A method and system for detecting abnormalities in refrigeration valves

Abstract

This application relates to the field of data processing technology, and in particular to a method and system for anomaly detection of refrigeration valves. The method includes: acquiring preprocessed operating data of the refrigeration valve; for any data point in the operating data, calculating the sparsity of the data point based on its distribution in the local area; simultaneously, constructing a temporal neighbor reference set for the data points, and using the LOF algorithm to obtain the initial anomaly score for each data point in the neighbor reference set; constructing a trend anomaly degree based on the initial anomaly scores of the data points in the temporal neighbor reference set and the changes in data values; adjusting a preset initial anomaly threshold based on the trend anomaly degree and the sparsity to obtain a dynamic threshold, wherein the dynamic threshold is negatively correlated with the trend anomaly degree; and judging the operating data of the refrigeration valve based on the dynamic threshold. This application has the effect of improving the accuracy of anomaly detection in refrigeration valves.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for detecting anomalies in refrigeration valves. Background Technology

[0002] Refrigeration valves are an indispensable and common component in refrigeration systems, used to control and regulate fluid flow. Common refrigeration valves include shut-off valves, expansion valves, and safety valves. To ensure the stable operation of refrigeration systems, it is usually necessary to monitor these valves. Traditional manual inspection methods are often outdated and cannot promptly identify problems in the system. Currently, with the development of sensor and data processing technologies, some machine learning-based methods are increasingly being used in the monitoring of industrial refrigeration systems. For example, the LOF (Local Outlier Factor) algorithm identifies "outliers"—points with significantly deviated local densities—by comparing the density of data points with that of their neighbors. This method can effectively uncover hidden anomalies under unsupervised conditions and is suitable for modeling multi-dimensional, nonlinear refrigeration system data.

[0003] Traditional LOF (List of Elements) algorithms have played a role in detecting anomalies in industrial data. However, they also have some unavoidable drawbacks. For example, changes in the operating state of a refrigeration system can alter industrial data, creating sparsely populated areas of data points. LOF algorithms typically classify these as anomalies, even though changes in operating conditions are normal occurrences during system operation and not genuine anomalies. Therefore, traditional LOF algorithms generally have low accuracy in detecting anomalies. Summary of the Invention

[0004] To improve the accuracy of monitoring refrigeration valve operation data, this application provides a method and system for detecting anomalies in refrigeration valves.

[0005] Firstly, this application provides a method for detecting abnormalities in a refrigeration valve, employing the following technical solution:

[0006] An anomaly detection method for a refrigeration valve includes: acquiring preprocessed operating data of the refrigeration valve; for any data point in the operating data, calculating the sparsity of the data point based on its distribution in the local area; simultaneously, constructing a temporal neighbor reference set for the data points, and using the LOF algorithm to obtain initial anomaly scores for each data point in the neighbor reference set; constructing a trend anomaly degree based on the initial anomaly scores of the data points in the temporal neighbor reference set and changes in data values; adjusting a preset initial anomaly threshold based on the trend anomaly degree and the sparsity to obtain a dynamic threshold, wherein the dynamic threshold is negatively correlated with the trend anomaly degree, and judging the operating data of the refrigeration valve based on the dynamic threshold.

[0007] The beneficial effects are as follows: First, by using a spatially nearest reference set to calculate sparsity, the density differences of data points in spatial distribution can be effectively distinguished, avoiding misjudgments of sparse regions caused by changes in operating conditions due to reliance solely on local density, thereby reducing false alarms caused by operating condition migration. Second, by introducing data trend analysis based on a temporally nearest reference set and the initial anomaly score of LOF, the degree of trend anomaly is calculated, enabling detection to consider not only the spatial distribution characteristics at a single moment but also comprehensively reflect the changing trend of data in the time dimension, thus identifying anomaly patterns of slow shifts or sudden changes. Furthermore, by constructing a dynamic threshold negatively correlated with the degree of trend anomaly, the threshold is increased to suppress false alarms when sparse but temporally stable, while the threshold is decreased to provide early warnings when temporal changes are significant, thus reducing both false and false alarms. This dynamic threshold adjustment mechanism overcomes the poor adaptability problem caused by the static threshold of the traditional LOF algorithm, achieving adaptive detection of anomalies under different operating conditions.

[0008] Optionally, the step of calculating the sparsity of a data point based on its distribution in the local area includes: constructing a spatially nearby reference set; obtaining the data differences between each data point in the spatially nearby reference set and the current data point; and using the average of the data differences between multiple data points as the sparsity of the pixel.

[0009] The beneficial effects are: constructing a spatially nearby reference set, then calculating the data difference between it and the current data point and taking the mean as the sparsity, can accurately reflect the local density characteristics of the data points at the spatial distribution level.

[0010] Optionally, the steps for obtaining the data differences between each data point in the spatial neighboring reference set and the current data point include: calculating the value difference based on the differences between each data point in the spatial neighboring reference set and each data point in the current data point; calculating the directional difference based on the cosine angle between each data point in the spatial neighboring reference set and the current data point; and using the product of the value difference and the directional difference as the data difference.

[0011] The beneficial effects are as follows: the value difference directly measures the degree of deviation of the amplitude of the current data point from the data points in the spatially adjacent reference set in various physical quantities (such as pressure, temperature, flow rate, etc.), which can reflect the numerical changes of the operating parameters; secondly, the direction difference measures the consistency of the distribution direction of the data points in the multidimensional feature space through the cosine angle. When the cosine angle is large, it indicates that the data distribution trend is significantly different, which can reveal structural changes that cannot be found from the numerical difference alone.

[0012] Optionally, the steps of constructing a spatial proximity reference set include: for any data point, obtaining a preset number of data points that are spatially close to the data point as initial data points, and taking the mean of the Euclidean distances between the initial data points and the current data point as the proximity distance; and taking the set of data points whose Euclidean distance to the current data point is less than the proximity distance as the spatial proximity reference set.

[0013] The beneficial effects are as follows: First, by selecting a preset number of initial data points that are spatially close to the current data point, it ensures that these candidate points have high similarity in the multidimensional feature space, providing a reasonable basis for subsequent calculations of sparsity and data differences. Second, by using the mean of the Euclidean distance between the initial data points and the current data point as the nearest neighbor threshold, data points with excessively large distances are further filtered out. This can eliminate outliers in the initial set caused by occasional noise, extreme fluctuations, or sudden changes in operating conditions, thereby reducing the bias in local density estimation.

[0014] Optionally, the step of calculating the value difference based on the differences between the data points of the spatially nearby reference set and the data points of the current data point includes: for any data point, calculating the absolute value of the difference between the current data point and the data point for that data point; using the normalized result of the absolute value as the standard difference, and using the mean of the standard differences of the data points as the value difference.

[0015] Optionally, the step of constructing the trend anomaly degree based on the initial anomaly scores of data points in the temporally adjacent reference set and the changes in data values ​​includes: any two adjacent data points in the temporally adjacent reference set constitute a temporal reference point pair; construct a local score difference based on the difference in the initial anomaly scores in the temporal reference point pair; construct a local data change based on the difference in each data item of the two data points in the temporally adjacent reference set; and construct the trend anomaly degree based on the local data change and the local score difference.

[0016] Multiple time-series reference point pairs are constructed, and the differences in initial anomaly scores and local data changes between two adjacent data points are compared. The overall trend anomaly is reflected by combining multiple local score differences and local data changes.

[0017] Optionally, the sum of local data changes can be used as the overall data change, the sum of multiple local score differences can be used as the overall score difference, the product of the overall score difference and the overall data change can be used as the anomaly index, and the normalized result of the anomaly index can be used as the degree of trend anomaly.

[0018] Optionally, the step of adjusting the preset initial anomaly threshold to obtain the dynamic threshold based on the degree of trend anomaly and sparsity includes: using the ratio of the sparsity of data points to the degree of trend anomaly as a first control index, and using the mean of the first control index corresponding to each data point as a second control index; using the ratio of the normalized result of the first control index to the second control index as an adjustment factor, and using the product of the adjustment factor and the initial anomaly threshold as the dynamic threshold.

[0019] Secondly, this application provides an abnormality detection system for a refrigeration valve, employing the following technical solution:

[0020] An anomaly detection system for a refrigeration valve includes a processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the anomaly detection method for a refrigeration valve as described above.

[0021] The beneficial effect is that the above-mentioned abnormal detection method for refrigeration valves is generated into a computer program and stored in the memory so that it can be loaded and executed by the processor. Thus, the system can be made based on the memory and the processor, which is convenient to use.

[0022] This application has the following technical effects:

[0023] In this application, sparsity is constructed based on the spatial distribution of data points, and trend anomaly index is constructed based on the changes in the initial anomaly scores of data points in the time series and the changes in data. The trend anomaly degree can reflect the degree of anomaly in the changes of data in the time series, thereby reducing the false alarms of data anomalies caused by the migration of working conditions. Attached Figure Description

[0024] Figure 1 This is a flowchart of a method for detecting abnormalities in a refrigeration valve according to an embodiment of this application.

[0025] Figure 2 This is a flowchart of step S2 in an embodiment of the present application for an abnormal detection method for a refrigeration valve. Detailed Implementation

[0026] This application discloses an anomaly detection method for a refrigeration valve. It constructs a sparsity index reflecting the density of the local area where the current data point is located based on the spatial location of the data point. Simultaneously, it constructs a temporally adjacent reference set that is temporally adjacent to the current data point. By analyzing the initial anomaly score and data changes in the temporally adjacent reference set, a trend anomaly index reflecting the temporal anomaly degree of the current data point is constructed. Finally, the initial anomaly threshold is adjusted by combining the two to obtain a dynamic threshold. Based on the dynamic threshold, the state of the current data point is determined, thereby judging the operating state of the refrigeration valve.

[0027] Reference Figure 1 An abnormality detection method for a refrigeration valve includes steps S1-S3.

[0028] S1: Obtain the pre-processed operating data of the refrigeration valve.

[0029] To achieve comprehensive perception of the refrigeration valve's operating status and accurate identification of abnormal behavior, multiple key operating parameters of the refrigeration valve need to be collected in real time during the normal operation of the refrigeration system. Specific parameters collected include: pressure values ​​at the valve's inlet and outlet, valve body temperature, valve drive current, and flow rate through the valve body. The inlet and outlet pressures are used to calculate the pressure difference to reflect the system load; the valve body temperature and drive current reflect the valve's operating intensity and heat dissipation; and the flow rate directly reflects the refrigerant's throughput efficiency. These parameters are collected in real time using pre-deployed industrial-grade pressure sensors, temperature sensors, current transformers, and flow meters. The sensors have a measurement accuracy of 0.1% or higher and are connected to the data acquisition module via RS485 or CAN (Controller Area Network) bus.

[0030] The data acquisition frequency is determined based on the rate of change in the refrigeration valve's operating conditions, typically set to once every 10 seconds to cover common periodic adjustment behaviors. Each data point generated during the acquisition process includes the aforementioned multi-dimensional data, thus yielding the raw data.

[0031] To improve the accuracy and robustness of subsequent anomaly detection and reduce noise interference, the raw data is preprocessed. Preprocessing operations include missing value imputation and normalization. Missing value imputation and normalization are standard techniques in this field and will not be elaborated upon here. The preprocessed raw data will be used as the input for subsequent steps.

[0032] S2: For any data point in the running data, calculate the sparsity of the data point based on the distribution status in the local area where the data point is located.

[0033] For any given data point, determine whether the local area where the data point is located belongs to a sparse region based on its spatial location, so as to further determine whether the data point is a real anomalous data.

[0034] Reference Figure 2 Step S2 includes steps S21-S23.

[0035] S21: Construct a spatially proximate reference set.

[0036] In the process of constructing a spatial proximity reference set, firstly, for any data point, a predetermined number of data points that are spatially close to the data point are selected as initial data points.

[0037] For constructing the nearest neighbor set of any data point, a preset number of data points should be selected as initial data points based on the distance between that data point and other data points. In this embodiment, the preset number of initial data points is 30. In other embodiments, those skilled in the art can adjust this based on experience and actual production conditions; for example, the preset number can be set to 40 or 50.

[0038] Then, the mean of the Euclidean distances between the initially selected data points and the current data point is taken as the nearest neighbor distance, and the set of data points whose Euclidean distance to the current data point is less than the nearest neighbor distance is taken as the spatial nearest neighbor reference set.

[0039] This step is primarily used to filter the initial data points, reducing the introduction of outliers into the spatially adjacent reference set and improving the accuracy and robustness of subsequent sparsity calculations. For example, suppose one or two initial data points have an Euclidean distance of 30 to the current data point, while the Euclidean distances of the other data points are concentrated between 10 and 15. This step can then remove data points with Euclidean distances far exceeding 30, thereby improving the accuracy of the sparsity calculation.

[0040] S22: Obtain the data differences between each data point in the spatially neighboring reference set and the current data point, and use the average of the data differences corresponding to multiple data points as the sparsity of the pixels.

[0041] In one embodiment, the step of obtaining the data difference between each data point in the spatial neighboring reference set and the current data point includes: calculating the value difference based on the differences between each data point in the spatial neighboring reference set and each data point in the current data point; calculating the direction difference based on the cosine angle between each data point in the spatial neighboring reference set and the current data point; and taking the product of the value difference and the direction difference as the data difference.

[0042] The steps for calculating the difference between data points in a spatially nearby reference set and data points in the current data point include: for any data point, calculating the absolute value of the difference between the current data point and that data point; using the normalized result of the absolute value as the standard difference, and using the mean of the standard differences of each data point as the value difference.

[0043] The directional difference is calculated based on the cosine angle between the data points of each data point in the spatially adjacent reference set and the data points of the current data point. The cosine angle between two data points indicates the consistency of the direction of the vectors corresponding to the two data points. The larger the cosine angle between the two data points, the more dispersed and irregular the two data points are in space, which suggests that the current data point may be in a sparse region.

[0044] By combining value difference and direction difference to calculate the data difference between the current data point and the data points in the spatially neighboring reference set, the accuracy and robustness of sparsity calculation are improved.

[0045] Based on the above steps, the data difference between each pixel in the spatially nearest reference set and the current data point is calculated, and the mean of the data difference is used as the sparsity of the current data point.

[0046] Specifically, the formula for calculating sparsity can be expressed as:

[0047] In the formula, Representing data points The sparsity of neighborhood data points, Representing data points The number of corresponding spatially adjacent reference set data points. Representing data points The Item data value, Representing data points Space Proximity Reference Set The first data point Item data value, Representing data points The corresponding spatially adjacent reference set of all data points The standard deviation of the data values ​​is primarily used to standardize the numerator. Representing data points The first in the corresponding spatial neighboring reference set The minimum cosine angle value of each data point; This indicates the dimension of the data contained in the data point. Taking this application as an example, the data points in this application include the pressure value at the front end of the refrigeration valve, the pressure value at the rear end, the valve body temperature, the valve's drive current, and the flow rate through the valve body. It is 5.

[0048] This section reflects the difference between one data point of the current data point and one of the data points of the corresponding spatially adjacent reference set. The differences of multiple data points corresponding to the data point are summed and averaged to obtain the overall value difference of each data point between the current data point and the data points.

[0049] It represents the minimum cosine angle between the current data point and the corresponding data point, indicating the directional difference in the distribution of the two data points.

[0050] This section reflects the data differences between two data points from two dimensions: the difference in data values ​​and the difference in the direction of distribution. The data differences are summed and averaged to reflect the overall distance between the current data point and spatially neighboring reference data points. The larger this distance, the greater the sparsity of the current data point's location; conversely, the smaller the distance, the less sparsity.

[0051] S3: Construct a temporal neighbor reference set for data points, and use the LOF algorithm to obtain the initial anomaly score of each data point in the neighbor reference set; construct the trend anomaly degree based on the initial anomaly score of the data points in the temporal neighbor reference set and the changes in data values.

[0052] Sparsity can be used to obtain the density of data points distributed in space. Data points with high sparsity may be real anomalous data points, or they may be normal data migration caused by changes in operating conditions.

[0053] Therefore, in this embodiment, a temporal proximity reference set for data points is constructed.

[0054] In the step of constructing a temporal proximity reference set for data points, for any given data point, a preset number of data points preceding that data point are acquired based on its acquisition time, forming the temporal proximity reference set. The data points in the temporal proximity reference set are close to the acquisition time of the current data point. In this embodiment, the preset acquisition number is 30. This number can be determined based on actual production conditions and the experience of those skilled in the art.

[0055] After the temporal proximity reference set is constructed, the LOF algorithm is used to obtain the initial anomaly score of each data point in the proximity reference set. The trend anomaly degree is constructed based on the initial anomaly score of the data points in the temporal proximity reference set and the changes in data values.

[0056] In constructing the trend anomaly level, any two adjacent data points in the temporal proximity reference set constitute a temporal reference point pair. Local score differences are constructed based on the differences in initial anomaly scores within the temporal reference point pair; local data changes are constructed based on the differences in various data points between the two data points in the temporal proximity reference set; and the trend anomaly level is constructed based on the local data changes and local score differences. The sum of local data changes is taken as the overall data change, the sum of multiple local score differences is taken as the overall score difference, the product of the overall score difference and the overall data change is taken as the anomaly index, and the normalized result of the anomaly index is taken as the trend anomaly level.

[0057] Specifically, the formula for calculating the degree of trend anomaly can be expressed as:

[0058] In the formula, Represents pixels The degree of trend abnormality; Indicates the number of data points in the temporally adjacent reference set; Representing data points The corresponding time-adjacent reference set Initial anomaly scores for each data point; Representing data points The corresponding time-adjacent reference set Initial anomaly scores for each data point; Representing data points The corresponding time-adjacent reference set The first pixel Item data; Representing data points The corresponding time-adjacent reference set The first pixel Item data; This represents the first time series data point in the reference set. The mean of the data items is mainly used for standardization of the molecule. Indicates the number of data points in the time-series adjacent reference set; This indicates the dimension of the data contained in the data point.

[0059] In the formula, the first The data point and the first Each data point is formed by two adjacent data points, constituting a time series reference point pair. This represents the local score difference between two data points in a time-series reference point pair. This section calculates the total difference in anomaly scores for all time series reference points, i.e., the overall score difference. The larger this value, the greater the anomaly score for each data point in the time series, and thus the more likely the current data point is to be a genuine anomaly.

[0060] This represents the difference between a later data point and an earlier data point in a time series reference point pair for a given data item. The average difference for each data item is calculated to obtain the local data change for that data point. Then, the local changes for each time series reference point pair are summed to obtain the overall data change.

[0061] This section represents the anomaly index; This section is mainly used to limit the range of values ​​for the anomaly index, so as to avoid the calculated result of the anomaly index being too large.

[0062] S4: Based on the degree of trend anomaly and sparsity, adjust the preset initial anomaly threshold to obtain a dynamic threshold. The dynamic threshold is negatively correlated with the degree of trend anomaly. The operating data of the refrigeration valve is judged based on the dynamic threshold.

[0063] In the process of obtaining the dynamic threshold, the ratio of the sparsity of data points to the degree of trend anomaly is used as the first control index, and the mean of the first control index corresponding to each data point is used as the second control index; the ratio of the normalized result of the first control index to the second control index is used as the adjustment factor, and the product of the adjustment factor and the initial anomaly threshold is used as the dynamic threshold.

[0064] Specifically, the formula for calculating the dynamic threshold can be expressed as: In the formula, Representing data points The corresponding adjustment factor; Representing data points The degree of sparsity; Representing data points The degree of trend abnormality; This represents the average ratio of the sparsity of all data points in the entire dataset to the degree of trend anomaly.

[0065] In this formula, when the sparsity is large but the trend anomaly is small, it means that although the current data point is in a region of sparse data density, there is no significant anomaly in the time series, which may be due to data sparsity caused by the migration of operating conditions. Therefore, for this type of case, the adjustment factor is greater than 1, which leads to a larger dynamic threshold and avoids false alarms.

[0066] Meanwhile, if a data point shows a significant trend anomaly, it may indicate a genuine data anomaly. This situation will result in a smaller numerator and an adjustment factor less than 1, which in turn will reduce the dynamic threshold, prompting the system to issue an early warning and reducing missed reports.

[0067] In addition, during the normal operation of the refrigeration valve, the sparsity is relatively small. If the data shows a slow anomaly in the time sequence, it will still cause the dynamic threshold to decrease, triggering an early alarm. This reduces the situation where the traditional LOF algorithm cannot detect slow data shifts, thus improving the accuracy and robustness of the system detection.

[0068] This application also discloses an anomaly detection system for a refrigeration valve, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an anomaly detection method for a refrigeration valve according to this application.

[0069] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0070] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for detecting abnormalities in a refrigeration valve, characterized in that, Obtain the pre-processed operating data of the refrigeration valve. For any data point in the operating data, calculate the sparsity of the data point based on the distribution status in the local area where the data point is located. Simultaneously, a temporal neighbor reference set for data points is constructed, and the LOF algorithm is used to obtain the initial anomaly score for each data point in the neighbor reference set. The degree of trend anomaly is constructed based on the initial anomaly scores of data points in the temporally adjacent reference set and the changes in data values. The dynamic threshold is obtained by adjusting the preset initial anomaly threshold based on the degree of trend anomaly and the sparsity. The dynamic threshold is negatively correlated with the degree of trend anomaly. The operating data of the refrigeration valve is judged based on the dynamic threshold. The steps for calculating the sparsity of a data point based on its distribution in the local area include: constructing a spatially nearby reference set; obtaining the data differences between each data point in the spatially nearby reference set and the current data point; and using the average of the data differences between multiple data points as the sparsity of the pixel. The steps for obtaining the data differences between each data point in the spatial neighboring reference set and the current data point include: calculating the value difference based on the differences between each data point in the spatial neighboring reference set and each data point in the current data point; calculating the directional difference based on the cosine angle between each data point in the spatial neighboring reference set and the current data point; and taking the product of the value difference and the directional difference as the data difference. The steps for calculating the difference between data points in a spatial neighboring reference set and data points in the current data point include: for any data point in the spatial neighboring reference set, calculating the absolute value of the difference between the current data point and the data point in the spatial neighboring reference set; using the normalized result of the absolute value as the standard difference, and using the mean of the standard differences of the data points as the value difference. The steps for constructing the degree of trend anomaly based on the initial anomaly scores and changes in data values ​​of data points in a temporally adjacent reference set include: forming a temporally reference point pair by any two adjacent data points in the temporally adjacent reference set; constructing a local score difference based on the difference in the initial anomaly scores of the temporally adjacent reference point pair; constructing a local data change based on the differences in various data points of the two data points in the temporally adjacent reference set; and constructing the degree of trend anomaly based on the local data change and the local score difference. The sum of local data changes is taken as the overall data change, the sum of multiple local score differences is taken as the overall score difference, the product of the overall score difference and the overall data change is taken as the anomaly index, and the normalized result of the anomaly index is taken as the degree of trend anomaly.

2. The method for detecting abnormalities in a refrigeration valve according to claim 1, characterized in that, The steps for constructing a spatial proximity reference set include: for any data point, obtaining a predetermined number of data points that are spatially close to the data point as initial data points, and taking the mean of the Euclidean distance between the initial data points and the current data point as the proximity distance; and taking the set of data points whose Euclidean distance to the current data point is less than the proximity distance as the spatial proximity reference set.

3. The method for detecting abnormalities in a refrigeration valve according to claim 1, characterized in that, The steps for obtaining a dynamic threshold by adjusting the preset initial anomaly threshold based on the degree of trend anomaly and sparsity include: using the ratio of the sparsity of data points to the degree of trend anomaly as a first control index, and using the mean of the first control index corresponding to each data point as a second control index; using the ratio of the normalized result of the first control index to the second control index as an adjustment factor, and using the product of the adjustment factor and the initial anomaly threshold as the dynamic threshold.

4. An anomaly detection system for a refrigeration valve, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement an anomaly detection method for a refrigeration valve according to any one of claims 1-3.

Images