Power consumption anomaly detection method and electronic device

Abstract

The application discloses a power consumption anomaly detection method and electronic equipment, and relates to the technical field of testing, which comprises the following steps: determining a power consumption distribution information based on a current performance index; the power consumption distribution information is automatically adjusted based on load change, so that the increase of power consumption caused by the improvement of the performance index will not be misjudged as an anomaly, and the false positive rate can be reduced. In addition, the abnormal index cumulative value of multiple periods is combined for judgment, so that only in the case that the actual power consumption value continuously deviates, the cumulative value will exceed the preset alarm threshold, and the abnormal judgment result is more accurate. In the case that the abnormal index cumulative value exceeds the preset alarm threshold, recheck is performed based on the second power consumption distribution information, the second power consumption distribution information can reflect the long-term power consumption distribution of the target device, so that whether the target device is abnormal can be accurately identified by comparing the actual power consumption value with the second power consumption distribution information, and the false positive rate is reduced.

Description

Technical Field

[0001] This application relates to the field of testing technology, and in particular to methods and electronic devices for detecting abnormal power consumption. Background Technology

[0002] As data centers continue to expand in scale, power consumption monitoring of electronic nodes within computing clusters plays an increasingly important role in operation and maintenance management. For example, by monitoring and analyzing power consumption in real time, it is possible to promptly detect whether a node is in an abnormal state, thereby preventing the escalation of hardware failures or resource waste caused by abnormal energy consumption. Currently, common methods for detecting power consumption anomalies mainly include judgment methods based on fixed thresholds; however, such schemes have a high false alarm rate. Summary of the Invention

[0003] This application provides a power consumption anomaly detection method, apparatus, electronic device, storage medium, and program product to solve the problem of high false alarm rate in related technologies.

[0004] This application provides a method for detecting abnormal power consumption, including: In the current cycle, collect the actual power consumption and performance data of the target device; The performance index data is input into the pre-built first quantile regression model to obtain the first power consumption distribution information output by the first quantile regression model. The first power consumption distribution information is used to indicate the power consumption distribution information of the target device under normal state under the performance index data. Based on the actual power consumption value and the first power consumption distribution information, determine the abnormal indicator value of the target device; Based on the cumulative abnormal indicator values ​​of the previous period before the current period and the abnormal indicator values ​​of the current period, determine the cumulative abnormal indicator value of the target device in the current period. If the cumulative value of the abnormal indicator of the target device in the current period is greater than the preset alarm threshold, the target device is determined to be in an abnormal state based on the actual power consumption value and the preset second power consumption distribution information. The second power consumption distribution information is determined based on the actual power consumption values ​​of multiple periods before the current period. If the target device is determined to be in an abnormal state, execute the first alarm operation.

[0005] This application also provides a power consumption anomaly detection device, including: The acquisition module is used to collect the actual power consumption and performance index data of the target device in the current cycle; The output module is used to input performance index data into a pre-built first quantile regression model to obtain the first power consumption distribution information output by the first quantile regression model. The first power consumption distribution information is used to indicate the power consumption distribution information of the target device under normal conditions under the performance index data. The determination module is used to determine the abnormal indicator value of the target device based on the actual power consumption value and the first power consumption distribution information; to determine the cumulative abnormal indicator value of the target device in the current period based on the cumulative abnormal indicator value of the previous period before the current period and the abnormal indicator value of the current period; and if it is determined that the cumulative abnormal indicator value of the target device in the current period is greater than the preset alarm threshold, to determine whether the target device is in an abnormal state based on the actual power consumption value and the preset second power consumption distribution information, wherein the second power consumption distribution information is determined based on the actual power consumption values ​​of multiple periods before the current period. The alarm module is used to execute the first alarm operation when it is determined that the target device is in an abnormal state.

[0006] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for implementing the steps of any of the above-described power consumption anomaly detection methods when executing the computer program.

[0007] This application also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of any of the above-described power consumption anomaly detection methods.

[0008] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described power consumption anomaly detection methods.

[0009] This application, after collecting performance index data and actual power consumption values, can determine the first power consumption distribution information where the target device's power consumption is normal under the current performance index data based on the first quantile regression model. Then, the actual power consumption value is compared with the first power consumption distribution information to determine the abnormal index value of the current actual power consumption. To accurately determine whether the target device is abnormal, this solution does not only judge whether the instantaneous actual power consumption value is normal, but also evaluates it through the cumulative value of abnormal indicators. Therefore, this solution calculates a cumulative value of abnormal indicators in the current period. If the cumulative value of abnormal indicators is greater than a preset alarm threshold, the target device is considered to be highly likely to be abnormal. To further accurately determine whether the target device is abnormal, this solution also compares the actual power consumption value with the second power consumption distribution information of historical periods to determine whether it conforms to the long-term operating characteristics of the target device. Based on this comparison result, it determines whether it is abnormal, and executes the first alarm operation in the case of abnormality.

[0010] Because the load on the target device constantly changes, using a fixed threshold can easily misjudge normal load increases as abnormal, leading to a large number of false alarms. This solution, however, determines a power consumption distribution based on current performance data. This power consumption distribution automatically adjusts based on load changes; therefore, power consumption increases caused by performance improvements will not be misjudged as abnormal, reducing the false alarm rate. Furthermore, this solution does not perform single-point judgments but combines the cumulative values ​​of abnormal indicators over multiple periods. This ensures that only when the actual power consumption value continuously deviates will the cumulative value exceed the preset alarm threshold, resulting in more accurate anomaly detection. Further, when the cumulative value of an abnormal indicator exceeds the preset alarm threshold, this solution does not immediately issue an alarm. Instead, it verifies the situation based on a second power consumption distribution information constructed through long-term statistics to determine whether the target device is abnormal. This second power consumption distribution information reflects the long-term power consumption distribution of the target device. Therefore, by comparing the actual power consumption value with the second power consumption distribution information, it can accurately identify whether the target device is experiencing an anomaly or is in a special event processing state, further reducing the false alarm rate. Attached Figure Description

[0011] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A schematic diagram of a computing cluster architecture provided for an embodiment of this application; Figure 2 A flowchart illustrating a power consumption anomaly detection method provided in an embodiment of this application; Figure 3 A schematic diagram illustrating the factors to consider in a power consumption anomaly detection method provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a power consumption anomaly detection device provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0014] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0015] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] The power consumption anomaly detection method provided in this application can be applied in computing clusters, such as... Figure 1 As shown, a computing cluster can include multiple electronic nodes, including but not limited to computing nodes (e.g., servers), switching nodes (e.g., switches), and storage nodes (e.g., storage devices). The motherboards of the computing nodes, switching nodes, and storage nodes are all equipped with hardware management and control components, such as a Baseboard Management Controller (BMC) and an ILO. Each hardware management and control component can be used to detect whether its assigned node is abnormal. Alternatively, the computing cluster can include monitoring nodes and service nodes. The monitoring nodes can be used to detect whether each service node is abnormal, and the service nodes can be any of the aforementioned electronic nodes.

[0017] Embodiments of this application provide a method for detecting abnormal power consumption in electronic nodes, which can be executed by a substrate management controller or a monitoring node. The following example uses a monitoring node. Figure 2 As shown, the specific processing steps of the power consumption anomaly detection method may include: Step S201: In the current cycle, collect the actual power consumption value and performance index data of the target device.

[0018] The target device can be any service node in the aforementioned computing cluster. Performance metrics data may include one or more of the following: CPU utilization, memory bandwidth utilization, disk I / O latency, network throughput, etc.

[0019] Specifically, within the current cycle, the monitoring node can extract the actual power consumption value from the power supply unit (PSU) of the target device according to a preset sampling period. The target device can collect performance indicator data from its own performance counters through the operating system interface or the hardware low-level interface, and send the collected performance indicator data to the monitoring node. The monitoring node can organize the performance indicator data into the form of a feature vector. For example, the preset sampling period can be 1 second or less.

[0020] Step S202: Input the performance index data into the pre-built first quantile regression model to obtain the first power consumption distribution information output by the first quantile regression model.

[0021] The first power consumption distribution information can be used to indicate the power consumption distribution information of the target device when it is in a normal state under the performance index data.

[0022] Specifically, the monitoring node can input the feature vector into the first quantile regression model to obtain the first power consumption distribution information output by the first quantile regression model. In this scheme, the quantiles used in the first quantile regression model can include 95% and 5%.

[0023] Step S203: Determine the abnormal indicator value of the target device based on the actual power consumption value and the first power consumption distribution information.

[0024] Specifically, the monitoring node can compare the actual power consumption value with the first power consumption distribution information, determine the comparison result, and determine the corresponding abnormal indicator value based on the comparison result.

[0025] Optionally, the first power consumption distribution information may include a first maximum power consumption value and a first minimum power consumption value. The first maximum power consumption value may be used to indicate the statistical upper limit that the target device should not exceed when in a normal state under the performance indicators of the current period, for example, it may be the power consumption value corresponding to the 95th percentile. The first minimum power consumption value is used to indicate the statistical lower limit that the target device should not fall below when in a normal state under the performance indicators of the current period, for example, it may be the power consumption value corresponding to the 5th percentile. Accordingly, the monitoring node may compare the actual power consumption value, the first maximum power consumption value, and the first minimum power consumption value to determine the comparison result, and then determine the corresponding abnormal indicator value based on the comparison result. For example, step S203 may include the following steps: Step 1: Determine the power consumption reference difference based on the first maximum power consumption value and the first minimum power consumption value.

[0026] Step 2: Determine the first power consumption difference based on the actual power consumption value and the first maximum power consumption value.

[0027] Step 3: Determine the second power consumption difference based on the actual power consumption value and the first minimum power consumption value.

[0028] Step 4: Determine the abnormal indicator values ​​of the target device in the current period based on the first power consumption difference, the second power consumption difference, and the power consumption baseline difference.

[0029] Specifically, the monitoring node can determine the difference between the first maximum power consumption value and the first minimum power consumption value as the power consumption baseline difference, the difference between the actual power consumption value and the first maximum power consumption value as the first power consumption difference, and the difference between the first minimum power consumption value and the actual power consumption value as the second power consumption difference. Finally, the first power consumption difference and the second power consumption difference can be compared with the power consumption baseline difference to determine the comparison result, and the corresponding abnormal indicator value can be determined based on the comparison result.

[0030] Thus, in some cases, a slight deviation of the actual power consumption from the boundary may not necessarily indicate an anomaly, and directly determining an anomaly based solely on exceeding the boundary is inaccurate. This solution does not directly identify an anomaly by comparing the actual power consumption with the boundary value and determining whether the actual power consumption exceeds or falls below the boundary. Instead, it provides an anomaly index value to indicate the degree of anomaly, making it more accurate.

[0031] Optionally, the difference between the actual power consumption value and the first maximum power consumption value can represent the extent to which the actual power consumption deviates from the normal upper bound; the difference between the actual power consumption value and the first minimum power consumption value can represent the extent to which the actual power consumption deviates from the normal lower bound; and the difference between the first maximum power consumption value and the first minimum power consumption value can represent the width of the normal power consumption range under the current performance indicators. This scheme uses the width of the normal power consumption range as a normalization benchmark and normalizes the first power consumption difference and the second power consumption difference respectively, which can convert the magnitude into a dimensionless relative proportion. Accordingly, step four may include: Step 1: Calculate the first ratio between the first power consumption difference and the power consumption reference difference.

[0032] Step 2: Calculate the second ratio between the second power consumption difference and the power consumption reference difference.

[0033] Step 3: Determine the maximum value between the first ratio and the first preset value as the first abnormal indicator value.

[0034] Step 4: Determine the maximum value between the second ratio and the first preset value as the second abnormal indicator value.

[0035] Among them, the first abnormal indicator value and the second abnormal indicator value together constitute the abnormal indicator value of the target equipment in the current period.

[0036] Specifically, the monitoring node can determine the ratio between the first power consumption difference and the power consumption reference difference as the first ratio, and the ratio between the second power consumption difference and the power consumption reference difference as the second ratio. The first preset value can be 0. The monitoring node can determine the maximum value between the first ratio and the first preset value as the first abnormal indicator value, and the maximum value between the second ratio and the first preset value as the second abnormal indicator value. In this way, the first abnormal indicator value and the second abnormal indicator value constitute the abnormal indicator values ​​of the target device in the current period, which can respectively represent the magnitude of deviation from the normal range.

[0037] For example, the first abnormal indicator value can be expressed as follows: (1) The second abnormal indicator value can be expressed as follows: (2) in, This represents the first abnormal indicator value, and t represents the current period. This represents the actual power consumption value. This represents the first maximum power consumption value. First minimum power consumption value, This indicates the second abnormal indicator value. and All parameters are dimensionless, while other parameters are dimensional; for example, the unit could be watts.

[0038] Step S204: Determine the cumulative value of abnormal indicators of the target device in the current period based on the cumulative value of abnormal indicators of the previous period before the current period and the value of abnormal indicators in the current period.

[0039] Specifically, the monitoring node can calculate the cumulative value of abnormal indicators in each period. In the current period, the cumulative value of abnormal indicators in the current period can be determined by the sum of the cumulative value of abnormal indicators in the previous period and the cumulative value of abnormal indicators in the current period. When the current period is the first period, the cumulative value of abnormal indicators in the previous period can be 0.

[0040] Optionally, since abnormal indicator values ​​can include two categories, the cumulative abnormal indicator values ​​can also include two categories. For example, the cumulative abnormal indicator values ​​of the previous period can include a first cumulative abnormal indicator value and a second cumulative abnormal indicator value, and the cumulative abnormal indicator values ​​of the current period can include a third cumulative abnormal indicator value and a fourth cumulative abnormal indicator value. The first and third cumulative abnormal indicator values ​​are one type of cumulative abnormal indicator value, used to indicate information of continuous deviation from the upper bound of normal power consumption, and the second and fourth cumulative abnormal indicator values ​​are another type of cumulative abnormal indicator value, used to indicate information of continuous deviation from the lower bound of normal power consumption. Accordingly, step S205 can include: Step 1: Determine the cumulative value of the first candidate abnormal indicator based on the cumulative value of the first abnormal indicator, the value of the first abnormal indicator, and the preset tolerance coefficient.

[0041] The preset tolerance coefficient is used to suppress transient noise fluctuations.

[0042] Step 2: Determine the maximum value between the first candidate abnormal indicator accumulation and the second preset value as the third abnormal indicator accumulation value of the target device in the current period.

[0043] Step 3: Determine the cumulative value of the second candidate abnormal indicator based on the cumulative value of the second abnormal indicator, the value of the second abnormal indicator, and the preset tolerance coefficient.

[0044] Step 4: Determine the maximum value between the second candidate abnormal indicator accumulation and the second preset value as the fourth abnormal indicator accumulation value of the target device in the current period.

[0045] Among them, the cumulative values ​​of the third and fourth abnormal indicators constitute the cumulative values ​​of the abnormal indicators of the target equipment in the current period.

[0046] Specifically, the monitoring node can first calculate the sum of the first abnormal indicator cumulative value and the first abnormal indicator value, then subtract a preset tolerance coefficient from the sum to obtain the first candidate abnormal indicator cumulative value, and then determine the maximum value between the first candidate abnormal indicator cumulative value and the second preset value as the third abnormal indicator cumulative value. The second preset value can be 0.

[0047] Similarly, the monitoring node can first calculate the sum of the second abnormal indicator cumulative value and the second abnormal indicator value, then subtract the preset tolerance coefficient from the sum to obtain the second candidate abnormal indicator cumulative value, and then determine the maximum value between the second candidate abnormal indicator cumulative value and the second preset value as the fourth abnormal indicator cumulative value.

[0048] For example, the cumulative value of the third anomaly indicator can be expressed as follows: (3) The cumulative value of the fourth abnormal indicator can be expressed as follows: (4) in, This represents the cumulative value of the third abnormal indicator. This represents the cumulative value of abnormal indicators. This indicates the first abnormal indicator value. This indicates the second abnormal indicator value. This represents a preset tolerance coefficient, for example, the value range can be [0.05, 0.1], which ensures that fluctuations below the preset tolerance coefficient do not participate in the accumulation process. This represents the cumulative value of the first candidate anomaly indicator. This is the cumulative value of the second candidate anomaly index. The parameters in formulas (3) and (4) are dimensionless parameters.

[0049] In this way, if the actual power consumption in a certain cycle exceeds the preset power consumption threshold, but returns to normal in the next cycle, an alarm will be triggered immediately using the fixed threshold method. However, in this solution, due to the decay effect of subsequent normal values, the accumulated amount may quickly drop back and may not exceed the threshold, thus avoiding an alarm and reducing the number of false alarms.

[0050] Step S205: If it is determined that the cumulative value of the abnormal indicators of the target device in the current period is greater than the preset alarm threshold, the target device is determined to be in an abnormal state based on the actual power consumption value and the preset second power consumption distribution information.

[0051] The second power consumption distribution information is determined based on the actual power consumption values ​​of multiple previous cycles. These multiple cycles can be cycles in which the target device is determined not to be in an abnormal state, ensuring that the second power consumption distribution information is the power consumption distribution information of the target device in a normal state. This, in turn, ensures that subsequent judgments on whether the target device is in an abnormal state can be more accurate. The preset alarm threshold range can be [3, 5].

[0052] Specifically, the monitoring node can compare the cumulative value of abnormal indicators in the current period with the preset alarm threshold. If it is determined that the cumulative value of abnormal indicators in the current period is greater than the preset alarm threshold, the actual power consumption value can be compared with the preset second power consumption distribution information to obtain the comparison result. Based on the comparison result, the state corresponding to the comparison result can be determined. This state can be used to indicate whether the target device is in an abnormal state.

[0053] Optionally, step S205 may include: Step 1: If the cumulative value of the abnormal indicator in the current period is greater than the preset alarm threshold, determine whether the actual power consumption value matches the second power consumption distribution information.

[0054] Step two: If the actual power consumption value does not match the second power consumption distribution information, the target device is determined to be in an abnormal state.

[0055] Alternatively, in step three, if the actual power consumption value matches the second power consumption distribution information, it is determined that the target device is not in an abnormal state.

[0056] Specifically, the cumulative value of abnormal indicators in the current period can include the cumulative value of the third abnormal indicator and the cumulative value of the fourth abnormal indicator. The monitoring node can compare the cumulative value of the third abnormal indicator and the cumulative value of the fourth abnormal indicator with the preset alarm threshold respectively, obtain the comparison result, and then determine the status of the target device based on the comparison result.

[0057] For example, if the cumulative value of the third or fourth abnormal indicator exceeds the preset alarm threshold, it indicates that the actual power consumption value deviates too much from the upper or lower bound of power consumption. Since excessive deviation can have various causes, to avoid false alarms, the actual power consumption value can be compared with the second power consumption distribution information under normal operation. If the actual power consumption value is not within the second power consumption distribution information, the target device is determined to be in an abnormal state; for example, it may be due to a hardware failure such as fan failure or leakage current. If the actual power consumption value is within the second power consumption distribution information, the target device is determined not to be in an abnormal state.

[0058] Alternatively, the preset alarm threshold may include a first preset alarm threshold and a second preset alarm threshold. When it is determined that the cumulative value of the third abnormal indicator is greater than the first preset alarm threshold, or the cumulative value of the fourth abnormal indicator is greater than the preset alarm threshold and greater than the second preset alarm threshold, it is determined whether the actual power consumption value matches the second power consumption distribution information.

[0059] Step S206: If it is determined that the target device is in an abnormal state, execute the first alarm operation.

[0060] Specifically, when a monitoring node determines that a target device is in an abnormal state, it can directly execute a preset first alarm operation. For example, it can send alarm information corresponding to the first alarm operation to the management device and indicate the level of the first alarm operation. The management device can be a terminal device of a technician.

[0061] The power consumption anomaly detection method of this application, after collecting performance index data and actual power consumption values, can determine the first power consumption distribution information where the power consumption of the target device is normal under the current performance index data based on a first quantile regression model. Then, the actual power consumption value is compared with the first power consumption distribution information to determine the current abnormal index value of the actual power consumption. In order to accurately determine whether the target device is abnormal, this solution does not only judge whether the instantaneous actual power consumption value is normal, but also evaluates it through the cumulative value of abnormal index. Therefore, this solution calculates an abnormal index cumulative value in the current period. If the abnormal index cumulative value is greater than a preset alarm threshold, it is considered that the target device is very likely to be abnormal. In order to further accurately determine whether the target device is abnormal, this solution also compares the actual power consumption value with the second power consumption distribution information of historical periods to determine whether it conforms to the long-term operating characteristics of the target device. Based on this comparison result, it determines whether it is abnormal, and executes the first alarm operation in the case of abnormality.

[0062] When the load on the target device changes continuously, using a fixed threshold can easily misjudge normal load increases as abnormal, leading to a large number of false alarms. This solution, however, determines a power consumption distribution based on current performance metrics. This power consumption distribution automatically adjusts based on load changes; therefore, increases in power consumption due to improved performance metrics will not be misjudged as abnormal, reducing the false alarm rate.

[0063] Furthermore, this solution does not make a single-point judgment, but combines the cumulative values ​​of abnormal indicators over multiple periods to make a judgment. This ensures that the cumulative value will only exceed the preset alarm threshold when the actual power consumption value continues to deviate, making the abnormal judgment result more accurate.

[0064] Furthermore, when the cumulative value of the abnormal indicator exceeds the preset alarm threshold, this solution does not immediately issue an alarm. Instead, it reviews the second power consumption distribution information constructed based on long-term statistics to determine whether the target device is abnormal. The second power consumption distribution information can reflect the long-term power consumption distribution of the target device. Therefore, by comparing the actual power consumption value with the second power consumption distribution information, it is possible to accurately identify whether the target device is abnormal or in a special event processing state, thereby further reducing the false alarm rate.

[0065] In some optional implementations, if the actual power consumption value is determined to match the second power consumption distribution information, the monitoring node may also perform the following steps: Step 1: Mark the current period as a global migration event.

[0066] Step 2: Count at least one candidate cycle that is marked as a global migration event.

[0067] Step 3: Determine the average power consumption and standard deviation of the global migration event based on the actual power consumption values ​​of the target device in at least one candidate period.

[0068] Step 4: Update the second power distribution information based on the average power consumption and standard deviation of the global migration events.

[0069] Specifically, if the actual power consumption value is determined to be within the second power consumption distribution information, it indicates that the target device is currently undergoing a global migration process. In this case, the current period can be marked as a global migration event. A global migration event refers to a shift in the overall power consumption distribution of the target device, but this change is a normal state migration caused by global factors such as business load, resource configuration, or operating environment, rather than an anomaly caused by a local hardware failure. For example, a global migration event could be an event where the overall business volume increases due to promotional activities, leading to a surge in power consumption. When it is determined in a subsequent period that the target device is no longer in a global migration process, statistical operations can be performed to identify at least one candidate period marked as a global migration event. Then, based on the actual power consumption values ​​corresponding to each of the at least one candidate period, the average power consumption value and standard deviation of the global migration event can be determined. Finally, the monitoring node can determine the third minimum power consumption value by subtracting a preset multiple of the standard deviation from the average power consumption value, and determine the third maximum power consumption value by adding a preset multiple of the standard deviation to the average power consumption value. The third minimum power consumption value and the third maximum power consumption value are then determined as the latest second power consumption distribution information.

[0070] The second power consumption distribution information may include a fourth maximum power consumption value and a fourth minimum power consumption value, represented as follows: , The preset multiple can be 3. Accordingly, the second power consumption distribution information can be represented as follows: ,in, This is the average of the actual power consumption values ​​over multiple cycles. This represents the standard deviation of the actual power consumption values ​​over multiple cycles. This solution can... Update to the average power consumption value of the global migration event. Updated to the standard deviation of global migration events.

[0071] In this way, by statistically analyzing the actual power consumption values ​​of multiple candidate periods, the impact of single-point fluctuations can be eliminated. Furthermore, the average power consumption value and standard deviation obtained based on multiple candidate periods can better represent the true and stable power consumption level of the new state after the migration, thereby ensuring the accuracy of baseline updates.

[0072] Based on this, this solution can also determine the start time and duration of the global migration event according to each candidate period. In addition, for each period, the start time, duration, average power consumption, standard deviation, and trigger direction of the cumulative value of abnormal indicators (e.g., upward or downward; upward means the cumulative value of the third abnormal indicator is greater than the preset alarm threshold, and downward means the cumulative value of the fourth abnormal indicator is greater than the preset alarm threshold) of the global migration event can be recorded as a data entry for the convenience of subsequent operation and maintenance.

[0073] In some optional implementations, based on the determination that the target device is in a global migration event, the preset alarm threshold can also be updated. The update operation can be as follows: Step 1: Determine candidate alarm thresholds based on preset baseline alarm thresholds, power consumption baseline difference, standard deviation, preset reference power consumption difference, and reference standard deviation.

[0074] Step 2: Determine the target alarm threshold from the candidate alarm thresholds, the preset maximum alarm threshold, and the minimum alarm threshold.

[0075] Step 3: Update the preset alarm threshold to the target alarm threshold.

[0076] Specifically, the monitoring node can calculate the candidate alarm threshold based on the preset baseline alarm threshold, power consumption baseline difference, standard deviation, preset reference power consumption difference, and reference standard deviation.

[0077] Then, the monitoring node can compare the candidate alarm thresholds, the preset maximum alarm threshold, and the minimum alarm threshold to determine the target alarm threshold. For example, the target alarm threshold can be expressed as follows: (5) in, Indicates the target alarm threshold. This represents the baseline alarm threshold, and its value can range from 3 to 10. The standard deviation of global migration events. Indicates the power consumption reference difference. Indicates the reference standard deviation. For reference power consumption difference, This represents the minimum alarm threshold, and its value can range from 1 to 3. This represents the maximum alarm threshold, and its value can range from 10 to 30. The preset minimum value can be 10. -6 Used to prevent division by zero. This represents the candidate alarm threshold. In formula (5) , , , All are dimensionless parameters. , , , , For dimensional parameters, the unit can be watt.

[0078] In this way, this solution calculates an adjustment coefficient by comparing the standard deviation and power consumption baseline difference that change in real time with the performance index data, and then provides a new candidate alarm threshold based on the adjustment coefficient. To avoid situations where the power consumption baseline difference or standard deviation is extremely small, resulting in extremely large or small thresholds and causing no alarm or frequent alarms, this solution uses constraints on the maximum and minimum alarm thresholds to avoid such situations.

[0079] In some optional implementations, in the current cycle, as long as it is determined that the target device is not in an abnormal state, the average power consumption value and standard deviation can be recalculated using the actual power consumption value of the current cycle and the actual power consumption values ​​of multiple cycles before the current cycle (cycles in which the target device was not in an abnormal state), and the second power consumption distribution information can be updated based on the average power consumption value and standard deviation.

[0080] In some optional implementations, to ensure the accuracy of alarm operations, the results of quantile regression models updated at different frequencies can be compared. Maintaining consistency across different results can improve the reliability of alarm operations. Accordingly, if the target device is determined to be in an abnormal state, the monitoring node may not execute the first alarm operation, but instead execute the following specific steps: Step 1: Input the performance index data into the pre-built second quantile regression model to obtain the third power consumption distribution information output by the second quantile regression model.

[0081] The update frequency of the second quantile regression model is lower than that of the first quantile regression model, and the second power consumption distribution information includes the second maximum power consumption value and the second minimum power consumption value.

[0082] For example, after obtaining each set of feature vectors and their corresponding actual power consumption values, the first quantile regression model is updated using gradient descent. Specifically, if the actual power consumption value is greater than the first maximum power consumption value, the gradient can be determined by the product of the feature vector and the first quantile (i.e., the quantile corresponding to the first maximum power consumption value, which could be 95%). If the actual power consumption value is less than the first minimum power consumption value, the gradient can be determined by the product of the feature vector and the target difference (the first quantile minus 1). Furthermore, for any model parameter in the first quantile regression model, the difference between the old model parameter and the gradient can be used to determine the new model parameter. The dimension of the model parameters is the same as the dimension of the feature vectors.

[0083] Alternatively, the monitoring node can be configured with an incremental training buffer. In each cycle, as long as it is determined that the target device is not in an abnormal state during that cycle, the feature vector and actual power consumption value for that cycle can be stored as a sample pair in the incremental training buffer. After acquiring a first preset number of sample pairs, an update operation is performed on the first quantile regression model. During each update operation, for each sample pair, the gradient can be calculated in a similar manner to the above. Then, after determining the gradients corresponding to the preset number of sample pairs, the average gradient is calculated. Finally, the average gradient is used to update the model parameters, and the update method is similar to the aforementioned update method, which will not be elaborated here.

[0084] For the second quantile regression model, an update operation is performed once every second preset number of sample pairs. The second preset number is greater than the first preset number.

[0085] Step 2: If the actual power consumption value is determined to be greater than the second maximum power consumption value, or the actual power consumption value is less than the second minimum power consumption value, then execute the second alarm operation.

[0086] The alarm level of the second alarm operation is higher than that of the first alarm operation.

[0087] Step 3: If the actual power consumption value is less than or equal to the second maximum power consumption value and greater than or equal to the second minimum power consumption value, execute the third alarm operation.

[0088] Among them, the alarm level of the third alarm operation is lower than the alarm level of the first alarm operation.

[0089] Specifically, if the actual power consumption value is determined to be greater than the second maximum power consumption value, or less than the second minimum power consumption value, it indicates that the second quantile regression model has determined that the actual power consumption value of the target device in the current period has exceeded the limit, which is consistent with the determination of the first quantile regression model. Therefore, the alarm level can be increased, and a second alarm operation can be executed. For example, an alarm message corresponding to the second alarm operation can be sent to the management device, and the level of the second alarm operation can be indicated.

[0090] If the actual power consumption is determined to be less than or equal to the second maximum power consumption and greater than or equal to the second minimum power consumption, it indicates that the first quantile regression model has degenerated due to learning some recent abnormal features too quickly, resulting in inaccurate judgments. Therefore, the alarm level can be lowered, and a third alarm operation can be executed. For example, an alarm message corresponding to the third alarm operation can be sent to the management device, along with the level of the third alarm operation.

[0091] In this way, technicians can understand the severity of the current anomaly based on the alarm action level and perform maintenance operations accordingly, thus improving maintenance efficiency. Furthermore, because the second quantile regression model updates slowly and is not sensitive to short-term fluctuations, it can effectively filter out false alarms caused by momentary load fluctuations or overly rapid model learning. That is, only truly serious and persistent anomalies will trigger high-level alarms, reducing the false alarm rate. In addition, using two different quantile regression models to determine the consistency of results allows for more accurate execution of alarm actions, further reducing the false alarm rate.

[0092] In some optional implementations, for each sample pair sampled, if it is determined that the target device is in an abnormal state during the corresponding period, the sample pair is removed, or its weight value is reduced. If it is determined that the actual power consumption value for the corresponding period is less than or equal to the second maximum power consumption value, and the actual power consumption value is greater than or equal to the second minimum power consumption value, the weight value of the sample pair for the current period can be increased.

[0093] The values ​​that can be decreased or increased can be preset.

[0094] Reducing the weight of a sample pair can specifically be done by: In the process of calculating the average gradient, the weight value of the gradient for that sample pair is reduced.

[0095] Specifically, increasing the weight value of a sample pair can be achieved by: Copy the sample pairs from the current period multiple times and store them in the incremental training buffer. Alternatively, increase the weight of the gradient for that sample pair during the calculation of the average gradient.

[0096] This prevents abnormal samples from affecting the accuracy of the quantile regression model and avoids failure to execute alarm operations in abnormal situations, which could lead to target device failure.

[0097] In some optional implementations, before performing step S201 above, the monitoring node can collect performance index data and actual power consumption values ​​at a preset sampling period during the first preset time window after the computing cluster deployment (e.g., the first 24 hours after deployment). The collected data is then used to train the quantile regression model to obtain a first quantile regression model. Alternatively, the average and standard deviation of the actual power consumption values ​​within the preset time window can be calculated based on the collected data, and these average and standard deviations can be used as the second power consumption distribution information. Or, after collecting a specified number of sample pairs, such as 86,400 sample pairs, the monitoring node can use these sample pairs to perform training to obtain the first quantile regression model and calculate the average and standard deviation of the actual power consumption values ​​for the specified number of sample pairs, using this as the second power consumption distribution information.

[0098] Thus, when the computing cluster is first deployed, the model parameters are not fully trained, and the output power consumption boundaries may be completely inaccurate. If anomaly detection is enabled directly at this time, it will generate a large number of meaningless false alarms. This solution trains the quantile regression model with sufficient samples and determines the second power consumption distribution information, thereby establishing a more accurate reference benchmark. This leads to greater accuracy in subsequent anomaly detection and a lower false alarm rate.

[0099] The aforementioned preset alarm thresholds, maximum alarm thresholds, minimum alarm thresholds, reference power consumption difference, and reference standard deviation can be determined by technicians after analyzing performance indicator data and actual power consumption values ​​collected within a preset time window. These values ​​may vary depending on the server model, sampling frequency, and service type, and are adjusted by technicians according to the actual situation.

[0100] In some optional implementations, the aforementioned second power consumption distribution information may include multiple sub-distribution information at different time scales. For example, it may include short-term sub-distribution information, medium-term sub-distribution information, long-term sub-distribution information, and total distribution information. The historical durations corresponding to the short-term, medium-term, and long-term sub-distribution information increase sequentially. For instance, the short-term sub-distribution information may be determined based on actual power consumption values ​​collected over the past hour, the medium-term sub-distribution information may be determined based on actual power consumption values ​​collected over the past day, the long-term sub-distribution information may be determined based on actual power consumption values ​​collected over the past seven days, and the total distribution information may be determined based on all previously collected actual power consumption values. It should be noted that the actual power consumption values ​​all originate from the period during which the target device is in normal operating condition.

[0101] For each sub-distribution information, it can be determined whether the target device is in an abnormal state. If the judgment results for all sub-distribution information are yes, then the second alarm operation is executed. If only one or zero judgment results are yes, then the third alarm operation is executed. Otherwise, the first alarm operation is executed.

[0102] In this way, if the anomaly is confirmed at all time scales, it indicates a very serious and persistent problem, triggering the highest level of alert action. If one or zero anomalies are detected at only a very few time scales, it may be just transient noise or short-term fluctuations, triggering the lowest level of alert action. Other cases indicate that the anomaly has a certain degree of persistence, triggering an intermediate level of alert action. Furthermore, operations and maintenance personnel can quickly determine the processing priority based on the alert level, improving operational efficiency.

[0103] The above-described embodiments, such as Figure 3As shown, this scheme uses multiple factors, including the first quantile regression model, the cumulative value of abnormal indicators, the second power consumption distribution information, and the second quantile regression model, to perform layer-by-layer verification, which can reduce the false alarm rate.

[0104] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0105] Embodiments of this application also provide a power consumption anomaly detection device, such as... Figure 4 As shown, it includes: The acquisition module 410 is used to acquire the actual power consumption value and performance index data of the target device in the current cycle; The output module 420 is used to input performance index data into a pre-built first quantile regression model to obtain the first power consumption distribution information output by the first quantile regression model. The first power consumption distribution information is used to indicate the power consumption distribution information of the target device under normal conditions under the performance index data. The determination module 430 is used to determine the abnormal indicator value of the target device based on the actual power consumption value and the first power consumption distribution information; determine the cumulative abnormal indicator value of the target device in the current period based on the cumulative abnormal indicator value of the previous period before the current period and the abnormal indicator value of the current period; if it is determined that the cumulative abnormal indicator value of the target device in the current period is greater than the preset alarm threshold, determine whether the target device is in an abnormal state based on the actual power consumption value and the preset second power consumption distribution information, wherein the second power consumption distribution information is determined based on the actual power consumption values ​​of multiple periods before the current period; The alarm module 440 is used to execute the first alarm operation when it is determined that the target device is in an abnormal state.

[0106] In some optional implementations, the first power consumption distribution information includes a first maximum power consumption value and a first minimum power consumption value; Module 430 is specifically used for: The power consumption reference difference is determined based on the first maximum power consumption value and the first minimum power consumption value; The first power consumption difference is determined based on the actual power consumption value and the first maximum power consumption value; The second power consumption difference is determined based on the actual power consumption value and the first minimum power consumption value; Based on the first power consumption difference, the second power consumption difference, and the power consumption baseline difference, the abnormal indicator values ​​of the target device in the current period are determined.

[0107] In some alternative implementations, the determining module 430 is specifically used for: Calculate the first ratio between the first power consumption difference and the power consumption reference difference; Calculate the second ratio between the second power consumption difference and the power consumption reference difference; The maximum value between the first ratio and the first preset value is determined as the first abnormal indicator value; The maximum value between the second ratio and the first preset value is determined as the second abnormal indicator value; Among them, the first abnormal indicator value and the second abnormal indicator value together constitute the abnormal indicator value of the target equipment in the current period.

[0108] In some optional implementations, the cumulative value of the abnormal indicators of the target device in the previous period includes the cumulative value of the first abnormal indicator and the cumulative value of the second abnormal indicator. Module 430 is specifically used for: The first candidate abnormal indicator cumulative value is determined based on the first abnormal indicator cumulative value, the first abnormal indicator value, and the preset tolerance coefficient, wherein the preset tolerance coefficient is used to suppress transient noise fluctuations. The maximum value between the first candidate abnormal indicator accumulation and the second preset value is determined as the third abnormal indicator accumulation value of the target device in the current period; The cumulative value of the second candidate abnormal indicator is determined based on the cumulative value of the second abnormal indicator, the value of the second abnormal indicator, and the preset tolerance coefficient. The maximum value between the second candidate abnormal indicator accumulation and the second preset value is determined as the fourth abnormal indicator accumulation value of the target device in the current period; Among them, the cumulative values ​​of the third and fourth abnormal indicators constitute the cumulative values ​​of the abnormal indicators of the target equipment in the current period.

[0109] In some alternative implementations, the determining module 430 is specifically used for: If the cumulative value of abnormal indicators in the current period is determined to be greater than the preset alarm threshold, it is determined whether the actual power consumption value matches the second power consumption distribution information. If the actual power consumption value does not match the second power consumption distribution information, the target device is determined to be in an abnormal state. Alternatively, if the actual power consumption value matches the second power consumption distribution information, it can be determined that the target device is not in an abnormal state.

[0110] In some alternative implementations, the device further includes an update module for: If the actual power consumption value matches the second power consumption distribution information, the current period is marked as a global migration event; Statistics are marked as at least one candidate period for global migration events; Based on the actual power consumption values ​​of the target device in at least one candidate period, determine the average power consumption value and standard deviation of the global migration event; The second power distribution information is updated based on the average power consumption and standard deviation of the global migration events.

[0111] In some alternative implementations, the update module is also used for: Candidate alarm thresholds are determined based on preset baseline alarm thresholds, power consumption baseline differences, standard deviations, preset reference power consumption differences, and reference standard deviations. The target alarm threshold is determined based on the candidate alarm thresholds, the preset maximum alarm threshold, and the minimum alarm threshold. Update the preset alarm threshold to the target alarm threshold.

[0112] In some optional implementations, the target alarm threshold is expressed as follows:

[0113] in, Indicates the target alarm threshold. Indicates the baseline alarm threshold. The standard deviation of global migration events. Indicates the power consumption reference difference. Indicates the reference standard deviation. For reference power consumption difference, This represents the minimum alarm threshold. Indicates the maximum alarm threshold. This is a preset minimum value used to prevent division by zero.

[0114] In some alternative implementations, the alarm module 440 is further configured to: When it is determined that the target device is in an abnormal state, the performance index data is input into the pre-built second quantile regression model to obtain the third power consumption distribution information output by the second quantile regression model. The update frequency of the second quantile regression model is lower than that of the first quantile regression model. The second power consumption distribution information includes the second maximum power consumption value and the second minimum power consumption value. If it is determined that the actual power consumption value is greater than the second maximum power consumption value, or the actual power consumption value is less than the second minimum power consumption value, a second alarm operation is performed, wherein the alarm level of the second alarm operation is higher than the alarm level of the first alarm operation. Alternatively, if it is determined that the actual power consumption value is less than or equal to the second maximum power consumption value and the actual power consumption value is greater than or equal to the second minimum power consumption value, a third alarm operation is performed, wherein the alarm level of the third alarm operation is lower than the alarm level of the first alarm operation.

[0115] For a description of the features in the embodiment corresponding to the power consumption anomaly detection device, please refer to the relevant description of the embodiment corresponding to the power consumption anomaly detection method, which will not be repeated here.

[0116] Embodiments of this application also provide an electronic device, such as... Figure 5 As shown, the device includes a memory 10 and a processor 20. The memory 10 stores a computer program, and the processor 20 is configured to run the computer program to perform the steps in any of the above-described embodiments of the power consumption anomaly detection method. The electronic device may be the monitoring node or hardware management control component described above.

[0117] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above-described embodiments of the power consumption anomaly detection method when running.

[0118] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0119] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described embodiments of the power consumption anomaly detection method.

[0120] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above-described power consumption anomaly detection method embodiments.

[0121] Any of the components, modules, units, parts, methods, and operations described herein can be implemented using software, firmware, hardware (fixed logic circuit systems), manual processing, or any combination thereof. Alternatively or additionally, any functionality described herein can be performed at least in part by one or more hardware logic components, such as, but not limited to, a central processing unit, a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), an application-specific standard product (ASSP), a system-on-a-chip (SoC), a complex programmable logic device (CPLD), a microcontroller unit (MCU), etc. The terms "system," "computing device," or "apparatus" as used herein encompass various means, devices, and machines for processing data, including, for example, one or more programmable processors, computers, SoCs, or combinations thereof. The apparatus may also include code that creates an execution environment for the computer program in question, such as code constituting processor firmware, a protocol stack, a database management system, an operating system, a cross-platform runtime environment, a virtual machine, or a combination thereof. The aforementioned computer program (also known as a program, software, software application, App, script, or code) can be written in any form of programming language, including compiled or interpreted languages, declarative or procedural languages, and can be deployed in any form, including as a standalone program or as a module, component, subroutine, object, or other unit suitable for a computing environment.

[0122] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0123] The above provides a detailed description of the power consumption anomaly detection method, apparatus, electronic device, storage medium, and program product provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A method for detecting abnormal power consumption, characterized in that, include: In the current cycle, collect the actual power consumption and performance data of the target device; The performance index data is input into a pre-constructed first quantile regression model to obtain the first power consumption distribution information output by the first quantile regression model. The first power consumption distribution information is used to indicate the power consumption distribution information of the target device when it is in a normal state under the performance index data. Based on the actual power consumption value and the first power consumption distribution information, the abnormal indicator value of the target device is determined; Based on the pre-acquired cumulative value of abnormal indicators from the previous period before the current period and the abnormal indicator value of the current period, the cumulative value of abnormal indicators of the target device in the current period is determined. If it is determined that the cumulative value of the abnormal indicator of the target device in the current period is greater than the preset alarm threshold, the target device is determined to be in an abnormal state based on the actual power consumption value and the preset second power consumption distribution information, wherein the second power consumption distribution information is determined based on the actual power consumption values ​​of multiple periods before the current period; If the target device is determined to be in the abnormal state, a first alarm operation is performed.

2. The power consumption anomaly detection method according to claim 1, characterized in that, The first power consumption distribution information includes a first maximum power consumption value and a first minimum power consumption value; The step of determining the abnormal indicator value of the target device based on the actual power consumption value and the first power consumption distribution information includes: The power consumption reference difference is determined based on the first maximum power consumption value and the first minimum power consumption value; The first power consumption difference is determined based on the actual power consumption value and the first maximum power consumption value; The second power consumption difference is determined based on the actual power consumption value and the first minimum power consumption value; Based on the first power consumption difference, the second power consumption difference, and the power consumption benchmark difference, the abnormal indicator value of the target device in the current period is determined.

3. The power consumption anomaly detection method according to claim 2, characterized in that, The step of determining the abnormal indicator value of the target device in the current period based on the first power consumption difference, the second power consumption difference, and the power consumption reference difference includes: Calculate a first ratio between the first power consumption difference and the power consumption reference difference; Calculate a second ratio between the second power consumption difference and the power consumption reference difference; The maximum value between the first ratio and the first preset value is determined as the first abnormal indicator value; The maximum value between the second ratio and the first preset value is determined as the second abnormal indicator value; The first abnormal indicator value and the second abnormal indicator value together constitute the abnormal indicator value of the target device in the current period.

4. The power consumption anomaly detection method according to claim 3, characterized in that, The cumulative value of the abnormal indicators of the target device in the previous period includes the cumulative value of the first abnormal indicator and the cumulative value of the second abnormal indicator; The step of determining the cumulative abnormal indicator value of the target device in the current period based on the pre-acquired cumulative abnormal indicator value of the previous period before the current period and the abnormal indicator value of the current period includes: Based on the cumulative value of the first abnormal indicator, the value of the first abnormal indicator, and a preset tolerance coefficient, a cumulative value of the first candidate abnormal indicator is determined, wherein the preset tolerance coefficient is used to suppress transient noise fluctuations. The maximum value between the first candidate abnormal indicator accumulation and the second preset value is determined as the third abnormal indicator accumulation value of the target device in the current period; The cumulative value of the second candidate abnormal indicator is determined based on the cumulative value of the second abnormal indicator, the value of the second abnormal indicator, and the preset tolerance coefficient. The maximum value between the second candidate abnormal indicator accumulation and the second preset value is determined as the fourth abnormal indicator accumulation value of the target device in the current period; The cumulative value of the third abnormal indicator and the cumulative value of the fourth abnormal indicator constitute the cumulative value of the abnormal indicators of the target device in the current period.

5. The power consumption anomaly detection method according to any one of claims 2 to 4, characterized in that, The step of determining whether the target device is in an abnormal state based on the actual power consumption value and preset second power consumption distribution information when the cumulative value of the abnormal indicator of the target device in the current period is greater than a preset alarm threshold includes: If the cumulative value of the abnormal indicator in the current period is determined to be greater than the preset alarm threshold, it is determined whether the actual power consumption value matches the second power consumption distribution information; If the actual power consumption value does not match the second power consumption distribution information, the target device is determined to be in an abnormal state. Alternatively, if the actual power consumption value matches the second power consumption distribution information, it can be determined that the target device is not in an abnormal state.

6. The power consumption anomaly detection method according to claim 5, characterized in that, If the actual power consumption value matches the second power consumption distribution information, the method further includes: Mark the current period as a global migration event; Statistics are marked as at least one candidate period of the global migration event; The average power consumption and standard deviation of the global migration event are determined based on the actual power consumption values ​​of the target device in at least one of the candidate periods. The second power distribution information is updated based on the average power consumption value and standard deviation of the global migration event.

7. The power consumption anomaly detection method according to claim 6, characterized in that, The method further includes: Candidate alarm thresholds are determined based on a preset baseline alarm threshold, the power consumption baseline difference, the standard deviation, a preset reference power consumption difference, and a reference standard deviation. The target alarm threshold is determined based on the candidate alarm threshold, the preset maximum alarm threshold, and the minimum alarm threshold. Update the preset alarm threshold to the target alarm threshold.

8. The power consumption anomaly detection method according to claim 7, characterized in that, The target alarm threshold is expressed by the following expression: in, This indicates the target alarm threshold. Indicates the baseline alarm threshold. The standard deviation of the global migration event is represented by the standard deviation of the global migration event. This represents the power consumption reference difference. This represents the reference standard deviation. The reference power consumption difference, This represents the minimum alarm threshold. This represents the maximum alarm threshold. This is a preset minimum value used to prevent division by zero.

9. The power consumption anomaly detection method according to any one of claims 1 to 4, characterized in that, The method further includes: When it is determined that the target device is in the abnormal state, the performance index data is input into the pre-constructed second quantile regression model to obtain the third power consumption distribution information output by the second quantile regression model. The update frequency of the second quantile regression model is lower than that of the first quantile regression model. The second power consumption distribution information includes a second maximum power consumption value and a second minimum power consumption value. If it is determined that the actual power consumption value is greater than the second maximum power consumption value, or the actual power consumption value is less than the second minimum power consumption value, a second alarm operation is performed, wherein the alarm level of the second alarm operation is higher than the alarm level of the first alarm operation. Alternatively, if it is determined that the actual power consumption value is less than or equal to the second maximum power consumption value, and the actual power consumption value is greater than or equal to the second minimum power consumption value, a third alarm operation is performed, wherein the alarm level of the third alarm operation is lower than the alarm level of the first alarm operation.

10. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the power consumption anomaly detection method as described in any one of claims 1 to 9 when executing the computer program.

Images