezer during the closed and silent period; A dual-channel signal generation module is used to calculate the ambient dew point temperature sequence based on the operating data and extract the unit temperature difference load fluctuation rate sequence of the refrigeration system. The data analysis module is used to construct an instantaneous cross-bispectral coupled potential energy sequence characterizing the nonlinear modulation relationship between the time change rate of the environmental dew point temperature sequence and the unit temperature difference load fluctuation rate sequence. The phase space entropy calculation module is used to construct a two-dimensional phase space hysteresis trajectory with the environmental dew point temperature sequence as the horizontal axis and the instantaneous cross-bispectral coupled potential energy sequence as the vertical axis, and to calculate the cross-bispectral hysteresis entropy of the dew point driven load fluctuation rate based on the probability density distribution and directional circulation area of ​​the two-dimensional phase space hysteresis trajectory. The fault prediction output module is used to verify whether the cross-bispectral hysteresis entropy of the dew point driven load volatility is within the historical benchmark range, and outputs the freezer operating status accordingly.

2. The freezer remote monitoring and fault prediction system according to claim 1, characterized in that, Capture the freezer's operational data during the closed, silent period, including: Acquire the original sensor records of the freezer within a preset historical time window; based on the door opening / closing status signal fed back by the door magnetic sensor and the light intensity signal fed back by the ambient light sensor, select a continuous time interval from the original sensor records where the door remains closed and the ambient light intensity is lower than a preset nighttime threshold, as the closed silent period; perform time series interpolation and resampling processing on the compressor power data, store ambient temperature data, store ambient relative humidity data, and freezer temperature data within the closed silent period to generate a time-aligned operating data sequence.

3. The freezer remote monitoring and fault prediction system according to claim 2, characterized in that, Calculate the environmental dew point temperature sequence based on the aforementioned operational data, including: The indoor ambient temperature data sequence and the indoor ambient relative humidity data sequence are extracted from the running data sequence; the indoor ambient relative humidity data sequence is processed by natural logarithmic operation using the saturated water vapor pressure empirical correlation algorithm, and a temperature fractional correction term is constructed in combination with the indoor ambient temperature data sequence; based on the ratio relationship between the result of the natural logarithmic operation and the temperature fractional correction term, the ambient dew point temperature sequence is calculated point by point in time.

4. The freezer remote monitoring and fault prediction system according to claim 3, characterized in that, Extract the unit temperature difference load fluctuation rate sequence of the refrigeration system, including: The compressor power data sequence in the operation data sequence is divided by the difference between the store ambient temperature data sequence and the cabinet temperature data sequence at each time point to generate a unit temperature difference load sequence. The unit temperature difference load sequence is subjected to low-pass filtering to obtain a smoothing trend component, and the smoothing trend component is subtracted from the unit temperature difference load sequence to obtain a high-frequency residual load sequence with thermal inertia background removed. The high-frequency residual load sequence is then subjected to root mean square statistical operation based on a sliding time window to calculate the unit temperature difference load fluctuation rate sequence characterizing the load texture roughness window by window.

5. The freezer remote monitoring and fault prediction system according to claim 4, characterized in that, Based on the time-varying rate of change of the environmental dew point temperature sequence and the unit temperature difference load fluctuation rate sequence, an instantaneous cross-spectral coupled potential energy sequence is constructed, including: A time-domain difference operation is performed on the environmental dew point temperature sequence to calculate the environmental dew point temperature change rate sequence; the environmental dew point temperature change rate sequence is squared to extract the environmental dew point change rate square sequence characterizing the nonlinear modulation intensity; the unit temperature difference load fluctuation rate sequence is multiplied point by point with the environmental dew point change rate square sequence to generate an instantaneous cross-bispectral coupling potential energy sequence characterizing the second-order phase coupling intensity between external wet stress and internal load texture.

6. The freezer remote monitoring and fault prediction system according to claim 5, characterized in that, A two-dimensional phase space hysteresis trajectory is constructed using the environmental dew point temperature sequence as the horizontal axis and the instantaneous cross-spectral coupling potential energy sequence as the vertical axis, including: The environmental dew point temperature sequence is used as the horizontal axis component to describe the external wet stress state, and the instantaneous cross-spectral coupled potential energy sequence is used as the vertical axis component to describe the internal dynamic response intensity. According to the time synchronization correspondence, the horizontal axis component and the vertical axis component at the same moment are combined to generate two-dimensional coordinate points, and the two-dimensional coordinate points at all moments are connected in chronological order to construct a two-dimensional phase space hysteresis trajectory describing the dynamic evolution path of the freezer during the closed silent period.

7. The freezer remote monitoring and fault prediction system according to claim 6, characterized in that, Based on the probability density distribution and directed circulation area of ​​the two-dimensional phase space hysteresis trajectory, the cross-bispectral hysteresis entropy of the dew-point driven load volatility is calculated, including: The planar region containing the two-dimensional phase space hysteresis trajectory is divided into several uniform micro-grids. The proportion of two-dimensional coordinate points falling into each micro-grid to the total number of points is counted to generate a joint probability distribution matrix. Based on the joint probability distribution matrix, the basic Shannon entropy value is calculated to quantify the dispersion of the two-dimensional phase space hysteresis trajectory distribution. The directed circulation area enclosed by the two-dimensional phase space hysteresis trajectory is calculated using a closed-loop integral algorithm to quantify the intensity of the system's ordered stress response to dew point changes. The cross-bispectral hysteresis entropy of the dew point-driven load volatility is calculated by division using the basic Shannon entropy value as the numerator and the value containing the weighted correction term of the directed circulation area as the denominator.

8. The freezer remote monitoring and fault prediction system according to claim 7, characterized in that, Verify whether the cross-spectral hysteresis entropy of the dew point-driven load volatility is within the historical benchmark range, and output the freezer operating status accordingly, including: Obtain several historical entropy value samples calculated during the freezer's historical normal operation cycle, and construct an effective historical benchmark interval based on the arithmetic mean and standard deviation of the historical entropy value samples; determine whether the cross-spectral hysteresis entropy of the current dew point driven load volatility falls within the historical benchmark interval; if the determination result is that it falls within the historical benchmark interval, the freezer is determined to be in a normal state with good coupling between anti-condensation control and wet load; if the determination result is that it does not fall within the historical benchmark interval, the freezer is determined to be in an abnormal state of control decoupling or failure.

Images