A real-time monitoring method of a data-driven power energy system

Abstract

This invention relates to the field of power energy monitoring technology, and more specifically, to a data-driven real-time monitoring method for power energy systems. The method includes: assessing the loss risk caused by equipment aging and failure based on equipment output loss due to aging, equipment failure risk, and maintenance costs; acquiring trend monitoring values ​​for each time window to determine the trend correction factor for the most recent time window; obtaining the equipment's aging index by multiplying the trend monitoring value and the trend correction factor for the most recent time window; determining the state weight of the most recent time window in response to the equipment's aging index exceeding an aging threshold; and adjusting the power energy system's allocation of monitoring resources for the equipment based on the equipment's loss risk function and the state weight of the most recent time window, thereby effectively improving the efficiency of power energy system monitoring resource allocation.

Description

Technical Field

[0001] This invention relates to the field of power energy monitoring technology. More specifically, this invention relates to a data-driven real-time monitoring method for power energy systems. Background Technology

[0002] A data-driven power energy system is a new type of system that relies on modern information technology to achieve intelligent decision-making, optimized operation, and efficient management of the power system by controlling the entire process of power production, transmission, and distribution through data management and combining artificial intelligence, big data, and the Internet of Things.

[0003] One of the core objectives of power energy systems is to achieve risk prediction through dynamic sensing, and equipment aging is a hidden risk source threatening system security. Power equipment is highly interconnected, and the aging of a single piece of equipment may trigger a chain reaction, causing operating parameters to gradually deviate from the normal range. Data-driven systems can locate aging equipment and reduce energy loss through end-to-end data correlation analysis.

[0004] The current periodic maintenance model may lead to resource waste or missed risks. Data-driven systems, by continuously monitoring equipment aging characteristic parameters and combining them with machine learning models such as ARIMA to predict the remaining lifespan of equipment, can achieve systematic maintenance. However, ARIMA models struggle to accurately extract aging signals from numerous complex factors, and traditional methods for assessing equipment failure losses are mostly post-hoc statistical analyses. This makes it difficult for the system to predict the potential consequences of aging in advance, ultimately resulting in low accuracy of real-time monitoring results for data-driven power energy systems and wasting monitoring resources.

[0005] Therefore, how to accurately obtain aging signals to assess the aging degree of equipment and thus accurately allocate monitoring resources is a problem that needs to be solved. Summary of the Invention

[0006] To address the technical problem of accurately acquiring aging signals to assess the aging degree of equipment and thus accurately allocate monitoring resources, this invention proposes a data-driven real-time monitoring method for power energy systems. This method includes the following steps: The system acquires historical data on equipment temperature, aging-related equipment output loss, equipment failure risk, and maintenance costs. Based on these data, a loss risk function is constructed to assess the risk of aging-related failures. The historical period is divided into multiple time windows. Trend monitoring values ​​for each time window are obtained based on the change in entropy values ​​of the temperature histogram within that window. A trend correction factor for the most recent time window is determined by comparing trend monitoring values ​​between adjacent time windows. The aging index of the equipment is obtained by multiplying the trend monitoring value of the most recent time window by the trend correction factor. If the aging index exceeds an aging threshold, the state weight of the most recent time window is determined based on the ratio of its trend monitoring value to the standard entropy value. Finally, the power system's allocation of monitoring resources for the equipment is adjusted based on the loss risk function and the state weight of the most recent time window.

[0007] This invention dynamically adjusts the system's monitoring resources by analyzing the degree of equipment aging and the potential losses caused by aging, effectively improving the utilization efficiency of monitoring resources. When analyzing potential losses caused by aging, this invention constructs a loss risk function based on equipment output loss, equipment failure risk, and maintenance costs, accurately obtaining the equipment's loss risk value and thus effectively increasing the weight of system monitoring resources for equipment loss risk. Furthermore, this invention measures the probability of equipment aging by the degree of temperature entropy change between adjacent time windows, accurately capturing abnormal temperature entropy changes. Based on this, this invention further distinguishes between aging and non-aging factors by the characteristic of increasing temperature entropy caused by aging, improving the accuracy of state weights and thus effectively improving the efficiency of power system monitoring resource adjustment.

[0008] According to the present invention, a data-driven real-time monitoring method for a power energy system includes the following steps: obtaining the unit output, unit output profit, failure duration, and aging-related loss coefficient between the equipment and other equipment during the historical period; and calculating the equipment output loss caused by aging during the historical period. ; This represents the loss of equipment output value due to aging during the historical period of the equipment. , , , These are the unit output, unit output value and profit, downtime, and aging-related loss coefficient of the equipment during the historical period.

[0009] This invention takes into account that the loss of output value caused by equipment aging is not limited to its own downtime, but may also affect related equipment. Therefore, this invention covers the direct loss part and the related loss part through multi-dimensional parameters, highlighting the cumulative nature of aging loss, so as to accurately obtain the equipment output value loss value caused by aging in the historical period of the equipment.

[0010] According to the present invention, a data-driven real-time monitoring method for a power energy system and a method for obtaining equipment failure risk values ​​within a historical time period are provided, comprising: obtaining the real-time load rate, rated load rate, historical failure rate within a historical time period, and industry average historical failure rate of each device in the power energy system, and calculating the equipment failure risk value within the historical time period. : ; , These are the device's real-time load rate and historical failure rate over a historical period, respectively. , These are the rated load rate of the equipment during a historical period and the industry average historical failure rate, respectively. It is an exponential function with base e.

[0011] This invention takes into account that the failure risk of power equipment is affected by both the equipment's operating load and its historical failure performance. Therefore, by measuring the equipment's health status by obtaining the equipment's real-time load rate deviation and historical failure rate deviation, the failure risk value of the equipment within a historical period can be accurately obtained.

[0012] According to the present invention, a data-driven real-time monitoring method for a power energy system, and a method for obtaining the maintenance cost of equipment over a historical period, include: obtaining the actual maintenance cost of the equipment at its most recent time, the equipment maintenance complexity score, the maintenance duration, and the average maintenance duration over a historical period; and calculating the maintenance cost of the equipment over the historical period. : ; , , These are the actual cost of the equipment's most recent repair, the equipment repair complexity score, and the repair time. This represents the average repair time for the equipment over a historical period. This is the time sensitivity coefficient.

[0013] According to the present invention, a data-driven real-time monitoring method for a power energy system includes constructing a loss risk function for the equipment based on the equipment output value loss due to aging, the equipment failure risk value, and the maintenance cost to assess the loss risk caused by the aging failure of the equipment, comprising: ; For the equipment loss risk function, This represents the equipment output value loss caused by aging during the historical period of the equipment. This refers to the maintenance costs of the equipment over a historical period. This represents the equipment failure risk value over a historical period. It is a linear normalization function.

[0014] According to the present invention, a real-time monitoring method for a data-driven power energy system is provided. The step of obtaining the trend monitoring value for each time window based on the change in entropy value of the temperature histogram within each time window includes: recording the absolute value of the difference between the entropy value of the first time window and the standard entropy value during normal equipment operation as the trend monitoring value of the first time window; then substituting this value into the trend monitoring model of the time window to obtain the trend monitoring value of the second time window, and so on, to obtain the trend monitoring value for each time window; wherein, the trend monitoring model of the time window is: , For the first Trend monitoring values ​​for each time window, Weighted by historical trends For the first Trend monitoring values ​​for each time window, For the first The entropy value of each time window. The standard entropy value, It is the absolute value symbol.

[0015] This invention takes into account that abnormal changes in equipment temperature are usually a gradual process. Therefore, by comprehensively evaluating the current state of the time window by obtaining the current state item and the historical trend accumulation item, it can accurately capture the short-term abnormal fluctuations and long-term slow deterioration trends of the equipment, thereby accurately obtaining the trend changes of the equipment time window.

[0016] According to a data-driven real-time monitoring method for a power energy system provided by the present invention, the step of determining the trend correction factor of the device's most recent time window based on the comparison results of trend monitoring values ​​between adjacent time windows includes: the first time window and the second time window are a first group of adjacent time windows; if the trend monitoring value of the second time window is less than the trend monitoring value of the first time window, then the group of adjacent windows is marked as 1; the comparison results of the trend monitoring values ​​of the remaining groups of adjacent time windows are obtained, and finally the number of all adjacent time window groups marked as 1 for the device is obtained; the ratio of the number of adjacent time window groups marked as 1 to the total number of adjacent time window groups is recorded as the trend correction factor of the device's most recent time window.

[0017] This invention takes into account that changes caused by some non-aging anomalies may also lead to large temperature fluctuations within a time window, while anomalies caused by aging have a gradual accumulation characteristic. Therefore, this invention analyzes the differences in trend changes between adjacent time windows to determine the probability that they conform to aging anomalies, and accurately corrects the trend monitoring values ​​within the time window.

[0018] According to a data-driven real-time monitoring method for a power energy system provided by the present invention, the step of determining the state weight of the most recent time window based on the ratio of the trend monitoring value of the equipment in the most recent time window to the standard entropy value includes: ; The state weights are for the most recent time window. The initial state weights for this most recent time window. For the first Trend monitoring values ​​for each time window, The standard entropy value, It is the absolute value symbol.

[0019] According to the present invention, a data-driven real-time monitoring method for a power energy system includes adjusting the allocation of monitoring resources for the equipment based on the equipment's loss risk function and the state weight of the most recent time window, comprising: through... The function integrates the loss risk function of the device and the product of the state weight of the most recent time window to obtain the attention level of the device in the most recent time window; monitoring resources are allocated to the device according to the attention level of the device in the most recent time window, and the attention level of the device in the most recent time window is positively correlated with the monitoring resources.

[0020] According to the present invention, a data-driven real-time monitoring method for a power energy system, wherein allocating monitoring resources to devices based on the attention level of the devices in their most recent time window includes: using the attention level of the devices in their most recent time window as input to an exponential function with base e to obtain the attention index of the devices in their most recent time window; obtaining the attention index and value of other devices of the same type as the devices in the power energy system for their most recent time windows, and using the ratio of the attention index of the devices in their most recent time window to the sum of the attention index and value as the attention weight of the devices; and using the attention weight of each device in the Transformer attention mechanism to realize the allocation of monitoring resources.

[0021] The present invention has the following beneficial effects: Based on the above technical solutions, this invention provides a data-driven real-time monitoring method for power energy systems. By analyzing the degree of equipment aging and the potential losses caused by aging, the method dynamically adjusts the system's monitoring resources, effectively improving the utilization efficiency of monitoring resources. When analyzing potential losses caused by aging, this invention constructs a loss risk function based on equipment output loss, equipment failure risk, and maintenance costs, accurately obtaining the equipment's loss risk value, thereby effectively increasing the weight of system monitoring resources on equipment loss risk. Furthermore, this invention measures the probability of equipment aging by the degree of temperature entropy change between adjacent time windows, accurately capturing abnormal temperature entropy changes. Based on this, this invention further distinguishes between aging and non-aging factors by the characteristic of increasing temperature entropy caused by aging, improving the accuracy of state weights and thus effectively improving the efficiency of power system monitoring resource adjustment. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the steps of a data-driven real-time monitoring method for a power energy system provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0024] Please see Figure 1 , Figure 1 This is a flowchart illustrating the steps of a data-driven real-time monitoring method for power energy systems provided in this invention. The method constructs a loss risk function for equipment to reflect the short-term immediate impact of equipment failures and obtains the state weights of the equipment to measure the long-term trend of equipment performance degradation caused by aging. Combining these two methods enables a comprehensive assessment of the short-term urgency and long-term trend of aging, thereby accurately screening the potential risks of highly aging and heavily loaded equipment. Monitoring resources are then allocated to the equipment based on the degree of potential risk, thereby improving the utilization rate of monitoring resources. The method specifically includes the following steps: S1: Obtain relevant parameters generated by aging during the historical period of the equipment.

[0025] Among these parameters, at least the loss of equipment output value, equipment failure risk, and maintenance cost should be included.

[0026] It should be noted that equipment aging can lead to aging failures, and the inability of the equipment to operate normally will cause related losses. Moreover, the losses will increase as the aging process intensifies.

[0027] Based on this, embodiments of the present invention can construct a loss risk function by quantifying the losses caused by equipment aging to assess the loss risk generated by the equipment aging failure, so as to analyze the severity of the equipment aging and allocate corresponding monitoring resources according to the severity of the equipment aging, thereby effectively improving the utilization rate of monitoring resources.

[0028] It should be further clarified that equipment output loss refers to the reduction in production output caused directly or indirectly by the decline in equipment performance or downtime due to aging. The more severe the aging, the greater the output loss. Failure risk refers to the probability of equipment failure; the more severe the aging, the higher the failure risk. Maintenance cost refers to the maintenance expenses (including labor, spare parts, etc.) incurred to maintain the normal operation of equipment; the more severe the aging, the higher the maintenance cost.

[0029] Therefore, embodiments of the present invention can obtain the equipment output value loss, equipment failure risk and maintenance cost caused by aging during the historical period of the equipment, so as to assess the risk of the equipment experiencing aging failure in the short term.

[0030] It is understandable that the historical period of a device is the period from when the device started operating until the current moment.

[0031] For example, in an embodiment of the present invention, obtaining the equipment output value loss caused by aging during the historical period of the equipment includes: obtaining the unit output, output value profit per unit output, failure duration and aging-related loss coefficient between the equipment and other equipment during the historical period; and calculating the equipment output value loss caused by aging during the historical period of the equipment.

[0032] Specifically, the unit output of equipment can be viewed through the power energy system over historical periods. The unit time for unit output can be set to 1 hour. For example, if the equipment is a wind turbine, the power generation of the wind turbine in one hour can be obtained. Since the status of the equipment may fluctuate, the average output over multiple unit times within the historical period can be used as the unit output.

[0033] Profit per unit of output can be derived from a company's cost-profit analysis report and corresponds to the unit output.

[0034] The unit of fault duration for the equipment within a historical period is the same as the unit of time. If the unit of time is one hour, then the unit of fault duration is hours. Each fault is counted from the start of the fault count until it can be used normally again. Finally, the sum of the fault durations corresponding to all fault counts within the historical period is taken as the total fault duration for the historical period.

[0035] The aging-related loss coefficient between a device and other devices over a historical period represents the output value loss rate of the remaining devices caused by the failure of that device. Specifically, it is obtained by: acquiring a list of directly related devices affected by the device failure; analyzing the load transfer ratio and derating ratio of each related device based on the power system topology; and calculating the total output capacity reduction ratio of the related devices due to the device failure, which is the aging-related loss coefficient. For example, if one generator in a generator set suddenly trips and stops working, the remaining generators will automatically take over the entire load. This increased load, exceeding the rated load, may cause overload or frequency fluctuations. In this case, to protect the safety of the remaining generators, the power generation system will proactively reduce the generator's output rate. If the output rate of each of the ten generators decreases by 10%, the total output value loss rate is 100%.

[0036] It is understood that the embodiments of the present invention analyze the failure losses caused by aging, therefore the parameters obtained above are all parameters caused by aging failures.

[0037] For example, to calculate the equipment output loss caused by aging over a historical period, please refer to the following formula: ; This represents the loss of equipment output value due to aging during the historical period of the equipment. This represents the unit output of the equipment within a historical period. This represents the profit per unit output of the equipment over a historical period. This represents the duration of the device's malfunctions over a historical period. This is the aging-related loss coefficient between this device and other devices during a historical period.

[0038] In this calculation method, This indicates the direct loss in terms of output value caused by the equipment failure per unit of time. This represents the combined value of the equipment failure and the output loss of other equipment caused by the equipment failure within a unit of time. This represents the total loss value of the entire production line caused by the equipment failure until repair is completed within a historical period. The higher the unit output, unit output profit, failure duration, and aging-related loss coefficient between the equipment and other equipment within the historical period, the higher the equipment output loss value caused by the equipment, and the more attention should be paid to the losses caused by the aging of the equipment.

[0039] For example, in an embodiment of the present invention, the method for obtaining the equipment failure risk value within a historical period includes: obtaining the real-time load rate, rated load rate, historical failure rate within a historical period, and industry average historical failure rate of each device in the power energy system, and calculating the equipment failure risk value within a historical period.

[0040] Specifically, the real-time load rate of each device can be collected through sensors, control systems and other devices; the historical failure rate of each device caused by aging and the industry average historical failure rate can be obtained through maintenance logs; and the rated load rate of the device can be obtained through the parameters on the device nameplate.

[0041] When obtaining the industry average historical failure rate, one can obtain the average failure rate of equipment of the same type and specification as the equipment to be evaluated within the same operating period from equipment operation statistical reports published by the power industry association, equipment failure statistics of similar power companies, or reliability data provided by equipment manufacturers. Alternatively, one can collect the historical failure rates of all similar equipment in the power energy system, remove the highest and lowest values, and take the arithmetic mean as the industry average historical failure rate.

[0042] For example, in this embodiment of the invention, when calculating the equipment failure risk value over a historical period, the load condition of the equipment can be reflected by comparing the real-time load rate with the rated load rate, and the failure condition of the equipment can be reflected by comparing the historical failure rate with the industry average failure rate. Combining the two can accurately obtain the equipment failure risk value over a historical period, as shown in the following formula: ; This represents the equipment failure risk value within a historical time period. This represents the real-time load rate of the device. This represents the historical failure rate of the equipment over a specific historical period. This is the rated load rate of the equipment. This represents the industry average historical failure rate for this equipment over a given period. It is an exponential function with base e, where e is the natural constant.

[0043] It is the normalized value of the historical failure rate of the equipment exceeding the rated load rate within a historical period. These are the normalized values ​​of the historical failure rate of the equipment over a historical period that exceed the industry average historical failure rate. The larger these two values ​​are, the greater the degree to which the equipment exceeds the rated value, and the greater the risk of equipment failure.

[0044] For example, in an embodiment of the present invention, the method for obtaining the maintenance cost of equipment within a historical period includes: obtaining the actual maintenance cost of the equipment in the most recent time, the equipment maintenance complexity score, the maintenance time, and the average maintenance time within the historical period; and calculating the maintenance cost of the equipment within the historical period.

[0045] Specifically, in power energy systems, when critical equipment fails, losses typically increase non-linearly with the duration of the failure. Therefore, the indirect losses increase non-linearly with increasing repair time, and the rate of increase accelerates with repair time. Furthermore, the repair complexity varies among different pieces of equipment, encompassing the inherent complexity of the equipment itself, the availability of spare parts, the technical difficulty of repair, and the repair risks. Based on these parameters, a repair cost function can be derived. The actual repair cost of the most recent repair, the equipment repair complexity score, the repair duration, and the average repair duration over historical periods are obtained through the corresponding work order maintenance records.

[0046] When obtaining equipment maintenance complexity scores, a scoring model can be constructed based on three indicators: the difficulty of obtaining faulty spare parts, the technical difficulty, and the maintenance environment. ; Rate the complexity of the equipment's most recent maintenance. This represents the value of the i-th indicator from the most recent maintenance of the equipment. This represents the weight of the i-th indicator in the most recent maintenance of the equipment.

[0047] Specifically, when setting the difficulty index for obtaining faulty spare parts, if the factory already has them, the index can be set to 0; otherwise, it can be set to 1. When setting the technical difficulty index, if ordinary workers can complete the task, the index can be set to 0; otherwise, it can be set to 1. When setting the maintenance environment index, it can be set to 0 for ordinary environments and 1 for environments such as high altitude, low temperature, high temperature, and corrosive environments. The weights of the three indicators can be set to 0.4, 0.4, and 0.2 respectively, and can be set according to actual needs. This embodiment of the invention does not impose too many restrictions here.

[0048] For example, in an embodiment of the present invention, the maintenance cost of the device over a historical period can be calculated using the following formula: ; This refers to the maintenance costs during the equipment's historical period. This represents the actual cost of the equipment's most recent repair. This is the duration of the equipment's most recent maintenance. This represents the average repair time for the equipment over a historical period. The time sensitivity coefficient, Rate the complexity of the equipment's most recent maintenance.

[0049] in, The time sensitivity coefficient can be set to 1.05; it indicates that the loss caused by maintenance time increases non-linearly. For critical equipment, such as main transformers and generator sets, where downtime losses increase sharply with time, the time sensitivity coefficient can be taken as 1.2 to 1.5; for auxiliary equipment where downtime losses increase relatively slowly, the time sensitivity coefficient can be taken as 1.0 to 1.2. Optionally, in this embodiment of the invention, it can be set to 1.05; the specific value can be set according to actual needs.

[0050] In this method of calculating maintenance costs, the maintenance cost of the equipment is measured by the direct cost of maintenance, maintenance complexity, and maintenance time. This allows for better allocation of monitoring resources to the equipment based on the direct cost, maintenance complexity, and maintenance time.

[0051] After obtaining the equipment output value loss, equipment failure risk and maintenance cost caused by aging during the historical period of the equipment based on the above steps, the equipment loss risk function can be constructed based on this, and then the following steps can be performed.

[0052] S2: Construct the loss risk function for the equipment.

[0053] For example, in an embodiment of the present invention, a loss risk function for the equipment is constructed based on the equipment output value loss due to aging, the equipment failure risk value, and the maintenance cost to assess the loss risk caused by the aging failure of the equipment, including: ; For the equipment loss risk function, This represents the equipment output value loss caused by aging during the historical period of the equipment. This refers to the maintenance costs of the equipment over a historical period. This represents the equipment failure risk value over a historical period. It is a linear normalization function.

[0054] During linear normalization, all devices at the current time can be obtained. The maximum and minimum values ​​in the range are used to obtain the loss risk function of the equipment through max-min normalization.

[0055] In the constructed equipment loss risk function, by combining the equipment output value loss value caused by aging, the equipment failure risk value, and the maintenance cost, the loss risk caused by equipment aging can be accurately assessed, laying the foundation for subsequent resource allocation.

[0056] S3: Determine the aging indicators of the equipment.

[0057] For example, the temperature of the device during a historical period is obtained, the historical period is divided into multiple time windows, and the trend monitoring value of each time window is obtained based on the change in the entropy value of the temperature histogram within each time window, so as to obtain the aging index of the device.

[0058] It should be noted that equipment aging is often closely related to internal thermal stress. Temperature directly reflects the aging state of the internal insulation materials. Prolonged or sustained high temperatures accelerate the physicochemical changes of the materials, leading to a decline in insulation performance.

[0059] Therefore, when analyzing the long-term trend of equipment performance degradation caused by equipment aging, the embodiments of the present invention can use the temperature entropy value to measure the disorder of data distribution to reflect the abnormal trend of the equipment. The greater the change in entropy value, the more serious the abnormal trend of the equipment.

[0060] For example, in order to perform targeted analysis on the temperature changes of the equipment at different times, embodiments of the present invention can divide the historical period of the equipment into multiple time windows and analyze the changes in entropy values ​​within each time window.

[0061] The length of the time window can be set to 7 days; the specific length of the time window can be set according to actual needs, and this embodiment of the invention does not impose too many restrictions here.

[0062] Specifically, when obtaining the entropy value of temperature change within a time window, the temperature value within the time window can be obtained, and the temperature value can be divided into multiple temperature intervals with an interval length of 10 degrees Celsius. The number of temperatures falling into each temperature interval can be counted to obtain a temperature histogram, and the entropy value of the temperature histogram can be determined. The entropy value of the temperature histogram can be calculated using the Shannon entropy formula, which will not be elaborated here.

[0063] For example, in an embodiment of the present invention, the trend monitoring value of each time window can be obtained based on the change in the entropy value of the temperature histogram within each time window. Specifically, this includes: recording the absolute value of the difference between the entropy value of the first time window and the standard entropy value when the equipment is operating normally as the trend monitoring value of the first time window, then substituting it into the trend monitoring model of the time window to obtain the trend monitoring value of the second time window, and substituting it into the trend monitoring model of the time window to obtain the trend monitoring value of the third time window, and so on, to obtain the trend monitoring value of each time window.

[0064] The standard entropy value during normal operation of the equipment can be obtained by taking the average of the entropy values ​​of the first four time windows after the new equipment has been running stably.

[0065] For example, the trend monitoring model for time windows is as follows: ; For the first Trend monitoring values ​​for each time window, Weighted by historical trends For the first Trend monitoring values ​​for each time window, For the first The entropy value of each time window. The standard entropy value, It is the absolute value symbol.

[0066] For equipment primarily involving mechanical wear, such as motors, aging is a slow process. Therefore, it's crucial to focus on long-term trends, with a historical trend weight of 0.8. A higher weight amplifies the impact of past entropy changes. For equipment like cables or relay protection devices that experience sudden changes after aging, more attention needs to be paid to instantaneous anomalies, and a historical trend weight of 0.5 is appropriate. The specific historical trend weight can be set according to actual needs, and this embodiment of the invention does not impose excessive limitations.

[0067] In the trend monitoring model construction method within this time window, the higher the degree of equipment aging, the greater the difference between the entropy value within the time window and the entropy value during normal operation. The larger, It accumulates the weighted sum of the entropy value and the standard entropy value difference of each time window within the historical period. The further back the time window, the smaller the weight. The higher the degree of equipment aging, the higher the accumulated entropy value change, and the larger the corresponding trend monitoring value.

[0068] It should be further noted that the entropy changes accumulated during equipment aging can be easily confused with non-aging anomalies. These anomalies may also manifest as temperature entropy anomalies within each time window, resulting in significant changes in entropy values.

[0069] Based on this, embodiments of the present invention can correct the trend monitoring value of the time window according to the entropy change characteristics of aging, and determine the trend correction factor of the device's most recent time window by comparing the trend monitoring values ​​between adjacent time windows.

[0070] For example, in this embodiment of the invention, determining the trend correction factor of the device's most recent time window based on the comparison results of trend monitoring values ​​between adjacent time windows includes: the first time window and the second time window are the first group of adjacent time windows; if the trend monitoring value of the second time window is less than the trend monitoring value of the first time window, then the group of adjacent windows is marked as 1; the comparison results of the trend monitoring values ​​of the remaining groups of adjacent time windows are obtained, and finally the number of all adjacent time window groups marked as 1 for the device is obtained; the ratio of the number of adjacent time window groups marked as 1 to the total number of adjacent time window groups is recorded as the trend correction factor of the device's most recent time window.

[0071] Some changes not caused by aging anomalies may also lead to large temperature fluctuations within a time window, while aging-related anomalies exhibit a gradual accumulation characteristic, manifested as a gradual increase in entropy differences over time, i.e., the trend monitoring value shows an increasing trend. Therefore, this embodiment of the invention analyzes the differences in trend changes between adjacent time windows, statistically analyzes the proportion of window groups with increasing trend monitoring values, obtains the probability that they conform to aging anomalies, and accurately corrects the trend monitoring values ​​within the time window. The trend correction factor ranges from [0,1], with a larger value indicating that the change in equipment temperature entropy value more conforms to the gradual accumulation characteristic of aging.

[0072] For example, the aging index of the device is obtained by multiplying the trend monitoring value of the most recent time window and the trend correction factor.

[0073] The aging index of the equipment can be obtained by following the above steps. If the aging index of the equipment is large, it means that the possibility of failure due to aging is increased. Therefore, it is necessary to determine the state weight of the most recent time window based on the ratio of the trend monitoring value of the most recent time window to the standard entropy value to reflect the aging status of the equipment.

[0074] S4: Determine the state weight of the most recent time window.

[0075] For example, in response to a device's aging index exceeding an aging threshold, the state weight of the most recent time window is determined based on the ratio of the trend monitoring value of the device's most recent time window to the standard entropy value.

[0076] The aging threshold can be set to 0.4; the specific aging threshold can be set according to actual needs.

[0077] For example, in an embodiment of the present invention, determining the state weight of the most recent time window based on the ratio of the trend monitoring value of the device in the most recent time window to the standard entropy value includes: ; The state weights are for the most recent time window. The initial state weights for this most recent time window. For the first Trend monitoring values ​​for each time window, The standard entropy value, It is the absolute value symbol.

[0078] The initial state weight can be set to 1; the specific initial state weight can be set according to actual needs.

[0079] In the method of calculating the state weight of the device in the most recent time window The larger the value, the more severe the aging of the equipment. Adjusting the monitoring resources of the equipment according to this value can enable the equipment with a more severe aging condition to obtain more attention in the monitoring resource attention mechanism.

[0080] Understandably, if the aging index of the equipment is not greater than the aging threshold, it means that the current entropy value fluctuates little, the equipment status is relatively stable, and there is no obvious aging trend. Therefore, there is no need to increase its monitoring priority or adjust resources.

[0081] For example, in an embodiment of the present invention, in response to the aging index of the device not being greater than the aging threshold, the original monitoring resource allocation system of the device is maintained.

[0082] After obtaining the equipment loss risk function and the state weight of the most recent time window based on the above steps, the potential losses caused by equipment aging risk can be assessed by combining the two. Based on this, monitoring resources can be adjusted to improve resource utilization, i.e., the following steps are performed.

[0083] S5: Adjust the allocation of monitoring resources for equipment in the power energy system.

[0084] Among these, the power energy system can adjust the allocation of monitoring resources for the equipment based on the equipment's loss risk function and the state weight of the most recent time window.

[0085] Specifically, the monitoring resources in the embodiments of the invention include, but are not limited to: sensor data acquisition frequency, data transmission bandwidth, computing resource allocation of edge computing nodes, processing priority of cloud analysis servers, and inspection frequency of maintenance personnel. Among these, the sensor data acquisition frequency is the core monitoring resource, and its adjustment directly determines the accuracy of the system's perception of the equipment status.

[0086] It should be noted that, in this embodiment of the invention, the loss risk function of the device obtained in the above steps and the state weight of the most recent time window can be combined into an adaptive attention weight matrix, and the attention mechanism of the Transformer architecture can be used to adjust the frequency of device monitoring and resource allocation.

[0087] For example, in an embodiment of the present invention, adjusting the allocation of monitoring resources for the device by the power energy system based on the device's loss risk function and the state weight of the most recent time window includes: through... The function integrates the loss risk function of the device and the product of the state weight of the most recent time window to obtain the attention level of the device in the most recent time window; monitoring resources are allocated to the device according to the attention level of the device in the most recent time window, and the attention level of the device in the most recent time window is positively correlated with the monitoring resources.

[0088] For example, in this embodiment of the invention, allocating monitoring resources to a device based on the attention level of the device's most recent time window includes: using the attention level of the device's most recent time window as input to an exponential function with base e to obtain the attention index of the device's most recent time window; obtaining the attention index and value of other devices of the same type as the device in the power energy system for their most recent time windows, and using the ratio of the device's most recent time window attention index to the sum of attention indexes as the attention weight of the device; and using the attention weights of each device in the Transformer attention mechanism to allocate monitoring resources.

[0089] Specifically, when using attention weights to allocate monitoring resources in the Transformer attention mechanism, the attention weights of each device can be used as scaling factors after the dot product of the query vector and the key vector. This allows for weighted aggregation of device state features within the multi-head attention mechanism. Devices with higher attention weights have a larger proportion of their state features in the aggregation result, and the system prioritizes monitoring that device.

[0090] Optionally, the monitoring resource allocation rules can be set as follows: when the attention weight is ≥0.3, the data collection frequency of the device is increased to the highest level, such as sampling once every 5 minutes, and a dedicated edge computing node is allocated for real-time analysis; when 0.15≤attention weight<0.3, the data collection frequency is increased to a medium level, such as sampling once every 15 minutes, and edge computing resources are shared for periodic analysis; when the attention weight is <0.15, the original data collection frequency is maintained, such as sampling once per hour, and a cloud-based batch analysis mode is adopted.

[0091] It is understandable that the attention weight of a device is the proportion of that device in the total monitoring resources. By adjusting the monitoring resources of devices through attention weight, more monitoring resources can be allocated to devices with higher aging levels in the attention weight matrix.

[0092] For example, when adjusting monitoring resources, resources can be allocated in descending order of the attention weight of the devices, with priority given to devices with higher attention weights.

[0093] For example, the monitoring resources can be adjusted by adjusting the attention weight of the device, so that the original monitoring frequency of the device is changed from once an hour to once every ten minutes.

[0094] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A data-driven real-time monitoring method for power energy systems, characterized in that, include: The equipment's temperature over a historical period is obtained, along with the equipment's output value loss due to aging, equipment failure risk, and maintenance costs. Based on the equipment output value loss due to aging, equipment failure risk, and maintenance costs, a loss risk function for the equipment is constructed to assess the loss risk caused by the equipment's aging failure. The historical period is divided into multiple time windows, and the trend monitoring value of each time window is obtained based on the change in the entropy value of the temperature histogram within each time window. The trend correction factor for the device's most recent time window is determined based on the comparison results of trend monitoring values ​​between adjacent time windows. The aging index of the equipment is obtained by multiplying the trend monitoring value of the most recent time window and the trend correction factor. In response to the aging index of the equipment exceeding the aging threshold, the state weight of the most recent time window is determined based on the ratio of the trend monitoring value of the most recent time window to the standard entropy value. The power system adjusts the allocation of monitoring resources for the equipment based on the equipment's loss risk function and the state weight of the most recent time window.

2. The real-time monitoring method for a data-driven power energy system according to claim 1, characterized in that, Methods for obtaining the equipment output value loss caused by aging during the historical period of equipment include: Obtain the unit output, unit output profit, downtime, and aging-related loss coefficient of the equipment in historical periods; Calculate the equipment output value loss caused by aging during the historical period of the equipment: ； This represents the loss of equipment output value due to aging during the historical period of the equipment. , , , These are the unit output, unit output value and profit, downtime, and aging-related loss coefficient of the equipment during the historical period.

3. The real-time monitoring method for a data-driven power energy system according to claim 1, characterized in that, Methods for obtaining equipment failure risk values ​​over historical periods include: Obtain the real-time load rate, rated load rate, historical failure rate within a historical period, and industry average historical failure rate of each device in the power energy system, and calculate the equipment failure risk value within the historical period. : ； , These are the device's real-time load rate and historical failure rate over a historical period, respectively. , These are the rated load rate of the equipment during a historical period and the industry average historical failure rate, respectively. It is an exponential function with base e.

4. The real-time monitoring method for a data-driven power energy system according to claim 1, characterized in that, Methods for obtaining equipment maintenance costs over historical periods include: Obtain the equipment's most recent actual maintenance cost, equipment maintenance complexity score, maintenance duration, and average maintenance duration over historical periods; calculate the equipment's maintenance cost over historical periods. : ； , , These are the actual cost of the equipment's most recent repair, the equipment repair complexity score, and the repair time. This represents the average repair time for the equipment over a historical period. This is the time sensitivity coefficient.

5. The real-time monitoring method for a data-driven power energy system according to claim 1, characterized in that, The method of constructing a loss risk function for the equipment based on the equipment output value loss due to aging, the equipment failure risk value, and the maintenance cost to assess the loss risk caused by the aging failure of the equipment includes: ； For the equipment loss risk function, This represents the equipment output value loss caused by aging during the historical period of the equipment. This refers to the maintenance costs of the equipment over a historical period. This represents the equipment failure risk value over a historical period. It is a linear normalization function.

6. The real-time monitoring method for a data-driven power energy system according to claim 1, characterized in that, The process of obtaining trend monitoring values ​​for each time window based on the change in temperature histogram entropy values ​​within each time window includes: The absolute value of the difference between the entropy value of the first time window and the standard entropy value when the equipment is running normally is recorded as the trend monitoring value of the first time window. This value is then substituted into the trend monitoring model for the time window to obtain the trend monitoring value for the second time window, and so on, to obtain the trend monitoring value for each time window. The trend monitoring model for the time window is as follows: , For the first Trend monitoring values ​​for each time window, Weighted by historical trends For the first Trend monitoring values ​​for each time window, For the first The entropy value of each time window. The standard entropy value, It is the absolute value symbol.

7. The real-time monitoring method for a data-driven power energy system according to claim 1, characterized in that, The step of determining the trend correction factor for the device's most recent time window based on the comparison results of trend monitoring values ​​between adjacent time windows includes: The first and second time windows are the first group of adjacent time windows. If the trend monitoring value of the second time window is less than the trend monitoring value of the first time window, the adjacent window group is marked as 1. The comparison results of the trend monitoring values ​​of the remaining adjacent time windows are obtained, and finally the number of adjacent time window groups marked as 1 for the device is obtained. The ratio of the number of adjacent time window groups marked as 1 to the total number of adjacent time window groups is recorded as the trend correction factor of the device's most recent time window.

8. A real-time monitoring method for a data-driven power energy system according to claim 6, characterized in that, The step of determining the state weight of the most recent time window based on the ratio of the trend monitoring value of the device to the standard entropy value includes: ； The state weights are for the most recent time window. The initial state weights for this most recent time window. For the first Trend monitoring values ​​for each time window, The standard entropy value, It is the absolute value symbol.

9. The real-time monitoring method for a data-driven power energy system according to claim 1, characterized in that, The adjustment of the power system's monitoring resource allocation for the equipment based on the equipment's loss risk function and the state weight of the most recent time window includes: pass The function combines the loss risk function of the device with the product of the state weights of the most recent time window to obtain the attention level of the device in the most recent time window; Monitoring resources are allocated to devices based on the level of attention they receive in their most recent time window, and the level of attention they receive in their most recent time window is positively correlated with the amount of monitoring resources available.

10. A real-time monitoring method for a data-driven power energy system according to claim 9, characterized in that, The method of allocating monitoring resources to devices based on their most recent time window of attention includes: The attention level of the device's most recent time window is used as the input of an exponential function with base e to obtain the attention index of the device's most recent time window; the attention index and value of other devices of the same type as the device are obtained in the power energy system, and the ratio of the attention index of the device's most recent time window to the sum of the attention index values ​​is used as the attention weight of the device; the attention weight of each device is used in the Transformer attention mechanism to realize the allocation of monitoring resources.

Images