Power component data monitoring method and system based on remote automation

Abstract

The present invention application discloses a power component data monitoring method and system based on remote automation, which belongs to the field of data analysis technology. The present invention application obtains comprehensive sensor monitoring signal records to deeply characterize the operating status description of the power component, adopts a lightweight remote monitoring algorithm combined with a fault monitoring guide label, improves the monitoring accuracy, pertinence and algorithm utilization efficiency, and the generated sensor monitoring attention vector provides a comprehensive basis for abnormality discrimination. Finally, by generating abnormal discrimination results under multiple monitoring attention tasks, the fault of the power component can be discovered in a timely and accurate manner, thereby effectively solving the problems existing in the above-mentioned traditional technologies, improving the reliability and stability of the power component, and ensuring the normal operation of related systems.

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis, and specifically relates to a power supply component data monitoring method and system based on remote automation. Background Art

[0002] With the continuous development of modern technology, the importance of power components in many fields has become increasingly prominent. In data centers, a large number of servers rely on stable power supply to ensure the normal storage and processing of data; on industrial production lines, various automation equipment requires stable power supply to ensure the uninterrupted operation of the production process; in smart grid systems, power components are the key factor in the stable operation of the grid. However, various problems may occur in the operation of power components, such as voltage fluctuations, current overload, high temperature, etc. If these problems are not discovered and handled in time, they may lead to serious consequences.

[0003] Traditional power component monitoring technology has many shortcomings. On the one hand, traditional monitoring can only obtain limited power component operation data and cannot fully and carefully reflect its operating status, resulting in some potential fault hazards being ignored. For example, only focusing on the average value of voltage while ignoring detailed information such as voltage fluctuation frequency and amplitude may miss early fault signals in voltage stability.

[0004] On the other hand, in terms of fault monitoring, traditional technologies lack the ability to accurately monitor different fault types. Most traditional methods use a more general monitoring method and do not provide customized monitoring for specific fault types (such as voltage spikes, current surges, etc.). This makes it difficult to accurately determine the type and root cause of a complex power component failure, which in turn affects the efficiency and accuracy of fault repair.

[0005] In addition, traditional monitoring algorithms are usually complex and consume a lot of computing resources, which will cause computing resource constraints in large-scale systems that need to monitor multiple power components at the same time, affecting the overall performance of the monitoring system. Moreover, traditional technologies are difficult to achieve efficient processing of multiple fault monitoring tasks under limited computing resources. Summary of the invention

[0006] The present invention application provides a power component data monitoring method and system based on remote automation, which can solve or partially solve the technical problems involved in the above-mentioned background technology.

[0007] The present invention provides a power component data monitoring method based on remote automation, which is applied to a remote automation monitoring system. The method includes: obtaining a sensor monitoring signal record of a target power component, wherein the sensor monitoring signal record is used to characterize the operating state description of the target power component; obtaining X lightweight remote monitoring algorithms and Y fault monitoring guidance tags corresponding to the X lightweight remote monitoring algorithms, wherein one of the fault monitoring guidance tags is used to match an attention monitoring attribute of a monitoring attention task, and one of the lightweight remote monitoring algorithms is associated with one or more of the fault monitoring guidance tags, wherein X and Y are both positive integers, and X is not greater than Y; using the X lightweight remote monitoring algorithms to obtain the sensor monitoring signal record and the associated fault monitoring guidance tags; Under the configuration of the fault monitoring guidance tag, a sensor monitoring attention vector corresponding to each fault monitoring guidance tag is generated, and a sensor monitoring attention vector includes an operating state vector of the target power supply component and a fault monitoring sampling vector of the monitoring attention task indicated by the corresponding fault monitoring guidance tag, the operating state vector is determined by knowledge embedding the operating state description represented by the sensor monitoring signal record, and the fault monitoring sampling vector is determined by knowledge embedding the attention monitoring attribute of the monitoring attention task indicated by the corresponding fault monitoring guidance tag; the sensor monitoring attention vectors corresponding to the Y fault monitoring guidance tags are respectively subjected to abnormal discrimination operations to generate abnormal discrimination results of the target power supply component under Y types of monitoring attention tasks.

[0008] The present invention application provides a remote automated monitoring system, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the above method.

[0009] The present invention application provides a readable storage medium, on which a program or instruction is stored, and the program or instruction implements the steps of the above method when executed by a processor.

[0010] In the present invention application, comprehensive sensor monitoring signal records are obtained to deeply characterize the operating status description of the power supply component, and a lightweight remote monitoring algorithm is used in combination with fault monitoring guidance tags to improve the accuracy, pertinence and utilization efficiency of the algorithm. The generated sensor monitoring attention vector provides a comprehensive basis for abnormality discrimination. Ultimately, by generating abnormal discrimination results under multiple monitoring attention tasks, the fault of the power supply component can be discovered in a timely and accurate manner, thereby effectively solving the problems existing in the above-mentioned traditional technologies, improving the reliability and stability of the power supply component, and ensuring the normal operation of related systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1A flow chart of a power supply component data monitoring method based on remote automation provided in the present application.

[0012] Figure 2 A schematic diagram of the structure of a remote automated monitoring system provided in the present invention application. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the present invention application to clearly and completely describe the technical solutions in the present invention application. Obviously, the described embodiments are part of the embodiments of the present invention application, rather than all the embodiments. Based on the embodiments in the present invention application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention application.

[0014] The terms "first", "second", etc. in the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the present application represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0015] Figure 1 A power component data monitoring method based on remote automation is shown, which is applied to a remote automation monitoring system. The method includes the following S110-S140.

[0016] S110, the remote automation monitoring system obtains the sensor monitoring signal record of the target power supply component.

[0017] In the present application, the sensor monitoring signal record is used to characterize the operating status description of the target power supply component.

[0018] In detail, in various complex power application scenarios, target power components play a vital role. For example, in large data centers, the stable operation of many servers depends on the power components to continuously and stably provide appropriate voltage and current. In industrial automation production lines, power components provide power sources for various automation equipment (such as high-precision machine tools, automated assembly robots, etc.). In the substation of the smart grid, the power components are the key part of the stable operation of the power grid.

[0019] Based on this, a series of sensors are deployed on the target power supply components. For power supply components in data centers, there may be high-precision voltage sensors to accurately measure fluctuations in output voltage, which is essential to prevent server failures due to voltage instability. Current sensors can monitor the output current of the power supply in real time to ensure that it meets the power needs of the server cluster. Temperature sensors are installed in key heat-generating parts of power supply components, such as near power conversion modules, because excessive temperatures may cause component performance degradation or even damage.

[0020] Similarly, in industrial automation production lines, voltage sensors for power components ensure that automation equipment is supplied with a stable voltage. Current sensors can monitor current changes during equipment operation to promptly detect overload or other abnormal operating conditions. Temperature sensors help prevent power components from malfunctioning due to long-term heating, which can affect production processes.

[0021] In the substation power supply components of the smart grid, voltage sensors must meet the accuracy requirements of high voltage measurement, current sensors need to be able to accurately measure the changes in large currents, and temperature sensors are also indispensable for monitoring the operating status of large power supply equipment in substations. These sensors continuously generate sensor monitoring signals that reflect the operating status of power supply components.

[0022] In actual application, the remote automation monitoring system is connected to these sensors through a communication network. For wired connections, such as Ethernet or RS-485 communication, the receiving end of the remote automation monitoring system can stably receive the signals from the sensors. If it is a wireless connection, such as Wi-Fi, ZigBee or 4G / 5G communication, the remote automation monitoring system can also obtain signals under the appropriate communication protocol.

[0023] The sensor monitoring signal records received by the remote automation monitoring system are a data set containing a variety of information. Taking the voltage sensor signal as an example, it may contain a series of voltage values ​​at different time points. The changes in these values ​​reflect the stability of the voltage output of the power supply component. The signal record of the current sensor is also a time series, showing the dynamic changes of the current. The signal record of the temperature sensor reflects the changing trend of the temperature inside the power supply component. These sensor monitoring signal records overall characterize the operating status description of the target power supply component and provide basic data for subsequent monitoring and analysis.

[0024] S120: The remote automated monitoring system obtains X lightweight remote monitoring algorithms and Y fault monitoring guidance tags corresponding to the X lightweight remote monitoring algorithms.

[0025] In the present invention application, one of the fault monitoring guidance tags is used to match the attention monitoring attribute of a monitoring attention task, one of the lightweight remote monitoring algorithms is associated with one or more of the fault monitoring guidance tags, X and Y are both positive integers, and X is not greater than Y.

[0026] When monitoring a large number of target power components, efficient and accurate monitoring algorithms are required. Lightweight remote monitoring algorithms aim to reduce the consumption of computing resources and improve the execution efficiency of algorithms without sacrificing accuracy. In large data centers, there may be hundreds or thousands of power components that need to be monitored simultaneously. If complex and computationally intensive algorithms are used, the performance of the monitoring system will be seriously affected. Similarly, in smart grids, there are a large number of substation power components, and lightweight algorithms can help achieve efficient monitoring of large-scale power components.

[0027] A fault monitoring guidance label is an identifier used to classify and guide monitoring tasks. A fault monitoring guidance label is used to match the attention monitoring attribute of a monitoring attention task. For example, in the monitoring of power supply components in a data center, there may be a fault monitoring guidance label of "voltage stability monitoring", and the matching attention monitoring attribute may be to pay attention to the fluctuation of voltage within a certain time range, including the fluctuation amplitude, frequency, etc. Another label may be "temperature anomaly monitoring", and the corresponding attention monitoring attribute may be to pay attention to whether the temperature of the key parts of the power supply component exceeds the normal range and the rate of temperature change, etc.

[0028] A lightweight remote monitoring algorithm is associated with one or more fault monitoring guidance tags. For example, a lightweight algorithm may be associated with both the fault monitoring guidance tags of "voltage stability monitoring" and "current overload monitoring". This means that the algorithm can process the sensor monitoring signal records according to these two different attention monitoring attributes to a certain extent.

[0029] In terms of the quantitative relationship, between X lightweight remote monitoring algorithms and Y fault monitoring guidance tags, X is not greater than Y. This reflects a mapping relationship between the algorithm and the monitoring task. There may be multiple fault monitoring guidance tags sharing the same lightweight remote monitoring algorithm. For example, in the monitoring of power components in an industrial automation production line, there may be 10 fault monitoring guidance tags (Y=10), such as "voltage spike monitoring", "current surge monitoring", "long-term low voltage monitoring", etc., while there may be only 8 lightweight remote monitoring algorithms (X=8), one of which may be applicable to both "voltage spike monitoring" and "long-term low voltage monitoring" fault monitoring guidance tags.

[0030] The values ​​of X and Y depend on the specific monitoring scenario and requirements. In a relatively simple small data center, the value of Y may be small because there are relatively few types of fault monitoring that need to be paid attention to. Correspondingly, X may also be small because fewer monitoring tasks may require fewer specific algorithms to handle. In a large smart grid system, the value of Y will be larger because comprehensive fault monitoring of substation power components is required, including monitoring tasks related to various electrical parameters and environmental factors. The value of X needs to be determined based on how to efficiently handle these monitoring tasks. Some lightweight algorithms with strong versatility may be used to cover multiple fault monitoring guidance tags to improve the overall efficiency of the monitoring system.

[0031] S130, the remote automated monitoring system generates a sensor monitoring attention vector corresponding to each fault monitoring guidance tag by using the X lightweight remote monitoring algorithms under the configuration of the sensor monitoring signal record and the associated fault monitoring guidance tag.

[0032] In the present application, a sensor monitoring attention vector includes an operating state vector of the target power component and a fault monitoring sampling vector of the monitoring attention task indicated by the corresponding fault monitoring guidance tag. The operating state vector is determined by knowledge embedding the operating state description represented by the sensor monitoring signal record, and the fault monitoring sampling vector is determined by knowledge embedding the attention monitoring attributes of the monitoring attention task indicated by the corresponding fault monitoring guidance tag.

[0033] For the description of the operating state represented by the sensor monitoring signal record of the target power supply component, knowledge embedding is required to determine the operating state vector. Taking the power supply component of the data center as an example, the voltage value sequence, current value sequence and temperature value sequence in the sensor monitoring signal record are converted into vector form through a predefined knowledge embedding method. For example, the voltage value is mapped to a point in a multidimensional vector space according to certain rules, and the coordinates of this point constitute part of the operating state vector. Similarly, the current and temperature values ​​are mapped in a similar way. This knowledge embedding method can take into account factors such as the normal operating range of the power supply component and the statistical characteristics of historical data, so that the operating state vector can accurately reflect the actual operating state of the power supply component.

[0034] For the attention monitoring attributes of the monitoring attention task indicated by each fault monitoring guidance label, knowledge embedding is also required to determine the fault monitoring sampling vector. In the monitoring of power supply components in industrial automation production lines, if the fault monitoring guidance label is "current overload monitoring", its attention monitoring attributes include overload threshold, overload duration, etc. Through knowledge embedding, these attributes are converted into vector form. For example, according to the size of the overload threshold and the quantitative standard of the overload duration, it is mapped to a vector in a specific vector space. This vector can accurately reflect the characteristics related to the monitoring attention task of "current overload monitoring".

[0035] In detail, a sensor monitoring attention vector contains the operating state vector of the target power component and the fault monitoring sampling vector of the monitoring attention task indicated by the corresponding fault monitoring guidance label. In the monitoring of power components in smart grid substations, taking the fault monitoring guidance label "voltage stability monitoring" as an example, the sensor monitoring attention vector contains, on the one hand, the operating state vector obtained by embedding knowledge into the voltage sensor signal record, which reflects the actual voltage fluctuation, mean and other operating state information; on the other hand, it contains the fault monitoring sampling vector obtained by embedding knowledge based on the attention monitoring attributes of "voltage stability monitoring" (such as the allowable voltage fluctuation range, fluctuation frequency limit, etc.). The sensor monitoring attention vector composed of these two parts provides a comprehensive information basis for subsequent abnormality discrimination operations.

[0036] Based on the above, each lightweight remote monitoring algorithm works under the configuration of sensor monitoring signal records and associated fault monitoring guidance tags. In the data center power component monitoring scenario, when a lightweight remote monitoring algorithm is associated with the two fault monitoring guidance tags of "voltage stability monitoring" and "temperature anomaly monitoring", it will extract relevant information from the sensor monitoring signal record according to the attention monitoring attributes corresponding to these two tags. For voltage stability monitoring, the algorithm will focus on the voltage fluctuations in the voltage sensor signal record and process it in combination with the fault monitoring sampling vector corresponding to "voltage stability monitoring". For temperature anomaly monitoring, the algorithm will obtain temperature change information from the temperature sensor signal record and analyze it based on the fault monitoring sampling vector corresponding to "temperature anomaly monitoring".

[0037] In this way, these lightweight remote monitoring algorithms can efficiently process sensor monitoring signal records through specific algorithm structures and logic. Compared with traditional complex algorithms, they can accurately generate sensor monitoring attention vectors corresponding to each fault monitoring guidance tag while reducing computing resource consumption. In the monitoring of power components in industrial automation production lines, since a large number of power components need to be monitored in real time, this efficiency and accuracy are crucial to ensure the normal operation of the production line. If the algorithm processing speed is slow or the accuracy is poor, it may lead to misjudgment or omission of power component failures, thereby affecting the production efficiency of the entire production line.

[0038] S140, the remote automated monitoring system performs abnormality identification operations on the sensor monitoring attention vectors corresponding to the Y fault monitoring guidance tags, respectively, and generates abnormality identification results of the target power supply component under Y types of monitoring attention tasks.

[0039] In this step, the remote automation monitoring system performs abnormality discrimination operations on the sensor monitoring attention vectors corresponding to the Y fault monitoring guidance tags. In the monitoring of power components in large data centers, for the sensor monitoring attention vector corresponding to "voltage stability monitoring", the system will compare and analyze the operating state vector (including actual voltage fluctuation information, etc.) and the fault monitoring sampling vector (including standard information on voltage stability, etc.) in the vector. If the voltage fluctuation in the operating state vector exceeds the normal range specified by the fault monitoring sampling vector, it may be determined that the voltage stability is abnormal.

[0040] In the monitoring of power components in smart grid substations, for the sensor monitoring attention vector corresponding to "temperature anomaly monitoring", the system will check the temperature value change in the operating state vector and the normal temperature range in the fault monitoring sampling vector. If the temperature value is continuously higher than the normal range or the temperature change rate is abnormal, it will be judged as a temperature anomaly.

[0041] It is worth mentioning that there are many methods for abnormal discrimination operation. For example, a threshold-based discrimination method can be used, that is, the threshold information in the fault monitoring sampling vector can be used to directly judge whether the relevant parameters in the operating state vector are out of range. A discrimination method based on a machine learning model can also be used, such as a pre-trained support vector machine (SVM) model or a neural network model, which takes the sensor monitoring attention vector as input and judges whether there is an abnormality through the output of the model. In the monitoring of power supply components in industrial automation production lines, for "current overload monitoring", if a threshold-based method is used, as long as the current value exceeds the pre-set overload threshold, it is judged as an overload abnormality; if a neural network model is used, more factors can be considered, such as historical current data, association with other devices, etc., so as to more accurately judge whether there is a real overload abnormality.

[0042] It can be understood that after the abnormal discrimination operation of the sensor monitoring attention vector corresponding to Y fault monitoring guidance tags, the remote automation monitoring system generates the abnormal discrimination results of the target power component under Y monitoring attention tasks. In the monitoring of power components in the data center, this means that for various monitoring attention tasks such as voltage stability, temperature anomalies, current overload, etc., the system can give clear abnormal discrimination results. If an abnormal situation occurs, the system can accurately point out which monitoring attention task has a problem, such as abnormal voltage stability or fault risk caused by excessive temperature.

[0043] In addition, these abnormality identification results provide a basis for subsequent treatment measures. In the monitoring of power components in smart grid substations, if the voltage is determined to be abnormal, the operation and maintenance personnel can take corresponding measures according to the specific circumstances of the abnormality (such as the degree and duration of the voltage being too high or too low), such as adjusting the voltage regulation equipment of the substation. In the monitoring of power components in industrial automation production lines, if a current overload abnormality is found, it may be necessary to check whether the relevant equipment is faulty or adjust the power distribution in the production process to avoid further equipment damage and production interruption.

[0044] In the above invention application, the operating state description, attention monitoring attributes, sensor monitoring attention vector, and fault monitoring sampling vector are core technical features. In order to facilitate those skilled in the art to implement the invention application more clearly and completely, these core technical features are explained below through relevant examples.

[0045] Example of operating status description

[0046] 1. Data center power supply components

[0047] 1. Voltage related

[0048] For example, the voltage of the power supply components in a data center is monitored, and data is collected once a second for 10 minutes (a total of 600 data points). The voltage-related part of the operating state vector can be expressed as: (V=[v_1, v_2, cdots, v_{600}]), where (v_i) represents the voltage value collected at the (i)th second (in volts).

[0049] At the same time, in order to describe the voltage state more comprehensively, statistical characteristics of the voltage can also be added, such as: mean (bar{V}=frac{1}{600}sum_{i=1}^{600}v_i); standard deviation (sigma_V=sqrt{frac{1}{600}sum_{i=1}^{600}(v_i-bar{V})^2}), then the characteristic vector of the voltage part becomes (V'=[v_1, v_2, cdots, v_{600}, bar{V}, sigma_V]).

[0050] 2. Current related

[0051] Similarly, with a period of 10 minutes, current data is collected every second to obtain a current value sequence (I=[i_1, i_2, cdots, i_{600}]) (in amperes). Add the maximum current (I_{max}=max{i_1, i_2, cdots, i_{600}}), the minimum current (I_{min}=min{i_1, i_2, cdots, i_{600}}) and the mean current change rate (bar{r_I}), where (r_{I_j}=frac{i_{j+1}-i_j}{i_j})((j=1, 2, cdots, 599)), (bar{r_I}=frac{1}{599}sum_{j=1}^{599}r_{I_j}). The eigenvector of the current part can be (I'=[i_1, i_2, cdots, i_{600}, I_{max}, I_{min}, bar{r_I}]).

[0052] 3. Temperature related

[0053] Temperature sensors are installed at three key parts of the power supply components (such as the power conversion module, heat sink and capacitor group), and temperature data is collected every 30 seconds, for a total of 20 times in 10 minutes. Assume that the temperature sequences of the three parts are (T_1=[t_{11}, t_{12}, cdots, t_{1, 20}]), (T_2=[t_{21}, t_{22}, cdots, t_{2, 20}]), (T_3=[t_{31}, t_{32}, cdots, t_{3, 20}]) (in degrees Celsius).

[0054] Calculate the average temperature at each location:

[0055] (bar{T_1}=frac{1}{20}sum_{k=1}^{20}t_{1k}),

[0056] (bar{T_2}=frac{1}{20}sum_{k=1}^{20}t_{2k}),

[0057] (bar{T_3}=frac{1}{20}sum_{k=1}^{20}t_{3k}), as well as the overall maximum temperature (T_{max}=max{max{T_1}, max{T_2}, max{T_3}}) and minimum temperature (T_{min}=min{min{T_1}, min{T_2}, min{T_3}}). The eigenvector of the temperature part is (T'=[t_{11}, t_{12}, cdots, t_{1, 20}, t_{21}, t_{22}, cdots, t_{2, 20}, t_{31}, t_{32}, cdots, t_{3, 20}, bar{T_1}, bar{T_2}, bar{T_3}, T_{max}, T_{min}]).

[0058] In summary, the operating state vector of the data center power supply component can be exemplified as: (S_{data-center}=[V', I', T']), which is a complex vector containing a large amount of voltage, current and temperature information, and can comprehensively describe the operating state of the power supply component.

[0059] 2. Power supply components for industrial automation production lines

[0060] 1. Load related (except current)

[0061] For power supply components of industrial automation production lines, in addition to basic voltage and current monitoring, it is also necessary to pay attention to the load conditions. For example, by measuring the load power through a power sensor, data is collected every 5 seconds for 5 minutes (a total of 60 data points), and a power sequence (P=[p_1, p_2, cdots, p_{60}]) is obtained (in watts). At the same time, according to the equipment operation mode of the production line, there are different load factors (lambda), such as in the startup phase (lambda_1=1.2), normal operation phase (lambda_2=1), and standby phase (lambda_3=0.5). These load factors can be combined with power data to form a load-related feature vector (P'=[p_1, p_2, cdots, p_{60}, lambda_1, lambda_2, lambda_3]).

[0062] 2. Voltage and frequency coupling (for special equipment)

[0063] In some equipment that has special requirements for voltage and frequency (such as some high-precision processing equipment), it is necessary to monitor voltage and frequency simultaneously. For example, collect voltage (v_i) and frequency (f_i) data every 2 seconds for 8 minutes (a total of 240 data points). Calculate the covariance of voltage and frequency (Cov(V,F)=frac{1}{240}sum_{i=1}^{240}(v_i-bar{V})(f_i-bar{F})), where (bar{V}) and (bar{F}) are the means of voltage and frequency, respectively. The eigenvector of this part is (VF'=[v_1,f_1,v_2,f_2,cdots,v_{240},f_{240},bar{V},bar{F},Cov(V,F)]).

[0064] Taking into account factors such as current (collected and processed in a manner similar to that of a data center (I'')), temperature (for example, collected at two key locations to obtain (T'')), the operating state vector of the power supply component of an industrial automation production line (S_{industry}=[P', VF', I'', T'']) is obtained.

[0065] The meanings of the above-mentioned formulas for describing the operating status are as follows.

[0066] Data Center Power Components - Voltage Related

[0067] (V=[v_1, v_2, cdots, v_{600}]) and (v_i) meaning: Here (V) is a vector used to represent the voltage data sequence collected within 10 minutes (collected once per second, a total of 600 data points). (v_i) represents the voltage value collected at the (i)th second, in volts.

[0068] (bar{V}=frac{1}{600}sum_{i=1}^{600}v_i) Meaning: This formula calculates the average value of the 600 collected voltage values. (sum_{i=1}^{600}v_i) means adding the 600 voltage values ​​(v_1) to (v_{600}) and then dividing by 600 to get the average value (bar{V}).

[0069] (sigma_V=sqrt{frac{1}{600}sum_{i=1}^{600}(v_i-bar{V})^2}) Meaning: This is the standard deviation of the calculated voltage value. First, calculate the difference ((v_i-bar{V})) between each voltage value (v_i) and the average value (bar{V}), square these differences and add them up (sum_{i=1}^{600}(v_i-bar{V})^2), then divide by 600, and finally take the square root to get the standard deviation (sigma_V). The standard deviation can reflect the degree of discreteness of the voltage data.

[0070] Data Center Power Components - Current Related

[0071] The meaning of (I=[i_1, i_2, cdots, i_{600}]) and (i_j) is: (I) is a vector representing the current data within 10 minutes (collected once per second), and (i_j) represents the current value collected at the (j)th second, in amperes.

[0072] (I_{max}=max{i_1,i_2,cdots,i_{600}}) and (I_{min}=min{i_1,i_2,cdots,i_{600}}) mean: (I_{max}) is the maximum value among these 600 current values, and (I_{min}) is the minimum value.

[0073] (r_{I_j}=frac{i_{j+1}-i_j}{i_j})

[0074] The sum (bar{r_I}=frac{1}{599}sum_{j=1}^{599}r_{I_j}) means: (r_{I_j}) represents the rate of change of the current from the (j)th second to the (j+1)th second, that is, the difference between two adjacent current values ​​divided by the current current value. (bar{r_I}) is the average value of the 599 current change rates.

[0075] Data Center Power Components - Temperature Related

[0076] (bar{T_1}=frac{1}{20}sum_{k=1}^{20}t_{1k}),

[0077] (bar{T_2}=frac{1}{20}sum_{k=1}^{20}t_{2k}),

[0078] (bar{T_3}=frac{1}{20}sum_{k=1}^{20}t_{3k}) Meaning: Collect temperature data at three key parts of the power supply component, once every 30 seconds, and collect 20 times in 10 minutes. Here, the average values ​​of the temperature data (T_1), (T_2), and (T_3) of the three parts are calculated respectively. For example, (bar{T_1}) is the sum of the 20 temperature values ​​of part 1 (t_{11}) to (t_{1, 20}) and divided by 20.

[0079] The meaning of (T_{max}=max{max{T_1}, max{T_2}, max{T_3}}) and (T_{min}=min{min{T_1}, min{T_2}, min{T_3}}) is: (T_{max}) is the maximum value among the maximum values ​​of the temperature data of the three parts, and (T_{min}) is the minimum value among the minimum values ​​of the temperature data of the three parts.

[0080] Industrial Automation Production Line Power Components - Load Related

[0081] (P=[p_1, p_2, cdots, p_{60}]) and (p_i) mean: (P) is a vector representing power data. The power sensor collects data every 5 seconds for 5 minutes (60 data points in total). (p_i) represents the power value collected for the (i)th time, in watts.

[0082] Meaning of load coefficient (lambda): The coefficient set according to different operating modes of the production line equipment (such as startup phase (lambda_1=1.2), normal operating phase (lambda_2=1), and standby phase (lambda_3=0.5)). These coefficients are combined with power data to form a load-related characteristic vector (P'), which can reflect the load conditions under different operating modes.

[0083] Industrial Automation Production Line Power Components - Voltage and Frequency Coupling

[0084] (Cov(V,F)=frac{1}{240}sum_{i=1}^{240}(v_i-bar{V})(f_i-bar{F})) Meaning: For devices with special requirements for voltage and frequency, voltage (v_i) and frequency (f_i) data are collected every 2 seconds for 8 minutes (a total of 240 data points). This formula calculates the covariance of voltage and frequency, which can reflect the degree of linear relationship between voltage and frequency. (bar{V}) and (bar{F}) are the average values ​​of voltage and frequency, respectively.

[0085] Example of numerical feature vectors for attention monitoring attributes

[0086] 1. Attention monitoring attribute vector for “voltage stability monitoring”

[0087] 1. Voltage fluctuation range

[0088] For the "voltage stability monitoring" of the power supply components in the data center, the normal voltage range is set to (V_{min}=110V) to (V_{max}=120V). It is represented as ([V_{min}, V_{max}]) in the feature vector.

[0089] 2. Allowable fluctuation frequency

[0090] For example, within 1 minute, the maximum number of voltage fluctuations allowed is (n=5), and the amplitude of each fluctuation cannot exceed (DeltaV=2V). This part is represented in the vector as ([n, DeltaV]).

[0091] 3. Limitation of fluctuation duration

[0092] If the voltage fluctuation lasts longer than (t=10s), it is considered abnormal. This feature is represented in the vector as ([t]).

[0093] In summary, the attention monitoring attribute vector of “voltage stability monitoring” is: (A_{voltage-stability}=[V_{min}, V_{max}, n, DeltaV, t]).

[0094] 2. Attention monitoring attribute vector for “temperature anomaly monitoring”

[0095] 1. Normal temperature range (multiple parts)

[0096] For data center power components, the normal temperature range of the power conversion module is (T_{1, min} = 40^{circ}C) to (T_{1, max} = 60^{circ}C), the normal temperature range of the heat sink is (T_{2, min} = 30^{circ}C) to (T_{2, max} = 50^{circ}C), and the normal temperature range near the capacitor group is (T_{3, min} = 35^{circ}C) to (T_{3, max} = 55^{circ}C). It is represented in the feature vector as ([T_{1, min}, T_{1, max}, T_{2, min}, T_{2, max}, T_{3, min}, T_{3, max}]).

[0097] 2. Temperature rise rate limit

[0098] For example, within 10 minutes, the temperature rise rate cannot exceed (r = 2^{circ}C / min). This part is represented in the vector as ([r]).

[0099] Therefore, the attention monitoring attribute vector of "temperature anomaly monitoring" can be:

[0100] (A_{temperature-anomaly}=[T_{1, min}, T_{1, max}, T_{2, min}, T_{2, max}, T_{3, min}, T_{3, max}, r]).

[0101] Fault Monitoring Sampling Vector Example

[0102] 1. Fault monitoring sampling vector based on “voltage stability monitoring”

[0103] Combining operating status and monitoring attributes: For example, the voltage part of the operating state vector (V'=[v_1, v_2, cdots, v_{600}, bar{V}, sigma_V]) and the attention monitoring attribute vector (A_{voltage-stability}=[V_{min}, V_{max}, n, DeltaV, t]) have been obtained. The fault monitoring sampling vector (F_{voltage-stability}) can be constructed as follows: First, calculate the number of points where the voltage exceeds the normal range:

[0104] (m=sum_{i=1}^{60}I(v_i<V_{min}text{or}v_i> V_{max})), where (I(cdot)) is the indicator function. Then, the number of fluctuations (n') (the actual number of fluctuations is counted by a certain algorithm) and the number of times the fluctuation amplitude exceeds (DeltaV) (n''). Fault monitoring sampling vector (F_{voltage-stability}=[m, n', n'', bar{V}, sigma_V]) is calculated.

[0105] (II) Fault monitoring sampling vector based on “temperature anomaly monitoring”

[0106] Comprehensive temperature-related information: For the temperature part in the operating state vector (T'=[t_{11}, t_{12}, cdots, t_{1, 20}, t_{21}, t_{22}, cdots, t_{2, 20}, t_{31}, t_{32}, cdots, t_{3, 20}, bar{T_1}, bar{T_2}, bar{T_3}, T_{max}, T_{min}]) and the attention monitoring attribute vector (A_{temperature-anomaly}=[T_{1, min}, T_{1, max}, T_{2, min}, T_{2, max}, T_{3, min}, T_{3, max}, r]).

[0107] Calculate the number of points outside the normal temperature range: (m_1=sum_{k=1}^{20}(I(t_{1k}<T_{1,min}text{or}t_{1k}> T_{1,max})+I(t_{2k}<T_{2,min}text{or}t_{2k}> T_{2,max})+I(t_{3k}<T_{3,min}text{or}t_{3k}> T_{3,max}))), temperature rise rate (r') (by calculating the actual temperature rise rate).

[0108] The fault monitoring sampling vector is:

[0109] (F_{temperature-anomaly}=[m_1, r', bar{T_1}, bar{T_2}, bar{T_3}, T_{max}, T_{min}]).

[0110] Among them, the relevant formulas of attention monitoring attributes have the following meanings.

[0111] Attention monitoring attribute vector for "voltage stability monitoring"

[0112] (V_{min}=110V) to (V_{max}=120V) Meaning: This is the set range of normal voltages for data center power components, (V_{min}) is the lower limit, and (V_{max}) is the upper limit.

[0113] (n = 5), (Delta V = 2V), and (t = 10s) mean: within 1 minute, (n = 5) means that the maximum number of voltage fluctuations allowed is 5 times, (Delta V = 2V) means that the amplitude of each fluctuation cannot exceed 2 volts, and (t = 10s) means that if the voltage fluctuation lasts for more than 10 seconds, it is considered abnormal. These parameters constitute the attention monitoring attribute vector (A_{voltage-stability}) of "voltage stability monitoring".

[0114] Attention monitoring attribute vector for "temperature anomaly monitoring"

[0115] (T_{1, min}=40^{circ}C) to (T_{1, max}=60^{circ}C), (T_{2, min}=30^{circ}C) to (T_{2, max}=50^{circ}C), (T_{3, min}=35^{circ}C) to (T_{3, max}=55^{circ}C) Meaning: These are the normal temperature ranges of data center power components near the power conversion module, heat sink and capacitor group.

[0116] (r = 2^{circ}C / min) means: In 10 minutes, the temperature rise rate cannot exceed (2^{circ}C / min). These parameters constitute the attention monitoring attribute vector (A_{temperature-anomaly}) of "temperature anomaly monitoring".

[0117] The meanings of the relevant formulas for the fault monitoring sampling vector are as follows.

[0118] Fault monitoring sampling vector based on "voltage stability monitoring"

[0119] (m=sum_{i = 1}^{600}I(v_i<V_{min} text{ or} v_i> V_{max})) Meaning: Here (I(cdot)) is an indicator function. If (v_i) is less than (V_{min}) or greater than (V_{max}) (i.e. the voltage value is outside the normal range), the value of (I) is 1, otherwise it is 0. (m) is the number of points where the voltage is outside the normal range.

[0120] Calculate the number of fluctuations (n') and the number of times (n'') the fluctuation amplitude exceeds (Delta V). Meaning: This is to count the number of actual voltage fluctuations (n') and the number of times (n'') the fluctuation amplitude exceeds the set value (Delta V) through a certain algorithm. These parameters, together with the average value (bar{V}) and standard deviation (sigma_V) of the voltage, constitute the fault monitoring sampling vector (F_{voltage-stability}).

[0121] Fault monitoring sampling vector based on "temperature anomaly monitoring"

[0122] (m_1=sum_{k = 1}^{20}(I(t_{1k}<T_{1,min} text{ or} t_{1k}> T_{1,max})+I(t_{2k}<T_{2,min} text{ or} t_{2k}> T_{2,max})+I(t_{3k}<T_{3,min} text{ or} t_{3k}> Meaning of T_{3, max}))): Here, the indicator function (I) is also used to calculate the sum of the number of points where the temperatures of the three parts exceed their respective normal ranges. (t_{1k}), (t_{2k}), and (t_{3k}) are the temperature values ​​of the three parts respectively, and (T_{1, min}), (T_{1, max}), (T_{2, min}), (T_{2, max}), (T_{3, min}), and (T_{3, max}) are the normal temperature ranges of the corresponding parts.

[0123] Temperature rise rate (r') means: It is obtained by calculating the actual temperature rise rate (r'). These parameters, together with the average temperature of each part (bar{T_1}), (bar{T_2}), (bar{T_3}), and the overall maximum temperature (T_{max}) and minimum temperature (T_{min}), constitute the fault monitoring sampling vector (F_{temperature-anomaly}).

[0124] Example of sensor monitoring attention vector

[0125] 1. Sensor monitoring attention vector corresponding to “voltage stability monitoring”

[0126] Combined operation status and fault monitoring sampling vector: Taking the data center power supply component as an example, the voltage-related part (V') in the operation status vector (S_{data-center}) and the fault monitoring sampling vector (F_{voltage-stability}) based on "voltage stability monitoring" are combined into the sensor monitoring attention vector (SA_{voltage-stability}=[V', F_{voltage-stability}]). This is a vector that combines the actual voltage operation status and the sampling information related to voltage stability fault monitoring, which can provide a comprehensive basis for subsequent abnormality judgment.

[0127] 2. Sensor monitoring attention vector corresponding to “temperature anomaly monitoring”

[0128] Constructing a complete vector: Similarly, for the data center power supply components, the temperature-related part (T') in the operating state vector (S_{data-center}) and the fault monitoring sampling vector (F_{temperature-anomaly}) based on "temperature anomaly monitoring" are combined to obtain the sensor monitoring attention vector (SA_{temperature-anomaly}=[T', F_{temperature-anomaly}]).

[0129] Among them, the relevant formulas of the sensor monitoring attention vector are as follows.

[0130] The meaning of the sensor monitoring attention vector corresponding to "voltage stability monitoring" (SA_{voltage-stability}=[V', F_{voltage-stability}]) is:

[0131] This vector combines the voltage-related part (V') in the operating state vector (S_{data-center}) and the fault monitoring sampling vector (F_{voltage-stability}) based on "voltage stability monitoring", integrating the actual state of voltage operation and the sampling information related to voltage stability fault monitoring for subsequent abnormality identification.

[0132] The sensor monitoring attention vector corresponding to "temperature anomaly monitoring" (SA_{temperature-anomaly}=[T', F_{temperature-anomaly}]) means:

[0133] The temperature-related part (T') in the operating state vector (S_{data-center}) and the fault monitoring sampling vector (F_{temperature-anomaly}) based on "temperature anomaly monitoring" are combined to provide a comprehensive basis for anomaly discrimination in temperature anomaly monitoring.

[0134] In summary, the present invention first characterizes the operating status of the target power component by acquiring the sensor monitoring signal record of the target power component, which enables the working condition of the power component to be captured comprehensively and meticulously. Whether it is a slight fluctuation in voltage or current, or a change in other parameters such as temperature, it can be recorded by the sensor monitoring signal, providing a rich data basis for subsequent analysis.

[0135] This design is highly flexible and targeted in obtaining X lightweight remote monitoring algorithms and corresponding Y fault monitoring guidance tags. The lightweight remote monitoring algorithm can process data efficiently without consuming too much computing resources. Since an algorithm can be associated with one or more fault monitoring guidance tags, the algorithm can be reused, reducing the workload of algorithm development and improving the utilization efficiency of the algorithm. Fault monitoring guidance tags are used to match the attention monitoring attributes of specific monitoring attention tasks, which helps to accurately focus on different types of fault monitoring tasks, such as classified monitoring of different focus points such as voltage stability, current overload or temperature anomaly, improving the accuracy and targeting of fault monitoring.

[0136] The above-mentioned lightweight remote monitoring algorithm is used to generate a sensor monitoring attention vector under the configuration of sensor monitoring signal records and associated fault monitoring guidance tags. This process integrates the operating state information and sampling information related to fault monitoring. The operating state vector is determined by embedding knowledge into the operating state description, and the fault monitoring sampling vector is determined by embedding knowledge into the attention monitoring attribute. Such a sensor monitoring attention vector integrates multiple key information. This vector generation method can not only more accurately reflect the state of the power supply component under different monitoring attention tasks, but also can mine deeper relationships between data, providing a more comprehensive and accurate basis for abnormality identification.

[0137] Finally, the sensor monitoring attention vectors corresponding to the Y fault monitoring guidance tags are respectively subjected to abnormal discrimination operations to generate abnormal discrimination results of the target power component under Y monitoring attention tasks. This multi-task abnormal discrimination method can comprehensively detect various possible fault conditions of the power component. By distinguishing different monitoring attention tasks, the fault type can be accurately located, whether it is an abnormality in voltage, an abnormality in current or other faults related to the power component, they can be discovered in time. This helps to take corresponding measures in the early stages of the fault to avoid further deterioration of the fault, thereby improving the reliability and stability of the power component and reducing the system downtime and possible losses caused by power component failure, such as preventing data loss due to power failure in data centers, or avoiding production interruptions due to power problems on industrial production lines.

[0138] In some examples, any one of the X lightweight remote monitoring algorithms is a target lightweight remote monitoring algorithm, and any fault monitoring guidance tag corresponding to the target lightweight remote monitoring algorithm is a target fault monitoring guidance tag. The method of using the X lightweight remote monitoring algorithms to generate a sensor monitoring attention vector corresponding to each fault monitoring guidance tag under the configuration of the sensor monitoring signal record and the associated fault monitoring guidance tag includes: using the target lightweight remote monitoring algorithm to generate first quantized embedded knowledge of the target power component under the configuration of the sensor monitoring signal record, and the first quantized embedded knowledge is used to characterize the operating state vector of the target power component; using the target lightweight remote monitoring algorithm to generate second quantized embedded knowledge of the target power component under the configuration of the target fault monitoring guidance tag, and the second quantized embedded knowledge is used to characterize the fault monitoring sampling vector of the monitoring attention task indicated by the target fault monitoring guidance tag; performing knowledge collision on the first quantized embedded knowledge and the second quantized embedded knowledge to generate the sensor monitoring attention vector corresponding to the target fault monitoring guidance tag.

[0139] According to the above, the sensor monitoring signal record contains various information of the target power supply component, such as voltage, current, temperature, etc. For the target lightweight remote monitoring algorithm, it works under the configuration of the sensor monitoring signal record.

[0140] Taking the power supply components of the data center as an example, the voltage value sequence, current value sequence and temperature value sequence in the sensor monitoring signal record need to be quantized. For example, in the processing of voltage values, each voltage value is mapped to a point in a multidimensional vector space according to predefined rules. This process involves considering factors such as the normal operating range of the power supply component and the statistical characteristics of historical data. A similar approach is used for current and temperature values ​​to convert these actual numerical information into vector form. The first quantized embedded knowledge finally generated can accurately characterize the operating state vector of the target power supply component. This operating state vector covers various aspects of the operating state description information of the power supply component and is an important component of the subsequent construction of the sensor monitoring attention vector.

[0141] In addition, the target fault monitoring guide label has a specific attention monitoring attribute, and the target lightweight remote monitoring algorithm generates a second quantized embedded knowledge under the configuration of the target fault monitoring guide label.

[0142] Taking the fault monitoring guidance label "voltage stability monitoring" as an example, its attention monitoring attributes include voltage fluctuation range, allowed fluctuation frequency, fluctuation duration limit, etc. Through knowledge embedding, these attributes are converted into vector form. For example, according to the upper and lower limits of the voltage fluctuation range, the limit of the fluctuation frequency, and the quantitative standard of the fluctuation duration, they are mapped to a vector in a specific vector space. This vector is the second quantitative embedded knowledge, which is used to characterize the fault monitoring sampling vector of the monitoring attention task indicated by the target fault monitoring guidance label. In other types of fault monitoring guidance labels such as "temperature anomaly monitoring", the corresponding second quantitative embedded knowledge is generated in a similar way according to the attention monitoring attributes such as the normal temperature range and the temperature rise rate limit.

[0143] Based on this, after obtaining the first quantized embedded knowledge (characterizing the operating state vector) and the second quantized embedded knowledge (characterizing the fault monitoring sampling vector), the two are subjected to knowledge collision.

[0144] This knowledge collision is not a simple addition or combination, but a fusion based on the logic and algorithm structure inside the target lightweight remote monitoring algorithm. Taking the operation corresponding to "voltage stability monitoring" as an example, the actual voltage fluctuation, mean and other operating state information contained in the operating state vector (from the first quantized embedded knowledge) and the fault monitoring sampling vector (second quantized embedded knowledge) obtained by knowledge embedding according to the attention monitoring attributes of "voltage stability monitoring" (such as the allowable voltage fluctuation range, fluctuation frequency limit, etc.) are integrated through specific algorithm operations. For example, the relevant parameters in the operating state vector may be adjusted or re-weighted according to the threshold information in the fault monitoring sampling vector, thereby generating a vector that integrates the two aspects of key information. This vector is the sensor monitoring attention vector corresponding to the target fault monitoring guidance tag. The operations corresponding to other fault monitoring guidance tags are similar in principle. Through knowledge collision, the operating state information and the sampling information related to fault monitoring are integrated into one vector, providing a comprehensive information basis for subsequent abnormal discrimination operations.

[0145] In this way, the sensor monitoring signal records and fault monitoring guidance labels are first converted into the first and second quantized embedded knowledge respectively through the target lightweight remote monitoring algorithm, and then the knowledge collision is performed to obtain the vector. This process can effectively integrate the operating status information and fault monitoring related information, improve the accuracy of the characterization of the operating status of the power supply component, provide a more comprehensive and accurate basis for subsequent abnormality judgment, help to accurately monitor the operating status of the power supply component, timely discover potential faults, and improve the reliability and stability of the power supply component.

[0146] In other preferred embodiments, the target lightweight remote monitoring algorithm includes Z knowledge collision networks in a cascade state for realizing knowledge vector embedding, and any knowledge collision network includes an operating status knowledge collision module and a fault monitoring knowledge collision module in parallel, and Z is a positive integer; wherein, the target lightweight remote monitoring algorithm includes a first original embedded vector relationship network to be optimized, and the Z knowledge collision networks include the uth knowledge collision network, u is a positive integer, and u is not greater than Z, and the uth knowledge collision network has an input embedded vector relationship network; if u is equal to 1, then the input embedded vector relationship network is the first original embedded vector relationship network, and if u is greater than 1, then the input embedded vector relationship network is the output result of the u-1th knowledge collision network of the target lightweight remote monitoring algorithm, and the output result is obtained after knowledge collision between the output result of the operating status knowledge collision module in the u-1th knowledge collision network and the output result of the fault monitoring knowledge collision module.

[0147] It can be understood that the target lightweight remote monitoring algorithm contains Z knowledge collision networks in a cascade state. These knowledge collision networks play a key role in the entire algorithm and are used to realize knowledge vector embedding. Each knowledge collision network contains a parallel operation status knowledge collision module and a fault monitoring knowledge collision module. This structural design helps to effectively process and integrate different types of knowledge.

[0148] For example, taking the previously mentioned "voltage stability monitoring" as an example, when constructing the voltage-related sensor monitoring attention vector, these modules will process the voltage operating status information and fault monitoring related information respectively.

[0149] The target lightweight remote monitoring algorithm includes a first original embedding vector relationship network to be optimized. For the u-th knowledge collision network (u is a positive integer and u is not greater than Z) among the Z knowledge collision networks, it has an input embedding vector relationship network.

[0150] When u is equal to 1, this input embedding vector relationship network is the first original embedding vector relationship network. This is the starting point of the entire knowledge collision network processing flow. For example, when processing the initial sensor monitoring signal record of the power component, the first original embedding vector relationship network will accept these raw data as input.

[0151] When u is greater than 1, the input embedding vector relationship network is the output result of the u-1th knowledge collision network of the target lightweight remote monitoring algorithm. This output result is obtained after the knowledge collision between the output result of the operating status knowledge collision module in the u-1th knowledge collision network and the output result of the fault monitoring knowledge collision module. This means that the output of each knowledge collision network will be used as the input of the next knowledge collision network, forming a cascade processing flow. Taking "temperature anomaly monitoring" as an example, if there are multiple knowledge collision networks involved in processing temperature-related information, the processed and fused temperature operating status and fault monitoring information output by the previous knowledge collision network will be used as the input of the next knowledge collision network, and the temperature-related information will be gradually processed and optimized in a more in-depth manner.

[0152] In each knowledge collision network, the operating state knowledge collision module and the fault monitoring knowledge collision module work in parallel. The operating state knowledge collision module mainly processes information related to the operating state of the power supply component, such as the voltage, current, temperature and other data extracted from the sensor monitoring signal record after being quantified and embedded in the knowledge. The fault monitoring knowledge collision module focuses on processing the attention monitoring attribute information related to the fault monitoring guidance label, such as the voltage fluctuation range, fluctuation frequency and other information in "voltage stability monitoring" after being quantified and embedded in the knowledge.

[0153] Then, the output results of these two modules will be collided with each other. Taking "current overload monitoring" as an example, the operating state knowledge collision module may output a vector related to the actual operating state of the current, such as a vector of the current numerical sequence, maximum value, minimum value and other information after processing. The fault monitoring knowledge collision module may output a vector of the threshold value, overload duration and other information related to the current overload judgment after processing. The knowledge collision of these two vectors will combine the information from these two aspects to obtain a more comprehensive and accurate intermediate result about current overload monitoring. If this intermediate result is generated in the u-th knowledge collision network (u is less than Z), it will be used as part of the input embedding vector relationship network of the next knowledge collision network (u+1th knowledge collision network) and continue the next round of processing until the processing flow of Z knowledge collision networks is completed, and finally a complete sensor monitoring attention vector is obtained.

[0154] It can be understood that the target lightweight remote monitoring algorithm structure in this embodiment helps to efficiently process the monitoring information of the power supply component. Through the cascaded knowledge collision network, the operating status information and fault monitoring information can be gradually and deeply integrated. The operating status knowledge collision module and the fault monitoring knowledge collision module work in parallel and perform knowledge collision design to make information processing more targeted and accurate. This algorithm structure can improve the accuracy and efficiency of generating sensor monitoring attention vectors, which in turn helps to more accurately monitor the faults of power supply components and improve the reliability and stability of the operation of power supply components.

[0155] In an optional embodiment, the use of the target lightweight remote monitoring algorithm to generate the first quantized embedded knowledge of the target power component under the configuration of the sensor monitoring signal record includes: using the operating status knowledge collision module in the u-th knowledge collision network to perform feature de-interference on the input embedded vector relationship network under the configuration of the sensor monitoring signal record, and generate the power component state encoding of the u-th knowledge collision network for the target power component; the Y-th knowledge collision network of the target lightweight remote monitoring algorithm is used to encode the power component state of the target power component, and is determined as the first quantized embedded knowledge; wherein, the operation of performing feature de-interference on the input embedded vector relationship network under the configuration of the sensor monitoring signal record is an operation of performing knowledge collision on the sensor monitoring signal record and the input embedded vector relationship network.

[0156] In the target lightweight remote monitoring algorithm, when the algorithm is used to generate the first quantized embedded knowledge under the configuration of the sensor monitoring signal record, it involves using the operation state knowledge collision module in the u-th knowledge collision network. First, the input embedded vector relationship network is feature de-interfered under the configuration of the sensor monitoring signal record. The feature de-interference operation here is actually an operation of performing knowledge collision on the sensor monitoring signal record and the input embedded vector relationship network.

[0157] Taking "voltage stability monitoring" as an example, the voltage part of the sensor monitoring signal record is set to include a sequence of voltage values ​​collected once per second within 10 minutes, such as [112, 113, 111, 114, ...] (in volts). The input embedding vector relationship network may contain some initial vector information related to voltage in the previous processing process, such as a rough estimate of the voltage fluctuation range vector [110, 120] (representing the previously estimated lower and upper limits of normal voltage). Through the operation state knowledge collision module, the two are subjected to knowledge collision (i.e., feature de-interference), which can remove some factors that may interfere with the accurate characterization of the operation state of the power supply component.

[0158] The specific knowledge collision algorithm can be an algorithm based on weight adjustment. For each voltage value in the sensor monitoring signal record, the weight is adjusted according to its relationship with the voltage range in the input embedding vector relationship network. For example, if the voltage value is close to the boundary of the voltage range, its weight in subsequent encoding may be reduced because it may indicate an approaching unstable state. After such processing, the power component state encoding of the u-th knowledge collision network for the target power component is generated. This power component state encoding is a coded representation of the operating state of the target power component under the current knowledge collision network. It contains processed key information related to the operating state, such as a voltage value sequence after adjusting the weight and some marks indicating the tendency of voltage stability.

[0159] After completing the above steps, the Yth knowledge collision network of the target lightweight remote monitoring algorithm is determined as the first quantized embedded knowledge for the power component state encoding of the target power component. The Yth knowledge collision network here is the last one in the entire cascade of knowledge collision networks (because Y is not greater than Z, and in this case it is the last network related to determining the first quantized embedded knowledge).

[0160] For example, in the case of "temperature anomaly monitoring", it is assumed that the temperature-related sensor monitoring signal records and related vector relationship networks are gradually processed by the previous multiple knowledge collision networks. The temperature part of the sensor monitoring signal record is the temperature data collected every 30 seconds at three key parts of the power supply component, such as [42, 45, 43, ...] (in degrees Celsius), and the related vector relationship network contains information such as the normal temperature range of different parts. The power supply component state code output by the last knowledge collision network (the Yth knowledge collision network) contains the information that most accurately characterizes the temperature-related operating state.

[0161] Specifically, the algorithm in this knowledge collision network may generate the final power component state encoding based on the temperature change trend obtained previously, the degree of proximity to the normal range, and other factors. This encoding is determined as the first quantitative embedded knowledge, which will be used to subsequently construct the part of the sensor monitoring attention vector related to the operating state vector. For example, this encoding may contain information such as the actual value of the temperature, the statistical characteristics of the temperature fluctuation, and the degree of deviation from the normal temperature range. This information will accurately reflect the operating status of the power component in terms of temperature.

[0162] With this design, the power supply component state code is generated by feature de-interference through the operation state knowledge collision module, and the state code of the last related knowledge collision network is determined as the first quantitative embedded knowledge, which can effectively remove interference information and accurately extract key information related to the operation state of the power supply component. In this way, it helps to improve the accuracy of the first quantitative embedded knowledge, and then lays the foundation for building a more accurate sensor monitoring attention vector, which helps to improve the accuracy and reliability of power supply component fault monitoring. For example, in voltage stability and temperature anomaly monitoring, this precise processing can better capture the details of the power supply component's operating status, timely discover potential fault hazards, and reduce system problems caused by power supply component failures.

[0163] In some exemplary embodiments, the use of the target lightweight remote monitoring algorithm to generate the second quantized embedded knowledge of the target power supply component under the configuration of the target fault monitoring guide tag includes: using the fault monitoring knowledge collision module in the u-th knowledge collision network to perform feature de-interference on the input embedded vector relationship network under the configuration of the target fault monitoring guide tag to generate the original second quantized embedded knowledge of the u-th knowledge collision network for the target power supply component; based on the original second quantized embedded knowledge, generating the fault monitoring code of the u-th knowledge collision network for the target power supply component; determining the fault monitoring code of the Y-th knowledge collision network of the target lightweight remote monitoring algorithm for the target power supply component as the second quantized embedded knowledge; wherein, the operation of performing feature de-interference on the input embedded vector relationship network under the configuration of the target fault monitoring guide tag is an operation of performing knowledge collision on the target fault monitoring guide tag and the input embedded vector relationship network.

[0164] In the target lightweight remote monitoring algorithm, when the algorithm is used to generate the second quantized embedded knowledge of the target power component under the configuration of the target fault monitoring guide label, the fault monitoring knowledge collision module in the u-th knowledge collision network is involved. First, the input embedded vector relationship network is feature de-interfered under the configuration of the target fault monitoring guide label. This operation is actually a knowledge collision between the target fault monitoring guide label and the input embedded vector relationship network.

[0165] Taking "voltage stability monitoring" as an example, the target fault monitoring guide tag has a vector composed of attributes such as voltage fluctuation range (setting the normal voltage range to 110-120V), allowed fluctuation frequency (setting a maximum of 5 fluctuations within 1 minute), and fluctuation duration limit (setting fluctuations lasting more than 10s to be considered abnormal). The input embedding vector relationship network may contain preliminary vector information related to voltage stability in the previous processing process, such as the previously estimated voltage fluctuation range vector [108, 122]. The fault monitoring knowledge collision module adopts a difference adjustment algorithm to calculate the difference between the normal voltage range in the target fault monitoring guide tag and the voltage range in the input embedding vector relationship network, and adjust each attribute according to the difference. For example, if the difference is large, the weight of the attribute in subsequent encoding is increased. Through this knowledge collision (feature de-interference), the original second quantized embedded knowledge of the u-th knowledge collision network for the target power component is generated, which contains the quantized information of the adjusted voltage stability-related attributes.

[0166] Furthermore, based on the original second quantized embedded knowledge, the fault monitoring code of the u-th knowledge collision network for the target power supply component is generated. Continuing with the example of "voltage stability monitoring", the original second quantized embedded knowledge contains quantized information of attributes such as the adjusted voltage fluctuation range, fluctuation frequency and fluctuation duration. The generation algorithm of the fault monitoring code can be an algorithm based on threshold comparison. For example, the voltage fluctuation range in the original second quantized embedded knowledge is compared with a preset more precise threshold. If the fluctuation range is close to the unstable threshold, it is marked as high risk in the fault monitoring code; for the fluctuation frequency, if it is close to or exceeds the maximum allowed value, it is marked as frequency abnormality risk in the code, etc. In this way, the generated fault monitoring code can more accurately reflect the fault monitoring status of the target power supply component related to the target fault monitoring guide label under the current knowledge collision network.

[0167] Then, the fault monitoring code of the target power component of the Yth knowledge collision network of the target lightweight remote monitoring algorithm is determined as the second quantized embedded knowledge. In the "voltage stability monitoring" scenario, after the previous multiple knowledge collision networks gradually process the voltage stability related information, the fault monitoring code output by the Yth knowledge collision network (set Y as the last relevant knowledge collision network) contains the information that can most accurately characterize the voltage stability fault monitoring status. This code contains information such as the degree of deviation between the final adjusted voltage fluctuation range and the standard range, the comparison result of the fluctuation frequency and the allowed frequency, and whether the fluctuation duration is close to the abnormal state. This information is determined as the second quantized embedded knowledge, which is used to subsequently construct the part of the sensor monitoring attention vector related to the fault monitoring sampling vector.

[0168] It can be seen that the process of generating original quantized knowledge through feature de-interference through the fault monitoring knowledge collision module, regenerating fault monitoring code, and finally determining the second quantized embedded knowledge can accurately extract key information related to power component fault monitoring based on the target fault monitoring guide label. This method can accurately process relevant information such as voltage stability monitoring, effectively remove interference, improve the accuracy of the second quantized embedded knowledge, provide a guarantee for constructing an accurate sensor monitoring attention vector, and thus improve the accuracy and reliability of power component fault monitoring.

[0169] In a replaceable embodiment, the fault monitoring knowledge collision module of the u-th knowledge collision network includes a collision activation function, and the collision activation function is used to indicate the result output by the fault monitoring knowledge collision module of the u-th knowledge collision network and the weight configuration when performing knowledge collision with the result output by the operation status knowledge collision module of the u-th knowledge collision network; generating the fault monitoring code of the u-th knowledge collision network for the target power supply component based on the original second quantized embedded knowledge includes: performing feature activation on the original second quantized embedded knowledge according to the collision activation function, and generating the fault monitoring code of the u-th knowledge collision network for the target power supply component.

[0170] There is a collision activation function in the fault monitoring knowledge collision module of the u-th knowledge collision network. This collision activation function plays a key role in the entire knowledge processing process. It is used to indicate the weight configuration of the output result of the fault monitoring knowledge collision module of the u-th knowledge collision network when the knowledge collision is carried out with the output result of the operating state knowledge collision module.

[0171] Taking "temperature anomaly monitoring" as an example, the original second quantized embedded knowledge output by the fault monitoring knowledge collision module is set to include the quantized representation of relevant information such as the normal temperature range information (such as 30-50°C) and the temperature rise rate limit (such as 2°C / min). The result output by the operation status knowledge collision module may include information such as the actual temperature value sequence (such as [32, 33, 34, ...]) and the statistical characteristics of temperature fluctuations. The collision activation function determines the weight of the output result of the fault monitoring knowledge collision module in the subsequent knowledge collision according to specific rules. For example, if the actual temperature value is close to the boundary of the normal range, the collision activation function may increase the proportion of the normal temperature range information in the weight configuration; if the temperature rise rate is close to the limit value, the weight of the temperature rise rate limit information will also be increased accordingly.

[0172] According to the collision activation function, the original second quantization embedded knowledge is feature activated to generate the fault monitoring code of the u-th knowledge collision network for the target power component. Continuing with "temperature anomaly monitoring" as an example, the collision activation function is set to determine the weight of each element in the original second quantization embedded knowledge according to the above rules. Then, the weighted summation algorithm is used to generate the fault monitoring code. For example, for the normal temperature range information, its weight is multiplied by a predefined basic value, and then the value calculated according to the temperature rise rate limit information and its weight is added, and finally a comprehensive fault monitoring code is obtained. This fault monitoring code can more accurately reflect the fault monitoring status of the target power component related to the target fault monitoring guide label under the current knowledge collision network. It comprehensively considers the importance of each element in the original second quantization embedded knowledge, and this importance is dynamically adjusted according to the actual situation (through the collision activation function).

[0173] In this way, by setting the collision activation function in the fault monitoring knowledge collision module, the weight of the output result of the fault monitoring knowledge collision module in the knowledge collision can be dynamically adjusted according to the actual situation. In scenarios such as temperature anomaly monitoring, this method can make the generated fault monitoring code more accurately reflect the fault monitoring status of the power supply component. This helps to improve the accuracy of the fault monitoring code, thereby improving the reliability of the entire power supply component monitoring system and the accuracy of fault monitoring.

[0174] In one possible example, the function label of the collision activation function is a device parameter activation label or a communication status activation label, the device parameter activation label has an outlier sub-label and a non-outlier sub-label, and the communication status activation label has a synchronously set feature enhancement factor; the feature activation of the original second quantized embedded knowledge according to the collision activation function to generate the fault monitoring code of the u-th knowledge collision network for the target power supply component includes: if the function label of the collision activation function is the device parameter activation label, and the device parameter activation label is an outlier sub-label, then the original second quantized embedded knowledge is determined as the fault monitoring code of the u-th knowledge collision network for the target power supply component; if the function label of the collision activation function is the device parameter activation label, and the device parameter activation label is in a non-outlier sub-label, then the fault monitoring code of the u-th knowledge collision network for the target power supply component is determined to be empty; if the function label of the collision activation function is the communication status activation label, the original second quantized embedded knowledge is knowledge enhanced according to the feature enhancement factor to generate the fault monitoring code of the u-th knowledge collision network for the target power supply component.

[0175] When the functional label of the collision activation function is the device parameter activation label, there are two cases: outlier sub-label and non-outlier sub-label.

[0176] Taking "voltage stability monitoring" as an example, the original second quantized embedded knowledge is set to include quantitative information such as voltage fluctuation range and fluctuation frequency. If the device parameter activation label is an outlier sub-label, for example, when monitoring the voltage, it is found that the voltage value deviates significantly from the normal range (such as the normal range is 110-120V, and a value of 130V appears), this obvious outlier indicates that the device may be in an abnormal state. At this time, the original second quantized embedded knowledge is directly determined as the fault monitoring code of the u-th knowledge collision network for the target power supply component. This is because the outlier device parameters themselves strongly suggest a fault state, and the original quantized knowledge is sufficient to reflect this abnormal situation.

[0177] On the contrary, if the device parameter activation label is in a non-outlier sub-label, for example, the voltage value has been fluctuating stably within the normal range of 110-120V, and the fluctuation frequency is also within the normal range, which indicates that the device is operating normally. In this case, it is determined that the fault monitoring code of the u-th knowledge collision network for the target power component is empty, because there is no abnormal situation that needs to be encoded.

[0178] When the function label of the collision activation function is a communication status activation label, the communication status activation label has a synchronously set feature enhancement factor. Taking "current overload monitoring" as an example, the original second quantization embedded knowledge is set to include information such as current overload threshold and current current value. If the communication status is good, for example, there is no communication delay and the data transmission is complete, the feature enhancement factor may be 1, indicating normal enhancement. At this time, the original second quantization embedded knowledge is enhanced according to the feature enhancement factor. For example, a simple multiplication algorithm is used to multiply each element in the original second quantization embedded knowledge by the feature enhancement factor 1 to obtain the enhanced knowledge, and then the u-th knowledge collision network is generated according to the enhanced knowledge. Fault monitoring code for the target power supply component. If the communication status is poor, such as there is a certain delay, the feature enhancement factor may be 0.8, indicating appropriate weakening, and the original second quantization embedded knowledge is processed according to the same multiplication algorithm to generate a fault monitoring code. This method of adjustment according to the communication status can make the fault monitoring code more in line with the actual equipment status monitoring requirements in the communication environment.

[0179] It can be seen that this embodiment classifies the generation of fault monitoring codes through device parameter activation tags and communication status activation tags. The outlier and non-outlier conditions of the device parameter activation tags can accurately determine the device status and generate appropriate fault monitoring codes, effectively distinguishing between obvious abnormalities or normal conditions. The feature enhancement factor of the communication status activation tag can adjust the original quantitative knowledge according to the communication status, so that the fault monitoring code is more in line with the device status in the actual communication environment. This improves the accuracy and adaptability of the fault monitoring code, thereby improving the reliability of power supply component monitoring.

[0180] Under some innovative design ideas, the first quantitative embedded knowledge includes W device state description levels under W description levels, and the second quantitative embedded knowledge includes W fault monitoring description levels under the W description levels, where W is a positive integer; the knowledge collision of the first quantitative embedded knowledge and the second quantitative embedded knowledge to generate the sensor monitoring attention vector corresponding to the target fault monitoring guidance label includes: performing description-level-based feature integration on the device state description level and the fault monitoring description level at the same description level, respectively, to generate the knowledge feature integration result of the target lightweight remote monitoring algorithm at each description level; performing first-order combination operations on the device state description level and the fault monitoring description level at the same description level, respectively, to generate the knowledge feature combination result of the target lightweight remote monitoring algorithm at each description level; performing global interaction on the knowledge feature integration results and knowledge feature combination results of the target lightweight remote monitoring algorithm at the W description levels, to generate the sensor monitoring attention vector corresponding to the target fault monitoring guidance label; wherein the knowledge collision includes the description-level-based feature integration, the first-order combination operation, and the global interaction.

[0181] The first quantitative embedded knowledge includes W device status description levels under W description levels, and the second quantitative embedded knowledge includes W fault monitoring description levels under these W description levels. Taking "voltage stability monitoring" as an example, set W=3, and these three description levels may be the voltage value level, the voltage fluctuation frequency level, and the voltage fluctuation duration level.

[0182] At the voltage value level, the device status description level contains a specific voltage value sequence (such as [112, 113, 111, ...]), and the fault monitoring description level contains information such as the normal voltage range (such as 110-120). A weighted average algorithm is used to perform feature integration based on the description level. A weight is assigned to each voltage value at the device status description level according to its relationship with the normal voltage range, and then the weighted average is calculated. A similar weighting process is also performed on the normal voltage range information at the fault monitoring description level to obtain the knowledge feature integration result at this description level. At the voltage fluctuation frequency level and fluctuation duration level, feature integration is performed according to a similar algorithm to generate the knowledge feature integration result of the target lightweight remote monitoring algorithm at each description level.

[0183] Next, the first-order combination operation is performed on the device state description level and the fault monitoring description level at the same description level. Taking "voltage stability monitoring" as an example, a linear combination algorithm is used at the voltage value level. The results of the device state description level after feature integration (such as the voltage value after weighted averaging) and the results of the fault monitoring description level (such as the adjusted normal voltage range information) are linearly combined in a certain ratio (such as 1:1) to generate the knowledge feature combination results at this description level. The same algorithm is also used at other description levels (such as the voltage fluctuation frequency level and the fluctuation duration level) to obtain the knowledge feature combination results of the target lightweight remote monitoring algorithm at each description level.

[0184] Finally, the knowledge feature integration results and knowledge feature combination results of the target lightweight remote monitoring algorithm at W description levels are globally interacted to generate the sensor monitoring attention vector corresponding to the target fault monitoring guidance label. Taking the previous three description levels as an example, a vector splicing algorithm is used to splice the knowledge feature integration results and knowledge feature combination results at each description level into a large vector. For example, the two results at the voltage value level are spliced ​​first, then the two results at the voltage fluctuation frequency level are spliced ​​at the back, and finally the two results at the voltage fluctuation duration level are spliced ​​at the end, thus generating a comprehensive vector, that is, the sensor monitoring attention vector corresponding to the target fault monitoring guidance label.

[0185] It can be seen that this embodiment generates a sensor monitoring attention vector by integrating features, combining operations, and globally interacting the first quantized embedded knowledge and the second quantized embedded knowledge at the description level. In specific examples such as voltage stability monitoring, this layered processing method can more finely integrate device status and fault monitoring information. It helps to improve the accuracy of the sensor monitoring attention vector, better reflect the actual status of the power supply component, and thus improve the accuracy and reliability of power supply component fault monitoring.

[0186] In the next step, the knowledge feature integration results and knowledge feature combination results of the target lightweight remote monitoring algorithm at the W description levels are globally interacted to generate a sensor monitoring attention vector corresponding to the target fault monitoring guidance label, including: performing a second-order combination operation on the knowledge feature integration results and knowledge feature combination results of the target lightweight remote monitoring algorithm at the same description level, respectively, to generate an intermediate knowledge vector of the target lightweight remote monitoring algorithm at each description level; performing attention processing on the intermediate knowledge vectors of the target lightweight remote monitoring algorithm at the W description levels, to generate a sensor monitoring attention vector corresponding to the target fault monitoring guidance label; wherein the global interaction includes the second-order combination operation and the attention processing.

[0187] In the target lightweight remote monitoring algorithm, when the knowledge feature integration results and knowledge feature combination results under W description levels are globally interacted, the second-order combination operation must be performed first. Taking "temperature anomaly monitoring" as an example, W=3 is set, and the three description levels are the temperature value level, the temperature rise rate level, and the temperature fluctuation amplitude level.

[0188] At the temperature numerical level, the knowledge feature integration result may be a sequence of temperature numerical values ​​after weighted averaging (such as [32, 33, 34]), and the knowledge feature combination result may be the result after comparison with the normal temperature range (such as the normal temperature range is 30-50, and the quantitative value of the relationship between the current temperature and the normal range). A nonlinear combination algorithm is used, such as multiplying the knowledge feature integration result by a coefficient (such as 0.6), multiplying the knowledge feature combination result by another coefficient (such as 0.4), and then adding them to obtain the intermediate knowledge vector at this description level. Similar nonlinear combination algorithms are also used at the temperature rise rate level and the temperature fluctuation amplitude level to generate the intermediate knowledge vector of the target lightweight remote monitoring algorithm at each description level.

[0189] Next, the intermediate knowledge vectors of the target lightweight remote monitoring algorithm at W description levels are processed with attention to generate the sensor monitoring attention vector corresponding to the target fault monitoring guidance label. Taking the previous temperature anomaly monitoring as an example, a threshold-based attention algorithm is adopted. Different thresholds are set. For example, for the temperature value level, if the value in the intermediate knowledge vector is close to the boundary of the normal temperature range, a higher attention weight is given; for the temperature rise rate level, if the rise rate in the intermediate knowledge vector is close to the abnormal threshold, a higher attention weight is also given. Then, according to these weights, the intermediate knowledge vectors are weighted and summed, and the results of each description level are spliced ​​together in a certain order to form a complete vector, that is, the sensor monitoring attention vector corresponding to the target fault monitoring guidance label.

[0190] In this way, the sensor monitoring attention vector is generated through the second-order combination operation and attention processing. In specific examples such as temperature anomaly monitoring, the second-order combination operation of the nonlinear combination algorithm can more accurately fuse different results. The threshold-based attention algorithm can highlight key information when generating vectors. This helps to improve the accuracy of the sensor monitoring attention vector, better reflect the status of power components, and improve the accuracy and reliability of fault monitoring.

[0191] In another possible step, the attention processing is performed on the intermediate knowledge vectors of the target lightweight remote monitoring algorithm at the W description levels to generate the sensor monitoring attention vector corresponding to the target fault monitoring guidance label, including: obtaining a spatiotemporal attention network; the spatiotemporal attention network includes W spatiotemporal attention operators; using the W spatiotemporal attention operators to perform spatiotemporal attention enhancement on the intermediate knowledge vectors of the target lightweight remote monitoring algorithm at the W description levels respectively to generate the spatiotemporal attention vectors of the target lightweight remote monitoring algorithm at the W description levels, and one spatiotemporal attention operator is used to perform spatiotemporal attention enhancement on the intermediate knowledge vectors of the target lightweight remote monitoring algorithm at one description level; using the spatiotemporal attention network to perform feature intersection on the spatiotemporal attention vectors of the target lightweight remote monitoring algorithm at the W description levels to generate the sensor monitoring attention vector corresponding to the target fault monitoring guidance label; wherein the attention processing includes the spatiotemporal attention enhancement and the feature intersection.

[0192] It can be understood that in the process of performing attention processing on the intermediate knowledge vectors of the target lightweight remote monitoring algorithm at W description levels to generate the sensor monitoring attention vector corresponding to the target fault monitoring guidance label, it is necessary to first obtain the spatiotemporal attention network. This spatiotemporal attention network contains W spatiotemporal attention operators.

[0193] Taking "voltage stability monitoring" as an example, W=3 is set. The three description levels are voltage value level, voltage fluctuation frequency level and voltage fluctuation duration level. For each intermediate knowledge vector at each description level, there is a corresponding spatiotemporal attention operator. These spatiotemporal attention operators are built based on specific mathematical models. For example, a spatiotemporal attention operator can be a structure based on a combination of a convolutional neural network (CNN) and a long short-term memory network (LSTM). The CNN part is used to extract the features of the intermediate knowledge vector in the spatial dimension, and the LSTM part is used to capture the features in the time series.

[0194] At the voltage numerical level, the intermediate knowledge vector may contain comprehensive information related to the voltage value obtained through previous operations, such as a numerical sequence after weighting, combination, and other operations. The spatiotemporal attention operator operates according to the characteristics of this intermediate knowledge vector. Specifically, the CNN part performs convolution operations on the values ​​in the vector, such as using a 3×1 convolution kernel with a step size of 1, and extracts the local relationship between the values ​​through convolution operations, which is similar to mining the distribution characteristics of voltage values ​​in space. The LSTM part processes the temporal order relationship of this numerical sequence. For example, if this numerical sequence is voltage values ​​collected in time order, LSTM can capture the trend of voltage values ​​changing over time. In this way, the spatiotemporal attention operator strengthens the intermediate knowledge vector of the target lightweight remote monitoring algorithm at this description level.

[0195] At the voltage fluctuation frequency level and fluctuation duration level, the corresponding spatiotemporal attention operators also operate according to similar principles. For the intermediate knowledge vector at the fluctuation frequency level, the CNN and LSTM parts will respectively operate on the numerical value of the fluctuation frequency and its time variation characteristics; the same is true for the fluctuation duration level, thereby generating the spatiotemporal attention vector of the target lightweight remote monitoring algorithm at these W description levels.

[0196] After obtaining the spatiotemporal attention vectors of the target lightweight remote monitoring algorithm at W description levels, the spatiotemporal attention network is used to perform feature crossover on these spatiotemporal attention vectors to generate the sensor monitoring attention vector corresponding to the target fault monitoring guidance label.

[0197] Continuing with "voltage stability monitoring" as an example, it is assumed that the spatiotemporal attention vectors at the voltage value level, voltage fluctuation frequency level, and voltage fluctuation duration level have been obtained. The feature cross-operation can adopt an algorithm based on tensor product (TensorProduct). First, the spatiotemporal attention vector at the voltage value level is regarded as a tensor, and the spatiotemporal attention vector at the voltage fluctuation frequency level is also regarded as a tensor, and the two tensors are subjected to a tensor product operation. This operation combines the elements in the two vectors according to specific rules, thereby mining the hidden feature relationship between the two description levels. For example, if there is some inherent correlation between the voltage value and the fluctuation frequency (such as the fluctuation frequency may have a specific change pattern when the voltage value is high), this relationship can be captured through the tensor product operation.

[0198] Then, the above results are tensor-producted with the spatiotemporal attention vector at the voltage fluctuation duration level to further integrate the feature relationships between the three description levels. In this process, the spatiotemporal attention network plays a role of coordination and guidance, ensuring that the feature crossover operation can accurately reflect the relationship between the various description levels. The final result is the sensor monitoring attention vector corresponding to the target fault monitoring guidance label. This vector integrates all the information of W description levels after spatiotemporal attention enhancement and feature crossover, and can more comprehensively and accurately reflect the state of the target power component under the target fault monitoring guidance label.

[0199] In other similar monitoring scenarios, such as "temperature anomaly monitoring", if the W description levels are the temperature value level, the temperature rise rate level, and the temperature fluctuation amplitude level, the above steps can also be followed. At the temperature value level, the spatiotemporal attention operator will strengthen the spatiotemporal attention according to the characteristics of the temperature value intermediate knowledge vector, and the same is true for the temperature rise rate and fluctuation amplitude levels. Then, the spatiotemporal attention network is used to perform feature crossover, and the information of these three levels is integrated to generate the sensor monitoring attention vector.

[0200] For example, in the "current overload monitoring" scenario, when the W description levels are set to the current value level, the current change rate level, and the overload duration level, the spatiotemporal attention operator will perform corresponding spatiotemporal attention enhancement operations on the intermediate knowledge vectors at each level, and then generate the final sensor monitoring attention vector through the feature crossover of the spatiotemporal attention network. This process helps to accurately capture the relationship between various current-related features in current overload monitoring and improve the accuracy of current overload judgment.

[0201] In this way, the spatiotemporal attention of the intermediate knowledge vector is strengthened through the spatiotemporal attention operator in the spatiotemporal attention network, and then the feature crossover is performed to generate the sensor monitoring attention vector. In specific monitoring examples such as voltage stability, temperature anomalies, and current overload, this method can fully explore the spatiotemporal relationship and hidden features between different description levels. Spatiotemporal attention strengthening uses structures such as CNN and LSTM to effectively capture the spatial and temporal characteristics of the vector, and feature crossover fuses multi-level information through the tensor product algorithm. This helps to improve the accuracy of the sensor monitoring attention vector, thereby improving the accuracy and reliability of power component fault monitoring.

[0202] In some independent embodiments, the sensor monitoring attention vectors corresponding to the Y fault monitoring guidance tags are respectively subjected to abnormality discrimination operations to generate abnormality discrimination results of the target power supply component under Y types of monitoring attention tasks, including: using a decision discrimination operator to perform a decision discrimination operation on each sensor monitoring attention vector to generate abnormality discrimination results of the target power supply component under the Y types of monitoring attention tasks; wherein, a sensor monitoring attention vector corresponding to one of the fault monitoring guidance tags is used to output the abnormality discrimination result of the target power supply component under one monitoring attention task, and the abnormality discrimination operation includes the decision discrimination operation.

[0203] In some other examples, the remote monitoring data is obtained by identifying the pending monitoring logs uploaded by the edge side sensor monitoring server, and the pending monitoring logs are also configured with the Y fault monitoring guidance tags. The edge side sensor monitoring server contains G fault monitoring guidance tags with screening authority, and the G fault monitoring guidance tags include multimodal fault monitoring guidance tags corresponding to the multimodal lightweight remote monitoring algorithm. The Y fault monitoring guidance tags are screened from the G fault monitoring guidance tags, G is a positive integer, and G is not less than Y; the method also includes: sending the abnormal judgment results of the target power supply component under the Y types of monitoring attention tasks to the edge side sensor monitoring server; wherein, the edge side sensor monitoring server is used to synchronize the output of the abnormal judgment results of the target power supply component under the Y types of monitoring attention tasks with the corresponding fault monitoring guidance tags.

[0204] It can be understood that when performing abnormality discrimination operations on the sensor monitoring attention vectors corresponding to Y fault monitoring guidance tags to generate abnormality discrimination results of the target power supply component under Y types of monitoring attention tasks, a decision discrimination operator is used.

[0205] Taking the sensor monitoring attention vector corresponding to "voltage stability monitoring" as an example, this vector is set to contain a comprehensive vector obtained through a series of previous processing of multiple information such as voltage value, voltage fluctuation range, fluctuation frequency, etc. The decision discriminant operator may be an algorithm structure based on thresholds and logical judgments. First, extract key elements related to voltage stability from the sensor monitoring attention vector, such as the actual voltage fluctuation value and the fluctuation frequency value. If under normal operation, the voltage fluctuation range is set to 110-120V, and the fluctuation frequency does not exceed 5 times within 1 minute. The decision discriminant operator will compare the actual voltage fluctuation value in the sensor monitoring attention vector with the range of 110-120V, and compare the fluctuation frequency value with the threshold of no more than 5 times in 1 minute.

[0206] If the actual voltage fluctuation exceeds the range of 110-120V or the fluctuation frequency exceeds 5 times, the decision discriminant operator will determine that the voltage stability is abnormal. This is the abnormal discrimination result of the target power component under the monitoring attention task of "voltage stability monitoring". For the sensor monitoring attention vectors corresponding to other fault monitoring guidance tags, such as "temperature abnormality monitoring", the sensor monitoring attention vector is set to contain information such as temperature value and temperature rise rate. The decision discriminant operator will perform similar comparisons and logical judgments according to the normal temperature range (for example, the normal temperature range of the power conversion module is 40-60°C, etc.) and the temperature rise rate limit (such as the temperature rise rate within 10 minutes does not exceed 2°C / min) and other standards, thereby obtaining the abnormal discrimination result of the target power component under the monitoring attention task of "temperature abnormality monitoring".

[0207] In "current overload monitoring", the sensor monitoring attention vector contains information such as current value and current change rate. The decision discrimination operator will make judgments based on the current overload threshold (such as setting the current over 10A as overload) and the normal range of the current change rate. If the current value exceeds 10A or the current change rate is abnormal, it is judged as a current overload abnormality, thereby completing the generation of abnormal discrimination results for the target power supply component under the monitoring attention task of "current overload monitoring".

[0208] In addition, the remote monitoring data is obtained by identifying the pending monitoring logs uploaded by the edge sensor monitoring server, and the pending monitoring logs are configured with Y fault monitoring guidance tags. The edge sensor monitoring server contains G fault monitoring guidance tags with screening authority, where the G fault monitoring guidance tags include multimodal fault monitoring guidance tags corresponding to the multimodal lightweight remote monitoring algorithm, and the Y fault monitoring guidance tags are screened from the G fault monitoring guidance tags (G is not less than Y).

[0209] For example, set G=10 fault monitoring guidance tags of the edge sensor monitoring server, which may include various fault monitoring guidance tags for different types of power components and different monitoring modes, such as voltage stability monitoring tags for different voltage levels, temperature anomaly monitoring tags for different devices, current overload monitoring tags under different load conditions, etc. From these 10 tags, select Y=8 fault monitoring guidance tags related to the current target power components and monitoring tasks, and these tags are configured to be uploaded to the monitoring log to be processed.

[0210] After the abnormality judgment results of the target power component under Y types of monitoring attention tasks are generated, these results are sent to the edge-side sensor monitoring server. The function of the edge-side sensor monitoring server is to synchronously output the abnormality judgment results of the target power component under Y types of monitoring attention tasks and the corresponding fault monitoring guidance tags. For example, when the abnormality judgment results (such as abnormality) of "voltage stability monitoring" and the corresponding fault monitoring guidance tags are received, these two information will be output at the same time, so that the operation and maintenance personnel can quickly and intuitively understand what kind of abnormal situation has occurred under which monitoring task, and the same is true for other monitoring attention tasks such as "temperature anomaly monitoring" and "current overload monitoring".

[0211] In this way, the abnormal discrimination operation is performed through the decision discrimination operator, and in specific monitoring examples such as voltage, temperature, and current, the abnormal discrimination results are accurately obtained based on thresholds and logical judgments. At the same time, the interaction mechanism between remote monitoring data and the edge-side sensor monitoring server, by screening the fault monitoring guide tags to upload and return the abnormal discrimination results for synchronous output, helps to improve the accuracy and efficiency of fault monitoring, facilitates operation and maintenance personnel to quickly locate and handle abnormal conditions of power components, and enhances the reliability of the entire power component monitoring system.

[0212] In other independent embodiments, the obtaining of the sensor monitoring signal record of the target power supply component includes: obtaining remote monitoring data of the target power supply component and a second original embedded vector relationship network to be optimized; using a feature de-interference algorithm to perform feature de-interference on the second original embedded vector relationship network under the configuration of the remote monitoring data to generate the sensor monitoring signal record.

[0213] Furthermore, the use of the feature de-interference algorithm to perform feature de-interference on the second original embedded vector relationship network under the configuration of the remote monitoring data to generate the sensor monitoring signal record includes: using the feature de-interference algorithm to perform feature de-interference on the second original embedded vector relationship network under the configuration of the remote monitoring data to generate an embedded vector prediction relationship network of the feature de-interference algorithm for the second original embedded vector relationship network; generating the sensor monitoring signal record through the second original embedded vector relationship network and the embedded vector prediction relationship network.

[0214] It can be understood that in the process of obtaining the sensor monitoring signal record of the target power component, it involves obtaining the remote monitoring data of the target power component and the second original embedded vector relationship network to be optimized. The remote monitoring data contains various information related to the operating status of the target power component, which may be data collected and transmitted by various sensors. The second original embedded vector relationship network to be optimized is an initial vector relationship structure, which may contain some predefined vector relationships related to the target power component.

[0215] For example, in a data center power component monitoring scenario, remote monitoring data may include voltage value sequences collected by voltage sensors, current value sequences collected by current sensors, and temperature value sequences collected by temperature sensors. These values ​​may be collected at certain time intervals, such as voltage values ​​collected every 1 second, current values ​​collected every 2 seconds, and temperature values ​​collected every 5 seconds. The second original embedded vector relationship network may contain some vectors related to the basic properties of the power component, such as vector relationships representing the rated voltage, rated current, and maximum allowable temperature of the power component.

[0216] The feature de-interference algorithm is used to perform feature de-interference on the second original embedded vector relationship network under the configuration of remote monitoring data to generate an embedded vector prediction relationship network of the feature de-interference algorithm for the second original embedded vector relationship network. The purpose of the feature de-interference algorithm is to remove factors that may interfere with the accurate characterization of the operating state of the target power component, thereby obtaining a more accurate vector relationship.

[0217] Taking voltage monitoring as an example, the voltage value sequence in the remote monitoring data is set to [110, 112, 111, 113, ...] (in volts), and the second original embedded vector relationship network contains a vector relationship with a rated voltage of 110V. The feature de-interference algorithm may adopt an algorithm based on deviation adjustment. For each voltage value, calculate its deviation from the rated voltage. For example, for the value of 112V, the deviation is 2V. Then adjust the vector relationship according to the size of the deviation and certain rules. If the deviation is within a certain range (such as ±5V), a linear adjustment algorithm may be used to adjust the weight in the vector relationship according to the size of the deviation. For example, when the deviation is 2V, adjust a certain weight in the vector relationship related to the rated voltage from the initial value 1 to 0.9. By performing such processing on all voltage values, an embedded vector prediction relationship network for the voltage part is obtained.

[0218] Similar operations are performed for current and temperature monitoring. In current monitoring, the current value sequence in the remote monitoring data is set to [5, 4.8, 5.2, 4.9, …] (in amperes), and the second original embedded vector relationship network contains a vector relationship with a rated current of 5A. According to the deviation between the current value and the rated current, a similar deviation adjustment-based algorithm is used. For example, for the value of 4.8A, the deviation is -0.2A. The weight in the vector relationship related to the rated current is adjusted according to the deviation, thereby obtaining the embedded vector prediction relationship network of the current part. In temperature monitoring, the temperature value sequence in the remote monitoring data is set to [30, 32, 31, 33, …] (in degrees Celsius), and the second original embedded vector relationship network contains a vector relationship with a maximum allowable temperature of 40°C. Similarly, according to the deviation between the temperature value and the maximum allowable temperature, the corresponding algorithm is used to adjust the vector relationship to obtain the embedded vector prediction relationship network of the temperature part. Combining the embedded vector prediction relationship networks of these different parts (voltage, current, temperature, etc.), the overall embedded vector prediction relationship network of the feature de-interference algorithm for the second original embedded vector relationship network is obtained.

[0219] Based on the above content, a sensor monitoring signal record is generated through the second original embedded vector relationship network and the embedded vector prediction relationship network. This step is to combine the vector relationship after feature denoising with the original vector relationship to obtain a sensor monitoring signal record that more comprehensively and accurately reflects the operating status of the target power component.

[0220] Specifically, for the voltage part, the rated voltage vector relationship in the original second original embedding vector relationship network is combined with the adjusted voltage vector relationship in the embedding vector prediction relationship network. For example, the original rated voltage vector relationship is [110V, weight 1], and the voltage vector relationship in the embedding vector prediction relationship network after feature denoising may be [110V, weight 0.9]. These two vector relationships are combined according to certain rules, such as weighted sum (110V*1+110V*0.9) / 2 to obtain a new voltage vector relationship, which will be included in the sensor monitoring signal record.

[0221] Similar operations are performed for the current and temperature parts. The original rated current vector relationship is combined with the adjusted current vector relationship, and the original maximum allowable temperature vector relationship is combined with the adjusted temperature vector relationship, and finally a comprehensive vector relationship containing voltage, current, temperature and other information is obtained. This comprehensive vector relationship is the sensor monitoring signal record. This sensor monitoring signal record can more accurately reflect the actual operating status of the target power supply component, providing a more reliable data basis for subsequent monitoring and analysis.

[0222] With this design, the original embedded vector relationship network is processed by the feature de-interference algorithm to generate the sensor monitoring signal record. In the specific voltage, current, and temperature monitoring examples, interference factors are removed and the vector relationship is adjusted based on algorithms such as deviation adjustment. This method can improve the accuracy of sensor monitoring signal records, combine the original relationship network with the predicted relationship network, and more comprehensively and accurately reflect the operating status of the power supply component, providing a more reliable data basis for subsequent power supply component monitoring, fault monitoring and other operations, and helping to improve the reliability and effectiveness of the entire power supply component monitoring system.

[0223] Further, Figure 2 This is a schematic diagram of the structure of a remote automated monitoring system 200 provided by the present invention. Figure 2 The remote automated monitoring system 200 shown includes a processor 210, which can call and run a computer program from a memory to implement the method of the present invention.

[0224] Alternatively, if Figure 2 As shown, the remote automated monitoring system 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method of the present invention.

[0225] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0226] Alternatively, if Figure 2 As shown, the remote automation monitoring system 200 may further include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices, specifically, to send information or data to other devices, or to receive information or data sent by other devices.

[0227] Optionally, the remote automated monitoring system 200 may implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as a processing module) or a device deployed with a storage engine in each method of the present invention, which will not be described in detail here for the sake of brevity.

[0228] It should be understood that the processor of the present invention may be an integrated circuit chip having signal processing capabilities.

[0229] It is understood that the memory in the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memory. It should be noted that the memory of the systems and methods described herein is intended to include but is not limited to suitable types of memory.

[0230] Based on the above, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above method are implemented.

[0231] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises one..." does not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the method and device in the embodiment of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0232] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention application in essence or the part that contributes to the prior art can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), including a number of instructions for a terminal (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention application.

[0233] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the inspiration of the present invention, ordinary technicians in this field can also make many forms without departing from the purpose of the present invention and the scope of protection of the present invention, all of which are within the protection of the present invention.

Claims

1. A power component data monitoring method based on remote automation, characterized in that: The method is implemented by a remote automated monitoring system, and the method comprises: Acquire a sensor monitoring signal record of a target power supply component, wherein the sensor monitoring signal record is used to characterize an operating state description of the target power supply component; Obtaining X lightweight remote monitoring algorithms and Y fault monitoring guidance tags corresponding to the X lightweight remote monitoring algorithms, wherein one fault monitoring guidance tag is used to match an attention monitoring attribute of a monitoring attention task, and one lightweight remote monitoring algorithm is associated with one or more fault monitoring guidance tags, wherein X and Y are both positive integers, and X is not greater than Y; Using the X lightweight remote monitoring algorithms, under the configuration of the sensor monitoring signal record and the associated fault monitoring guide tag, a sensor monitoring attention vector corresponding to each fault monitoring guide tag is generated, wherein one sensor monitoring attention vector includes an operation state vector of the target power component and a fault monitoring sampling vector of the monitoring attention task indicated by the corresponding fault monitoring guide tag, wherein the operation state vector is determined by knowledge embedding the operation state description represented by the sensor monitoring signal record, and the fault monitoring sampling vector is determined by knowledge embedding the attention monitoring attribute of the monitoring attention task indicated by the corresponding fault monitoring guide tag; Performing abnormality discrimination operations on the sensor monitoring attention vectors corresponding to the Y fault monitoring guidance tags respectively, and generating abnormality discrimination results of the target power supply component under Y monitoring attention tasks; Any one of the X lightweight remote monitoring algorithms is a target lightweight remote monitoring algorithm, any fault monitoring guidance tag corresponding to the target lightweight remote monitoring algorithm is a target fault monitoring guidance tag, and the use of the X lightweight remote monitoring algorithms under the configuration of the sensor monitoring signal record and the associated fault monitoring guidance tag to generate a sensor monitoring attention vector corresponding to each fault monitoring guidance tag includes: Using the target lightweight remote monitoring algorithm, under the configuration of the sensor monitoring signal record, generate first quantized embedded knowledge of the target power supply component, wherein the first quantized embedded knowledge is used to characterize the operating state vector of the target power supply component; Using the target lightweight remote monitoring algorithm under the configuration of the target fault monitoring guide tag, generating second quantized embedded knowledge of the target power supply component, wherein the second quantized embedded knowledge is used to characterize the fault monitoring sampling vector of the monitoring attention task indicated by the target fault monitoring guide tag; A knowledge collision is performed on the first quantized embedded knowledge and the second quantized embedded knowledge to generate a sensor monitoring attention vector corresponding to the target fault monitoring guidance label.

2. The method according to claim 1, characterized in that The target lightweight remote monitoring algorithm comprises Z knowledge collision networks in a cascade state for implementing knowledge vector embedding, any knowledge collision network comprises an operation state knowledge collision module and a fault monitoring knowledge collision module in parallel, and Z is a positive integer; The target lightweight remote monitoring algorithm includes a first original embedding vector relationship network to be optimized, the Z knowledge collision networks include a u-th knowledge collision network, u is a positive integer, and u is not greater than Z, and the u-th knowledge collision network has an input embedding vector relationship network; If u is equal to 1, then the input embedded vector relationship network is the first original embedded vector relationship network. If u is greater than 1, then the input embedded vector relationship network is the output result of the u-1th knowledge collision network of the target lightweight remote monitoring algorithm, and the output result is obtained after knowledge collision between the output result of the operating status knowledge collision module in the u-1th knowledge collision network and the output result of the fault monitoring knowledge collision module.

3. The method according to claim 2, characterized in that The method of using the target lightweight remote monitoring algorithm to generate first quantized embedded knowledge of the target power supply component under the configuration of the sensor monitoring signal record includes: Using the operation state knowledge collision module in the u-th knowledge collision network, under the configuration of the sensor monitoring signal record, the input embedding vector relationship network is subjected to feature de-interference to generate the power supply component state encoding of the u-th knowledge collision network for the target power supply component; Encoding the power supply component state of the target power supply component by the Yth knowledge collision network of the target lightweight remote monitoring algorithm is determined as the first quantized embedded knowledge; Among them, the operation of performing feature de-interference on the input embedded vector relationship network under the configuration of the sensor monitoring signal record is an operation of performing knowledge collision on the sensor monitoring signal record and the input embedded vector relationship network.

4. The method according to claim 2, characterized in that The method of using the target lightweight remote monitoring algorithm to generate second quantized embedded knowledge of the target power supply component under the configuration of the target fault monitoring guide tag includes: Using the fault monitoring knowledge collision module in the u-th knowledge collision network, under the configuration of the target fault monitoring guide label, the input embedding vector relationship network is subjected to feature de-interference to generate the original second quantized embedding knowledge of the u-th knowledge collision network for the target power supply component; Generating a fault monitoring code of the u-th knowledge collision network for the target power supply component according to the original second quantized embedded knowledge; Determine the Yth knowledge collision network of the target lightweight remote monitoring algorithm for the fault monitoring code of the target power supply component as the second quantized embedded knowledge; Among them, the operation of performing feature de-interference on the input embedded vector relationship network under the configuration of the target fault monitoring guidance label is an operation of performing knowledge collision on the target fault monitoring guidance label and the input embedded vector relationship network.

5. The method according to claim 4, characterized in that The fault monitoring knowledge collision module of the u-th knowledge collision network includes a collision activation function, and the collision activation function is used to indicate the weight configuration of the result output by the fault monitoring knowledge collision module of the u-th knowledge collision network when performing knowledge collision with the result output by the operation status knowledge collision module of the u-th knowledge collision network; The generating, based on the original second quantized embedded knowledge, the fault monitoring code of the u-th knowledge collision network for the target power supply component comprises: performing feature activation on the original second quantized embedded knowledge according to the collision activation function to generate the fault monitoring code of the u-th knowledge collision network for the target power supply component; The functional label of the collision activation function is a device parameter activation label or a communication state activation label, the device parameter activation label has an outlier sub-label and a non-outlier sub-label, and the communication state activation label has a synchronously set feature enhancement factor; The step of performing feature activation on the original second quantized embedded knowledge according to the collision activation function to generate the fault monitoring code of the u-th knowledge collision network for the target power supply component includes: If the function label of the collision activation function is the device parameter activation label, and the device parameter activation label is an outlier label, the original second quantization embedded knowledge is determined as the fault monitoring code of the u-th knowledge collision network for the target power supply component; If the function label of the collision activation function is the device parameter activation label, and the device parameter activation label is in a non-outlier sub-label, determining that the fault monitoring code of the u-th knowledge collision network for the target power supply component is empty; If the functional label of the collision activation function is the communication state activation label, the original second quantized embedded knowledge is enhanced according to the feature enhancement factor to generate the fault monitoring code of the u-th knowledge collision network for the target power supply component.

6. The method according to claim 1, characterized in that The first quantized embedded knowledge includes W device state description levels under W description levels, and the second quantized embedded knowledge includes W fault monitoring description levels under the W description levels, where W is a positive integer; performing knowledge collision on the first quantized embedded knowledge and the second quantized embedded knowledge to generate a sensor monitoring attention vector corresponding to the target fault monitoring guide label includes: The device state description level and the fault monitoring description level at the same description level are respectively integrated based on the description level to generate the knowledge feature integration result of the target lightweight remote monitoring algorithm at each description level; Performing first-order combination operations on the device state description level and the fault monitoring description level at the same description level, respectively, to generate knowledge feature combination results of the target lightweight remote monitoring algorithm at each description level; Performing global interaction on the knowledge feature integration results and knowledge feature combination results of the target lightweight remote monitoring algorithm at the W description levels to generate a sensor monitoring attention vector corresponding to the target fault monitoring guidance label; The knowledge collision includes the feature integration based on the description level, the first-order combination operation and global interaction.

7. The method according to claim 6, characterized in that The globally interacting the knowledge feature integration results and knowledge feature combination results of the target lightweight remote monitoring algorithm at the W description levels to generate the sensor monitoring attention vector corresponding to the target fault monitoring guidance label includes: Performing a second-order combination operation on the knowledge feature integration results and knowledge feature combination results of the target lightweight remote monitoring algorithm at the same description level, respectively, to generate an intermediate knowledge vector of the target lightweight remote monitoring algorithm at each description level; Attention processing is performed on the intermediate knowledge vectors of the target lightweight remote monitoring algorithm at the W description levels to generate a sensor monitoring attention vector corresponding to the target fault monitoring guidance label; wherein the global interaction includes the second-order combination operation and the attention processing.

8. The method according to claim 6, characterized in that The step of performing attention processing on the intermediate knowledge vectors of the target lightweight remote monitoring algorithm at the W description levels to generate a sensor monitoring attention vector corresponding to the target fault monitoring guidance label includes: Obtaining a spatiotemporal attention network; the spatiotemporal attention network includes W spatiotemporal attention operators; The W spatiotemporal attention operators are used to perform spatiotemporal attention enhancement on the intermediate knowledge vectors of the target lightweight remote monitoring algorithm at the W description levels, respectively, to generate the spatiotemporal attention vectors of the target lightweight remote monitoring algorithm at the W description levels, and one spatiotemporal attention operator is used to perform spatiotemporal attention enhancement on the intermediate knowledge vectors of the target lightweight remote monitoring algorithm at one description level; The spatiotemporal attention network is used to perform feature crossover on the spatiotemporal attention vectors of the target lightweight remote monitoring algorithm at the W description levels to generate a sensor monitoring attention vector corresponding to the target fault monitoring guidance label; wherein the attention processing includes the spatiotemporal attention enhancement and the feature crossover.

9. A remote automated monitoring system, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 8.

Images