A power supply diagnosis method, device and storage medium based on big data

Abstract

The present application provides a power supply diagnosis method, device and storage medium based on big data, in which the prior information is label information characterizing the performance of the target power supply, and the node describing the performance in the label information is located after the past node of the power supply monitoring data corresponding to the prior information. The power supply monitoring data in the debugging sample are respectively integrated with the power supply performance characterization vector of the corresponding prior information into integrated power supply monitoring data, so that the integrated power supply monitoring data in the integrated power supply monitoring data stream can include the information of the prior information. Because the prior information is close to the actual situation, the embedded mapping neural network is debugged based on the integrated power supply monitoring data stream, so that the embedded mapping neural network can learn the real information. When the embedded mapping neural network is used for reasoning data streams, the accuracy and reliability of power supply diagnosis can be increased.

Description

Technical Field

[0001] The present application relates to but is not limited to the field of data processing technology, and in particular to a power supply diagnosis method, device and storage medium based on big data. Background Art

[0002] In the field of power management and maintenance, accurate and timely diagnosis of power supply equipment is crucial. With the continuous development of big data technology, the use of big data for power supply diagnosis has become a trend. Traditional power supply diagnosis methods are usually based on limited monitoring data and empirical knowledge, and it is difficult to comprehensively and accurately evaluate the performance and status of the power supply. Although existing power supply diagnosis methods can collect and process a large amount of power supply monitoring data, they often face challenges such as high data dimension, information redundancy, and strong correlation when using these data for diagnosis. How to effectively extract key features from power supply monitoring data and combine them with power supply performance label information to improve the accuracy of diagnosis is the focus of current research.

[0003] In addition, existing power diagnosis methods often ignore the timing relationship and dependency between data when processing time series data. Power monitoring data is usually a series of data points arranged in chronological order, and the timing relationship and change trend between them are crucial for accurate diagnosis of power performance. Therefore, considering the timing characteristics of data in power diagnosis methods is of great significance to improve diagnostic accuracy.

[0004] In summary, the existing power supply diagnosis methods based on big data still have deficiencies in data processing, feature extraction, and timing relationship consideration. In order to improve the accuracy and reliability of power supply diagnosis, it is necessary to develop a new power supply diagnosis method to improve the accuracy of power supply diagnosis. Summary of the invention

[0005] In view of this, the embodiments of the present application at least provide a power supply diagnosis method, device and storage medium based on big data.

[0006] The technical solution of the embodiment of the present application is implemented as follows:

[0007] In a first aspect, an embodiment of the present application provides a power diagnosis method based on big data, the method comprising: obtaining a past power monitoring data stream of a target power supply; the past power monitoring data stream comprises power monitoring data of each past node of the target power supply in a plurality of past nodes, and each of the power monitoring data is arranged according to the node order of the respective past node; obtaining a priori information corresponding to each power monitoring data in the past power monitoring data stream, the priori information being label information characterizing the performance of the target power supply, and the node describing the performance in the label information being located after the past node of the power monitoring data corresponding to the priori information; mining a power performance characterization vector of each of the priori information to obtain a power performance characterization vector of each of the prior information; integrating each power monitoring data in the past power monitoring data stream with the corresponding power performance characterization vector of the prior information into integrated power monitoring data to obtain a past integrated power monitoring data stream; based on the past integrated power monitoring data stream, inferring the power monitoring data of each monitoring node of the target power supply in a plurality of uninterrupted monitoring nodes to obtain an inferred power monitoring data stream; the monitoring node is located after each of the past nodes.

[0008] In some embodiments, each power monitoring data in the past power monitoring data stream is respectively integrated with the power performance characterization vector of the corresponding prior information into integrated power monitoring data to obtain the past integrated power monitoring data stream, including: for each power monitoring data in the past power monitoring data stream, the power monitoring data is combined with the power performance characterization vector of the corresponding prior information to obtain the integrated power monitoring data corresponding to each of the power monitoring data; according to the arrangement position of the power monitoring data in the past power monitoring data stream, the integrated power monitoring data corresponding to each of the power monitoring data is arranged to generate the past integrated power monitoring data stream.

[0009] In some embodiments, the method further includes: arranging the power performance characterization vectors of the prior information corresponding to each power monitoring data according to the arrangement position of the power monitoring data in the past power monitoring data stream to generate a power performance characterization vector stream; for each power monitoring data in the past power monitoring data stream, combining the power monitoring data with the corresponding power performance characterization vector of the prior information to obtain integrated power monitoring data corresponding to each power monitoring data, including: combining the power monitoring data and the power performance characterization vector at the same position in the past power monitoring data stream and the power performance characterization vector stream to obtain integrated power monitoring data corresponding to each power monitoring data.

[0010] In some embodiments, the power monitoring data of each monitoring node of the target power supply in a plurality of uninterrupted monitoring nodes is inferred based on the past integrated power monitoring data stream to obtain the inferred power monitoring data stream, including: according to the power performance characterization vector included in each integrated power monitoring data in the past integrated power monitoring data stream, the past integrated power monitoring data stream is data segmented to obtain a plurality of data segments; wherein the power performance characterization vectors included in the integrated power monitoring data in the same data segment are the same, and the power performance characterization vectors included in the integrated power monitoring data in the two data segments before and after are different; based on the multiple data segments, the power monitoring data of each monitoring node of the target power supply in a plurality of uninterrupted monitoring nodes is inferred to obtain the inferred power monitoring data stream.

[0011] In some embodiments, based on the multiple data segments, the power monitoring data of each monitoring node of the target power supply in a plurality of uninterrupted monitoring nodes is inferred to obtain an inferred power monitoring data stream, including: embedding and mapping the multiple data segments respectively to obtain an embedded mapping vector corresponding to each of the data segments; based on the embedded mapping vector corresponding to each of the data segments, generating an embedded mapping vector stream corresponding to the past power monitoring data stream; based on the embedded mapping vector stream corresponding to the past power monitoring data stream, inferring the power monitoring data of each monitoring node of the target power supply in a plurality of uninterrupted monitoring nodes to obtain an inferred power monitoring data stream.

[0012] In some embodiments, the embedding mapping vector is generated by an embedding mapping neural network that has been debugged. The method also includes a debugging process of the embedding mapping neural network, and the debugging process of the embedding mapping neural network includes: obtaining an active debugging sample set, the active debugging sample set includes debugging samples corresponding to the target power supply in multiple past node sequences with intersections, the debugging samples corresponding to the past node sequences include power monitoring data of each past node of the target power supply in the past node sequences, and the power monitoring data in the debugging samples are arranged according to the order of each node; obtaining a passive debugging sample set, the passive debugging sample set includes debugging samples corresponding to the target power supply in multiple past node sequences that do not have intersections; for each debugging sample in the active debugging sample set, each power monitoring data in the debugging sample is respectively integrated with the power performance characterization vector of the corresponding prior information into integrated power monitoring data, and an integrated power monitoring data stream corresponding to each debugging sample in the active debugging sample set is generated; for each debugging sample in the passive debugging sample set, each power monitoring data in the debugging sample is respectively integrated with the power performance characterization vector of the corresponding prior information into Integrate power monitoring data to generate an integrated power monitoring data stream corresponding to each debugging sample in the passive debugging sample set; load the integrated power monitoring data stream corresponding to each debugging sample in the active debugging sample set into the embedding mapping neural network for embedding mapping to obtain multiple embedding mapping vectors, and obtain the embedded mapping vector stream corresponding to each debugging sample in the active debugging sample set; load the integrated power monitoring data stream corresponding to each debugging sample in the passive debugging sample set into the embedding mapping neural network for embedding mapping to obtain multiple embedding mapping vectors, and obtain the embedded mapping vector stream corresponding to each debugging sample in the passive debugging sample set; determine the commonality measurement results between the embedded mapping vector streams corresponding to each debugging sample in the active debugging sample set, and obtain the commonality measurement results corresponding to the active debugging sample set; determine the commonality measurement results between the embedded mapping vector streams corresponding to each debugging sample in the passive debugging sample set, and obtain the commonality measurement results corresponding to the passive debugging sample set; debug the embedding mapping neural network based on the commonality measurement results corresponding to the active debugging sample set and the commonality measurement results corresponding to the passive debugging sample set.

[0013] In some embodiments, the debugging of the embedding mapping neural network based on the commonality measurement results corresponding to the active debugging sample set and the commonality measurement results corresponding to the passive debugging sample set includes: performing a first numerical calculation on the commonality measurement results corresponding to multiple active debugging sample sets respectively to obtain a first commonality measurement result; performing the first numerical calculation on the commonality measurement results corresponding to multiple passive debugging sample sets respectively to obtain a second commonality measurement result; debugging the embedding mapping neural network based on an error between the first commonality measurement result and the second commonality measurement result; or, debugging the embedding mapping neural network based on the commonality measurement results corresponding to the active debugging sample set and the commonality measurement results corresponding to the passive debugging sample set includes: analyzing the statistical dispersion of the commonality measurement results corresponding to multiple active debugging sample sets respectively to obtain a first analysis result; analyzing the statistical dispersion of the commonality measurement results corresponding to multiple passive debugging sample sets respectively to obtain a second analysis result; and debugging the embedding mapping neural network based on the first analysis result and the second analysis result.

[0014] In some embodiments, the first analysis result is the variance or standard deviation of the commonality measurement results corresponding to multiple sets of active debugging samples; wherein the second analysis result is the variance or standard deviation of the commonality measurement results corresponding to multiple sets of passive debugging samples; debugging the embedding mapping neural network based on the first analysis result and the second analysis result includes: obtaining a first network cost of the embedding mapping neural network based on the sum of the first analysis result and the second analysis result, the first network cost is consistent with the change trend of the first analysis result, and the first network cost is consistent with the change trend of the second analysis result; adjusting the network configuration variables of the embedding mapping neural network based on the first network cost; or,

[0015] The debugging of the embedded mapping neural network based on the first analysis result and the second analysis result includes: calculating the mean of the commonality measurement results corresponding to the multiple active debugging sample sets to obtain a first commonality measurement result mean; calculating the mean of the commonality measurement results corresponding to the multiple passive debugging sample sets to obtain a second commonality measurement result mean; subtracting the first commonality measurement result mean from the second commonality measurement result mean to obtain a commonality measurement result mean difference; subtracting the commonality measurement result mean difference from a preset commonality measurement result threshold to obtain a target value, and weightedly summing the target value, the first analysis result and the second analysis result to obtain a second network cost; and adjusting the network configuration variables of the embedded mapping neural network based on the second network cost.

[0016] In some embodiments, the inferred power monitoring data stream is obtained by using a debugged data stream inference neural network, and the method further includes a data stream inference neural network debugging step, which includes: obtaining a power monitoring data stream debugging sample; the power monitoring data stream debugging sample includes power monitoring data of each past node of the target power supply in the first past node sequence, and each power monitoring data in the power monitoring data stream debugging sample is arranged according to the node order of each past node; each power monitoring data in the power monitoring data stream debugging sample is respectively integrated with the power performance characterization vector of the corresponding prior information into integrated power monitoring data to generate a debugging sample integrated power monitoring data stream; embedding and mapping the debugging sample integrated power monitoring data stream based on the debugged embedding mapping neural network to generate a debugging sample embedded mapping vector stream; loading the debugging sample embedded mapping vector stream into the data stream inference neural network to be debugged, inferring the power monitoring data of each past node of the target power supply in the second past node sequence, and generating an inference data stream; debugging the data stream inference neural network based on the inference data stream and the power monitoring data stream label.

[0017] In a second aspect, the present application provides a power supply diagnostic device based on big data, including: a monitoring data acquisition module, used to acquire a past power supply monitoring data stream of a target power supply; the past power supply monitoring data stream includes power supply monitoring data of each past node of the target power supply in a plurality of past nodes, and each of the power supply monitoring data is arranged according to the node order of the respective past node; a priori information acquisition module, used to acquire prior information corresponding to each power supply monitoring data in the past power supply monitoring data stream, the priori information being label information characterizing the performance of the target power supply, and the node describing the performance in the label information being located in the past of the power supply monitoring data corresponding to the priori information. After the node; a performance feature mining module, used to mine the power performance characterization vector of each of the prior information, and obtain the power performance characterization vector of each of the prior information; a performance feature integration module, used to integrate each power monitoring data in the past power monitoring data stream with the corresponding power performance characterization vector of the prior information into integrated power monitoring data, and obtain the past integrated power monitoring data stream; a power data inference module, used to infer the power monitoring data of each monitoring node of the target power in multiple uninterrupted monitoring nodes based on the past integrated power monitoring data stream, and obtain the inferred power monitoring data stream; the monitoring node is located after each of the past nodes.

[0018] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and the computer program is used by a processor to execute the steps in the above-described method.

[0019] This application has at least the following beneficial effects:

[0020] The power diagnosis method, device and storage medium based on big data provided by the embodiments of the present application, because the prior information is label information characterizing the performance of the target power supply, and the node describing the performance in the label information is located after the past node of the power monitoring data corresponding to the prior information, the power monitoring data in the debugging sample are respectively integrated with the power performance characterization vector of the corresponding prior information into integrated power monitoring data, so that the integrated power monitoring data in the integrated power monitoring data stream can include information of the prior information. Because the prior information is close to the actual situation, debugging the embedded mapping neural network based on the integrated power monitoring data stream can allow the embedded mapping neural network to learn the real information. When the embedded mapping neural network is used to infer the data stream, the accuracy and reliability of the power diagnosis can be increased.

[0021] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and are used together with the specification to illustrate the technical solution of the present application.

[0023] Figure 1 A schematic diagram of the implementation flow of a power supply diagnosis method based on big data provided in an embodiment of the present application.

[0024] Figure 2 A schematic diagram of the composition structure of a power supply diagnostic device based on big data provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions of the present application are further elaborated in detail below in conjunction with the drawings and embodiments. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0026] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. The terms "first / second / third" involved are merely used to distinguish similar objects and do not represent a specific order for the objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing this application and are not intended to limit this application.

[0028] The present application embodiment provides a power supply diagnosis method based on big data, which can be executed by a processor of a computer device. The computer device may refer to a server, a laptop, a tablet computer, a desktop computer, a smart TV, a mobile device (such as a mobile phone, a portable video player, a personal digital assistant, a dedicated messaging device, a portable gaming device) and other devices with data processing capabilities.

[0029] Figure 1 A schematic diagram of the implementation flow of a power supply diagnosis method based on big data provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method comprises the following steps:

[0030] Step S10: Acquire a past power monitoring data stream of a target power source; the past power monitoring data stream includes power monitoring data of each past node of the target power source in a plurality of past nodes, and each of the power monitoring data is arranged according to the node sequence of the respective past node.

[0031] In the embodiment of the present application, the target power supply is the power supply device that needs to be diagnosed. For example, if a power supply (PSU) installed on a specific server in a data center needs to analyze the performance and possible problems of the PSU, then this PSU is the "target power supply". The past power supply monitoring data stream is a collection of historical data that records the monitoring data of the target power supply in the past period of time. These data are arranged in chronological order to form a continuous data stream. For example, a monitoring log is set, which records the data obtained each time the target power supply is monitored in chronological order, and these data include voltage readings, current readings, temperature, etc. The monitoring log is equivalent to the past power supply monitoring data stream. The past node refers to the time point at which the power supply monitoring data is recorded in the past power supply monitoring data stream. Each past node corresponds to the state of the target power supply at a specific moment. The power supply monitoring data is the data obtained when the target power supply is measured at each past node. These data reflect the performance and state of the power supply at the corresponding moment. For example, in a power supply monitoring, the data monitored are: voltage 220V, current 5A, power factor 0.9. These data are power supply monitoring data, which describe the working state of the power supply at the corresponding moment. It can be understood that the above examples are only examples. In actual situations, power monitoring data can also include data such as temperature, frequency, waveform distortion, efficiency, capacity, etc., which are not specifically limited. In step S10, the node order refers to the order in which past nodes are arranged on the time axis. In the past power monitoring data stream, each power monitoring data is arranged according to the time sequence of the past node in which it is located. For example, on a time axis, multiple past nodes are marked, and each node has a corresponding timestamp, such as "10:00", "10:15", "10:30", etc. These nodes are arranged in the order in which they appear on the time axis to form a node order.

[0032] Step S20: obtaining a priori information corresponding to each power monitoring data in the past power monitoring data stream, wherein the a priori information is label information characterizing the target power performance, and the node describing the performance in the label information is located after the past node of the power monitoring data corresponding to the a priori information.

[0033] Specifically, prior information refers to information about power supply performance that is known before the actual monitoring of the target power supply or obtained through other means. This information usually exists in the form of tags and is used to describe the performance status of the power supply at a certain moment or over a period of time. Prior information may be obtained through historical records, expert evaluations, performance data of similar devices, etc., and is used to provide an important reference for subsequent power supply performance analysis and diagnosis. Prior information refers to information that is known after obtaining power supply monitoring data but before analyzing these data. This information describes the performance status of the target power supply at a node after the past node of the monitoring data. Prior information includes two key parts: node information and power supply performance information. In time series data, a node refers to the time point of data recording. In the power supply monitoring data stream, each node corresponds to a record of the power supply status. Power supply performance refers to the working status of the power supply at a specific node, which may include the actual values ​​or status descriptions (such as "normal", "overheated", etc.) of various performance indicators such as voltage, current, power factor, and temperature.

[0034] For example, when diagnosing the server power supply of a data center, the monitoring data such as voltage and current at each time node are recorded in the past power monitoring data stream. At the same time, it is also known whether the power supply has failed after these time nodes, as well as information such as the type and severity of the failure. This information is the prior information to be obtained in step S20, which describes the performance status of the power supply after a certain moment. The node describing the performance in the label information is located after the past node of the power monitoring data corresponding to the prior information, which means that the prior information is usually obtained after the actual power monitoring data. This is because the prior information often requires observation and analysis of the actual operation of the power supply before it can be obtained. For example, in the above example, it is necessary to collect the monitoring data of the power supply first, and then evaluate its performance status based on these data and the actual performance of the power supply, and obtain the corresponding prior information to achieve the purpose of prediction.

[0035] Assume a data center, where the server power supply is the target power supply that needs to be diagnosed. The computer device will perform the following operations according to step S20: The computer device first obtains the past power supply monitoring data stream from the monitoring system of the data center. These data streams contain the monitoring data of each node of the target power supply in the past period of time, such as the voltage and current readings every hour. For the monitoring data of each node, the computer device will search for the corresponding prior information. This information may be stored in another database, or generated based on historical data through an algorithm. The prior information includes a description of the power supply performance of the target power supply at a node after each past node of the monitoring data. For example, assuming that the monitoring data of the power supply is obtained at time node T1, the prior information may indicate that at time node T2 (T2>T1), the performance status of the power supply is "normal", and the voltage and current readings are within the normal range. The computer device associates this prior information with the corresponding monitoring data to form a more complete data set for subsequent power supply performance analysis and diagnosis.

[0036] By acquiring and integrating the prior information corresponding to each power supply monitoring data in the past power supply monitoring data stream, the computer equipment can more accurately understand the performance history and possible future status of the target power supply. This information is of great significance for timely discovering potential power supply problems, preventing power supply failures, and optimizing power supply management strategies. In addition, by combining prior information with monitoring data, a more comprehensive and reliable data foundation is provided for subsequent power supply performance analysis and diagnosis.

[0037] Step S30: mining each power performance characterization vector of the prior information to obtain each power performance characterization vector of the prior information.

[0038] In an embodiment of the present application, a power supply performance characterization vector refers to a set of values ​​or features that can represent or describe the power supply performance state. This set of values ​​or features is usually extracted from the monitoring data of the power supply, and may include actual values ​​or calculated values ​​of various power supply performance indicators such as voltage, current, power factor, temperature, etc. The power supply performance characterization vector is an important basis for power supply performance analysis and diagnosis. For example, assume that a power supply performance characterization vector includes a voltage reading, a current reading, and a temperature reading of a power supply. Specifically, this power supply performance characterization vector can be expressed as [220V, 5A, 45°C], where 220V is the voltage reading, 5A is the current reading, and 45°C is the temperature reading. This power supply performance characterization vector can characterize the performance state of the power supply at a specific moment.

[0039] For example, in a scenario of diagnosing a server power supply in a data center, the computer device performs in-depth mining of each prior information to extract a vector that can characterize the power supply performance. Assume that a prior information describes that the power supply overheated after a specific time node. The computer device first analyzes the monitoring data before and after this time node to extract features related to the overheating condition, such as temperature, fan speed, load current, etc. The computer device reduces the dimensionality of the data based on principal component analysis (PCA) while retaining the most important features. For example, the original monitoring data may include dozens of different sensor readings, but through PCA, the computer device can compress this data into a power supply performance characterization vector containing only a few key features. After PCA processing, the computer device obtains a power supply performance characterization vector containing three features: [temperature, fan speed, load current]. For the overheating condition, this power supply performance characterization vector may have specific values ​​similar to [70°C, 2000RPM, 80%], where 70°C represents temperature, 2000RPM represents fan speed, and 80% represents load current. This power supply performance characterization vector not only describes the state of the power supply when it is overheated, but can also be used for subsequent performance analysis and fault diagnosis. For example, the computer device can compare this vector with the vector in the normal state to determine whether the performance of the power supply has been abnormal. By mining the power supply performance characterization vector of each prior information, the computer device can have a deeper understanding of the performance status and potential problems of the power supply. These power supply performance characterization vectors not only provide a comprehensive description of the power supply status, but also provide strong support for subsequent performance analysis, fault diagnosis and preventive maintenance. In addition, by using dimensionality reduction techniques such as PCA, this technical solution can also effectively process high-dimensional monitoring data, improving the efficiency of data processing and analysis.

[0040] Step S40: integrating each power monitoring data in the past power monitoring data stream with the corresponding power performance characterization vector of prior information into integrated power monitoring data to obtain a past integrated power monitoring data stream.

[0041] Specifically, the computer device will first process the past power monitoring data stream, which contains various monitoring data of the power supply in the past period of time, such as voltage, current, power factor, temperature, etc. These data are arranged in chronological order and constitute a historical record of power supply performance. However, these raw monitoring data themselves are not sufficient to fully describe the performance status of the power supply, so they need to be combined with prior information.

[0042] The prior information contains power performance characterization vectors corresponding to the monitoring data. These characterization vectors are extracted from the prior information and can represent the performance status of the power supply under specific conditions. For example, a prior information may describe the performance characteristics of the power supply in a high temperature environment, and the corresponding power performance characterization vector may contain relevant characteristics such as temperature and heat dissipation efficiency. During the integration process, the computer device will match and combine each power monitoring data with the power performance characterization vector of the corresponding prior information. Specifically, the numerical value of the monitoring data can be matched with the characteristics of the characterization vector to form a more comprehensive data description. For example, if a certain monitoring data records the voltage value of the power supply, the computer device will find the part of the power performance characterization vector of the corresponding prior information that describes the voltage characteristics and combine the two. Through integration, each data point in the past power monitoring data stream is given richer information, no longer an isolated numerical value, but integrated data closely related to the power performance status. In this way, the computer device can more accurately analyze the performance history of the power supply, discover potential problems, and predict future performance trends. For example, suppose that a certain original power monitoring data records a voltage of 220V and a current of 5A. The corresponding power performance characterization vector of prior information includes the performance characteristics of the power supply in a high temperature environment, such as a heat dissipation efficiency of 0.8 (the higher the value, the better the heat dissipation). The integrated power monitoring data may be a comprehensive data description that combines this information, such as {voltage: 220V, current: 5A, heat dissipation efficiency: 0.8}.

[0043] It should be emphasized that the process of integrating power monitoring data is not a simple data splicing, but requires the use of certain algorithms and models. For example, a machine learning model can be used to fuse and extract features from monitoring data and prior information. The specific model selection can be determined based on actual needs and data characteristics, such as support vector machines (SVM), random forests, or neural networks. These models can automatically learn the laws and patterns in the data, thereby extracting more meaningful features for subsequent power performance analysis and diagnosis.

[0044] As an implementation mode, step S40: integrating each power monitoring data in the past power monitoring data stream with the corresponding power performance characterization vector of prior information into integrated power monitoring data to obtain the past integrated power monitoring data stream, including:

[0045] Step S41: for each power monitoring data in the past power monitoring data stream, the power monitoring data is combined with the corresponding power performance characterization vector of the prior information to obtain the integrated power monitoring data corresponding to each power monitoring data.

[0046] Step S41 is an important part of the power supply performance diagnosis process, which involves combining each power supply monitoring data in the past power supply monitoring data stream with the corresponding power supply performance characterization vector of the prior information. This process aims to integrate data from different sources to more comprehensively and accurately describe the performance status of the power supply.

[0047] Specifically, the computer device traverses each power monitoring data in the past power monitoring data stream. For each data point, the computer device finds the power performance characterization vector of the corresponding prior information. The power performance characterization vector contains key features that can represent the power performance, which can be extracted from a large amount of prior data through a machine learning model. Then, the computer device combines each power monitoring data with the corresponding power performance characterization vector. This combination process can be understood as fusing and splicing data of two different dimensions together to form a richer data description. For example, if a power monitoring data records the values ​​of voltage and current, and the corresponding power performance characterization vector contains features such as heat dissipation efficiency and power factor, then the combined integrated power monitoring data will contain all of this information.

[0048] To illustrate this process more specifically, in one example, it is assumed that the past power monitoring data stream contains three power monitoring data, which record voltage, current and temperature respectively. At the same time, the power performance characterization vector including the corresponding prior information includes characteristics such as heat dissipation efficiency, power factor and stability. In step S41, the computer device combines each power monitoring data with its corresponding power performance characterization vector. For example, the first power monitoring data (voltage) will be combined with the corresponding power performance characterization vector (heat dissipation efficiency, power factor, etc.) into an integrated power monitoring data. Similarly, the second and third power monitoring data will also be combined in a similar way.

[0049] In this way, a series of integrated power monitoring data can be obtained, each of which contains the original monitoring data and the corresponding power performance characterization vector information. These data provide a more comprehensive and accurate data basis for subsequent power performance analysis and diagnosis.

[0050] As an embodiment, the method further includes: arranging the power performance characterization vectors of the prior information corresponding to each power monitoring data according to the arrangement position of the power monitoring data in the past power monitoring data stream, and generating a power performance characterization vector stream.

[0051] Then, step S41, for each power monitoring data in the past power monitoring data stream, combines the power monitoring data with the corresponding power performance characterization vector of the prior information to obtain the integrated power monitoring data corresponding to each power monitoring data, which may include:

[0052] The power monitoring data and the power performance characterization vector at the same position in the past power monitoring data stream and the power performance characterization vector stream are combined to obtain integrated power monitoring data corresponding to each of the power monitoring data.

[0053] Specifically, the computer device arranges the power performance characterization vector of the prior information corresponding to each power monitoring data according to the arrangement position of the power monitoring data in the past power monitoring data stream. This process can be understood as that when the computer device processes a large amount of power monitoring data, it not only pays attention to the numerical information of each data point, but also pays attention to the position of these data points in the time series. Because the performance state of the power supply often changes over time, the position information of the data point is crucial to accurately understand the power performance. In order to incorporate this position information into the analysis, the computer device generates a power performance characterization vector stream. Each element in this vector stream is a power performance characterization vector corresponding to the power monitoring data at a specific position in the past power monitoring data stream. In this way, the power performance characterization vector stream retains the time series information in the original monitoring data, and incorporates the prior power performance characteristics related to each data point. Next, in step S41, the computer device combines each power monitoring data in the past power monitoring data stream with the corresponding power performance characterization vector of the prior information. The combination here is an accurate data integration process, and the computer device will pair and splice the power monitoring data and power performance characterization vector at the same position. This combination ensures that each integrated power monitoring data contains the original monitoring value and the power performance characteristics directly related to the value.

[0054] For example, if a data point in the past power monitoring data stream records the voltage value at a specific moment, the computer device will find the power performance characterization vector corresponding to the position of the data point (this vector may contain information such as the heat dissipation efficiency and power factor of the power supply at that moment), and combine the two into an integrated power monitoring data. In this way, the integrated data not only reflects the real-time value of the voltage, but also provides richer contextual information about the power performance status. In this way, the computer device can generate an integrated power monitoring data stream containing comprehensive information. This data stream contains both the original power monitoring data and the prior power performance characteristics that are closely related to each data point. This data integration provides a solid data foundation for subsequent power performance analysis, fault diagnosis, and preventive maintenance.

[0055] Step S42: Arranging the integrated power monitoring data corresponding to each power monitoring data according to the arrangement position of the power monitoring data in the past power monitoring data stream to generate a past integrated power monitoring data stream.

[0056] Specifically, in step S42, the computer device arranges the integrated power monitoring data corresponding to each power monitoring data according to the arrangement position of the power monitoring data in the past power monitoring data stream, and finally generates a past integrated power monitoring data stream. Among them, when processing the power monitoring data, the computer device not only pays attention to the numerical information of each data point, but also pays attention to the position of these data points in the time series. Because the performance state of the power supply changes over time, the position information of the data point is crucial for accurately understanding the power supply performance. In step S42, the computer device first reviews the integrated power monitoring data generated in step S41. These integrated data are a combination of the original power monitoring data and the power performance characterization vector of the corresponding prior information, and they contain richer power performance information. Then, the computer device rearranges these integrated power monitoring data according to their position information in the original past power monitoring data stream. This arrangement process ensures that the integrated data still maintains the time series characteristics of the original data, that is, the position of each integrated power monitoring data in the past integrated power monitoring data stream corresponds to its position in the original past power monitoring data stream.

[0057] For example: Assume that the original past power monitoring data stream contains voltage and current monitoring data arranged in chronological order, that is, [V1, I1, V2, I2, ...]. In step S41, the computer device finds the corresponding power performance characterization vector for each voltage and current data point, and combines them to generate integrated power monitoring data, such as [(V1, P1), (I1, Q1), (V2, P2), (I2, Q2), ...], where P and Q represent power performance characterization vectors. In step S42, the computer device rearranges these integrated power monitoring data in the order of the original data to generate a past integrated power monitoring data stream, that is, [(V1, P1), (I1, Q1), (V2, P2), (I2, Q2), ...]. This past integrated power monitoring data stream contains both voltage and current monitoring data and power performance characteristics related to each data point, providing a comprehensive and accurate data basis for subsequent power performance analysis and diagnosis. Through such data integration and processing, computer equipment can more comprehensively understand the performance status of the power supply and provide strong support for subsequent analysis and decision-making.

[0058] Step S50: Based on the past integrated power monitoring data stream, infer the power monitoring data of each monitoring node of the target power supply in a plurality of uninterrupted monitoring nodes to obtain an inferred power monitoring data stream; the monitoring node is located after each of the past nodes.

[0059] Specifically, the computer device uses the past integrated power monitoring data stream generated in the previous step as input. The past integrated power monitoring data stream contains various integrated monitoring data of the power supply in the past period of time. These data not only include the original monitoring values ​​such as voltage and current, but also incorporate the power performance characteristics extracted from the prior information. Based on these data, the computer device needs to infer the power monitoring data of the target power supply in multiple uninterrupted monitoring nodes in the future.

[0060] To achieve this goal, the computer device may use one or more machine learning models for reasoning. The machine learning model can learn the laws and patterns in the past integrated power monitoring data stream, and predict the future power monitoring data based on the learned knowledge. The specific model selection can be determined according to the actual needs and data characteristics. For example, you can choose to use a recurrent neural network (RNN) or a long short-term memory network (LSTM) that is suitable for processing time series data. Take LSTM as an example. It is a special RNN that can learn long-term dependencies and is very suitable for processing time series data. In step S50, the computer device can use the past integrated power monitoring data stream as the input of the LSTM model and train the model to learn the law of changes in power performance. Then, using the trained LSTM model, the computer device can infer the power monitoring data of the target power supply in multiple uninterrupted monitoring nodes in the future.

[0061] This process can be understood as the computer device building a model that can simulate power performance changes by learning past power performance data. Then, using this model, the computer device can predict what the power monitoring data may be at each monitoring node in the future. In this way, an inference power monitoring data stream is obtained, which describes the predicted performance status of the target power supply in the future.

[0062] It should be noted that the power monitoring data stream obtained by inference is not an absolutely accurate prediction result, but a possibility based on past data and model learning. However, by selecting a suitable model and optimizing algorithm parameters, the accuracy and reliability of inference can be improved, thus providing strong support for subsequent power performance analysis and diagnosis.

[0063] In one embodiment, step S50, based on the past integrated power monitoring data stream, inferring the power monitoring data of each monitoring node of the target power supply in a plurality of uninterrupted monitoring nodes to obtain the inferred power monitoring data stream, may specifically include:

[0064] Step S51: According to the power performance characterization vector included in each integrated power monitoring data in the past integrated power monitoring data stream, the past integrated power monitoring data stream is segmented to obtain multiple data segments; wherein the power performance characterization vector included in the integrated power monitoring data in the same data segment is the same, and the power performance characterization vectors included in the integrated power monitoring data in the two previous and next data segments are different.

[0065] Specifically, the computer device segments the past integrated power monitoring data stream, and the basis for this segmentation is the power performance characterization vector contained in each integrated power monitoring data. The power performance characterization vector is a set of key features that describe the power performance status, and may include multiple dimensions such as heat dissipation efficiency, power factor, and stability. When performing data segmentation, the computer device checks the power performance characterization vector in each integrated power monitoring data, and groups the integrated power monitoring data with the same power performance characterization vector into one data segment. In this way, the integrated power monitoring data in each data segment is consistent in power performance characterization, while the integrated power monitoring data between different data segments have differences in power performance characterization.

[0066] For example: Assume that the past integrated power monitoring data stream contains a series of integrated power monitoring data, each of which consists of a voltage value and a corresponding power performance characterization vector. The computer device first identifies all data with the same power performance characterization vector and divides them into the same data segment. In this way, if there are three consecutive data points in the data stream with the same power performance characterization vector A, and the next two data points have different power performance characterization vectors B, then the computer device will divide the data stream into two data segments, the first data segment contains three data points with vector A, and the second data segment contains two data points with vector B. Through such data segmentation processing, the computer device can identify the data segments under different power performance states in the past integrated power monitoring data stream, which provides a basis for the subsequent reasoning of power monitoring data based on these data segments. In step S52, the computer device will use these segmented data to infer the power monitoring data of the target power supply at the uninterrupted monitoring node, thereby obtaining the reasoning power monitoring data stream. This process involves the identification and analysis of changes in power performance states, and may rely on advanced data analysis tools such as machine learning models or neural networks to achieve accurate reasoning.

[0067] Step S52: Based on the multiple data segments, infer the power monitoring data of each monitoring node of the target power source in the multiple uninterrupted monitoring nodes to obtain an inferred power monitoring data stream.

[0068] Specifically, the computer device utilizes previously identified data segments that reflect monitoring data of the power supply under different performance states. Since the integrated power monitoring data within each data segment has the same power performance characterization vector, this indicates that they are in similar power performance states. Therefore, the computer device can infer the possible state of the power supply at a time point that is not directly monitored based on these data segments. To achieve this reasoning, the computer device can be implemented with the help of a machine learning model or a neural network. For example, a recurrent neural network (RNN) or a long short-term memory network (LSTM) can be used to process time series data. These network structures are particularly suitable for processing time-dependent data and can learn long-term dependencies in the data.

[0069] In the specific implementation, the computer device will first provide these data segments as input to a pre-trained machine learning model. The model has learned to extract features from past power monitoring data and predict future power status based on these features. By inputting multiple data segments, the model can capture the changing trend of the power performance status and infer the status of the power supply at the unmonitored time point.

[0070] For example, if past data shows that in a certain power performance state, the voltage value usually fluctuates within a certain range, then when the computer device detects a similar power performance characterization vector, it can predict that the current voltage value will also fall within this range. In this way, the computer device is able to generate an inferred power monitoring data for each uninterrupted monitoring node. Finally, the computer device combines these inferred power monitoring data in chronological order to form a continuous inferred power monitoring data stream. This data stream not only fills the gap between direct monitoring, but also provides a continuous and comprehensive view of power performance, which helps to timely detect potential power problems and perform preventive maintenance.

[0071] As a possible design, step S52, based on the multiple data segments, inferring the power monitoring data of each monitoring node of the target power supply in the multiple uninterrupted monitoring nodes to obtain an inferred power monitoring data stream, may specifically include:

[0072] Step S521: performing embedding mapping on the multiple data segments respectively to obtain an embedding mapping vector corresponding to each of the data segments;

[0073] Step S522: generating an embedded mapping vector stream corresponding to the past power monitoring data stream based on the embedded mapping vectors corresponding to each of the data segments;

[0074] Step S523: based on the embedded mapping vector flow corresponding to the past power monitoring data flow, inferring the power monitoring data of each monitoring node of the target power supply in a plurality of uninterrupted monitoring nodes to obtain an inferred power monitoring data flow.

[0075] In a possible design scheme, when the computer device executes step S52, it is further refined into three sub-steps: S521, S522 and S523. These sub-steps together constitute a detailed process of inferring the power monitoring data of the target power in multiple uninterrupted monitoring nodes based on multiple data segments. The core task of step S521 is to perform embedding mapping on multiple data segments respectively to obtain the embedding mapping vector corresponding to each data segment. In this context, embedding mapping refers to converting each data segment (containing a series of power monitoring data, reflecting the power performance status within a specific time period) into a vector representation in a high-dimensional space. This process aims to capture the intrinsic features and patterns in the data segments so that semantically similar (i.e., similar power performance status) data segments are closer in position in the vector space. In order to achieve embedding mapping, the computer device can use a pre-trained Transformer model. The Transformer model is an advanced neural network architecture that is particularly suitable for processing sequence data (such as text, time series, etc.). It uses the self-attention mechanism to capture long-term dependencies in the input sequence and gradually extracts and transforms the features of the input data through multiple stacked encoder and decoder layers.

[0076] In specific implementation, the computer device will input each data segment as an independent sequence into the encoder of the Transformer model. The encoder pays attention to each data point in the sequence through the self-attention mechanism, and generates the encoding representation of each position based on the information of all data points. These encoding representations are then further transformed through the feedforward neural network to finally obtain the embedding mapping vector corresponding to each data segment.

[0077] For example, suppose a data segment contains the voltage, current and temperature monitoring data of a power supply over a period of time. These data reflect the performance status of the power supply during the period of time. Through the embedding mapping of the Transformer model, the computer device can convert these monitoring data into a high-dimensional vector that captures the key features of the power supply performance status in the data segment. Different data segments will be mapped to different positions in the vector space according to their different power supply performance states. After completing step S521, the computer device will obtain a series of embedded mapping vectors corresponding to the data segments. These vectors constitute a high-level representation of the past power supply monitoring data stream, providing a basis for the reasoning process in subsequent steps S522 and S523. In step S522, these embedded mapping vectors will be combined into a continuous embedded mapping vector stream to capture the changing trend of the power supply performance status during the entire monitoring period. Finally, in step S523, the computer device will infer the power supply monitoring data of the target power supply in multiple uninterrupted monitoring nodes based on this embedded mapping vector stream, thereby obtaining an inferred power supply monitoring data stream.

[0078] Specifically, the goal of step S522 is to generate a continuous embedding mapping vector stream based on the embedding mapping vector corresponding to each data segment obtained in step S521, and this vector stream can represent the overall characteristics of the past power monitoring data stream. Specifically, the computer device first obtains the result output by step S521, that is, a series of discrete data segment embedding mapping vectors. These vectors correspond to different data segments, and each vector contains the encoding information of the power monitoring data in the corresponding data segment. Next, the computer device arranges these embedding mapping vectors in sequence according to the time sequence of these data segments in the original monitoring data stream. During the arrangement process, the computer device does not change the internal representation of each embedding mapping vector, but maintains their integrity, and only combines them in the time dimension. In this way, a sequence of embedding mapping vectors arranged in time order is formed, that is, an embedding mapping vector stream. This vector stream not only retains the characteristic information inside each data segment, but also reflects the correlation between the data segments through time order.

[0079] For example, suppose there are three data segments, corresponding to power monitoring data in three different time periods. After processing in step S521, each data segment is converted into an embedded mapping vector. In step S522, the computer device arranges the three embedded mapping vectors in the corresponding time sequence to form an embedded mapping vector stream. This vector stream represents the overall characteristics of the power monitoring data in these three time periods, and provides the necessary information basis for the subsequent reasoning steps.

[0080] It should be noted that step S522 does not involve further processing or modification of the embedded mapping vectors, but rather combines them in an appropriate manner for use in subsequent steps. The key to this step is to maintain the time series characteristics of the embedded mapping vectors, thereby ensuring that the generated embedded mapping vector stream can accurately reflect the overall characteristics of the past power monitoring data stream.

[0081] The goal of step S523 is to infer the power monitoring data of the target power supply in multiple uninterrupted monitoring nodes based on the embedded mapping vector flow corresponding to the past power monitoring data flow, and finally generate an inferred power monitoring data flow. This step makes full use of machine learning models, especially deep learning models, to mine the potential features and rules in the embedded mapping vector flow. Specifically, the computer device will use one or more machine learning models to analyze and infer the embedded mapping vector flow in step S523. These models may be recurrent neural networks (RNNs), long short-term memory networks (LSTMs), gated recurrent units (GRUs) or Transformers, etc. These models are good at processing sequence data and can capture long-term dependencies in sequences.

[0082] Taking LSTM as an example, the model can learn long-term dependency information in sequence data and avoid gradient vanishing or exploding in long-term dependency problems through its special gating mechanism. In step S523, LSTM can be used to receive the embedding mapping vector stream as input and output the inference power monitoring data stream. Each unit of LSTM receives the output of the previous unit and the input at the current moment (i.e., the current vector in the embedding mapping vector stream), and then combines the two to generate the output at the current moment. This process will proceed step by step along the time series until all vectors in the embedding mapping vector stream are processed. Finally, the output sequence of the LSTM model constitutes the inference power monitoring data stream, which reflects the prediction or inference of the future power state of the computer device based on past power monitoring data. This inference power monitoring data stream can be used to evaluate the current performance status of the power supply, predict possible problems in the future, or provide decision support for the maintenance and management of the power supply. It should be noted that the specific machine learning model used in step S523 may be adjusted according to the actual application scenario and data characteristics. In any case, the goal of this step is to use the machine learning model to extract useful information from the embedding mapping vector stream and generate an inference power monitoring data stream that can accurately reflect the power performance status.

[0083] As an implementation mode, the above-mentioned embedding mapping vector is generated by using a debugged embedding mapping neural network. Based on this, the method provided in the embodiment of the present application also includes a debugging process of the embedding mapping neural network, and the debugging process of the embedding mapping neural network includes the following steps:

[0084] Step S61: Obtain an active debugging sample set, which includes debugging samples corresponding to the target power supply in multiple past node sequences with intersections, and the debugging samples corresponding to the past node sequences include power monitoring data of each past node of the target power supply in the past node sequences. The power monitoring data in the debugging samples are arranged according to the order of each node.

[0085] Among them, a past node sequence refers to a series of monitoring nodes arranged in chronological order, which record the status information of the target power supply at different time points. Each node contains the power monitoring data at a specific moment, such as the values ​​of parameters such as voltage and current. The intersection refers to a part of the nodes shared between different past node sequences. In other words, if there are two or more past node sequences that have common monitoring nodes at certain time points, then these common nodes constitute the intersection between these sequences. The existence of an intersection means that these sequences record the same power status information in certain time periods, which is very important for extracting consistent and comparable features when debugging embedding mapping neural networks.

[0086] In order to generate the embedding mapping vector more accurately, a carefully debugged embedding mapping neural network is required. The debugging process directly affects the embedding mapping quality of the subsequent power monitoring data. For step S61, its core task is to obtain a set of active debugging samples. The active debugging sample set contains debugging samples of the target power supply on multiple past node sequences with intersections. The debugging samples corresponding to each past node sequence include the power monitoring data of each past node of the target power supply in the past node sequence. These data are arranged in the order of the nodes to form an orderly power monitoring data stream.

[0087] To understand this step more specifically, for example, there is a target power supply that is monitored and records the power supply status data at different time points. These time points constitute the so-called past node sequence. When data is collected on multiple such sequences (there are some common time points between them, that is, the intersection), a set of positive debugging samples is obtained. This set is called positive because these samples reflect the normal or expected state of power supply performance to some extent, which is very valuable for training neural networks. By letting the neural network learn these positive samples, it can be expected that it will be able to make correct predictions or mappings when it encounters similar data in the future. In actual operation, computer equipment will be responsible for collecting and processing this data. Power monitoring data is obtained from various sources (it may be sensors, log files, etc.) and organized into debugging samples in chronological order. Then, these samples are combined into a large data set, namely the set of positive debugging samples, for use in subsequent steps.

[0088] Step S62: Obtain a passive debugging sample set, where the passive debugging sample set includes debugging samples of the target power source corresponding to a plurality of past node sequences that have no intersection.

[0089] Among them, unlike the positive debugging samples, the negative debugging sample set includes debugging samples corresponding to the target power supply in multiple past node sequences that do not have intersections. By obtaining these negative debugging samples, the computer device can train the embedding mapping neural network to identify and handle abnormal situations in power supply performance. This is crucial to improving the robustness and generalization ability of the neural network, because in actual applications, the neural network is likely to encounter abnormal situations that are different from the training data.

[0090] Step S63: for each debugging sample in the active debugging sample set, each power monitoring data in the debugging sample is respectively integrated with the power performance characterization vector of the corresponding prior information into integrated power monitoring data, and an integrated power monitoring data stream corresponding to each debugging sample in the active debugging sample set is generated.

[0091] Specifically, each debugging sample contains a series of power monitoring data, which reflects the state of each node of the target power supply in the past node sequence. These power monitoring data do not exist in isolation, but are associated with prior information. Prior information usually refers to information that is known before the current analysis or learning, which can provide valuable insights into the power supply performance. In step S63, the computer device accesses the prior information library and extracts the power performance characterization vector corresponding to each power monitoring data. These characterization vectors are quantitative representations of power performance characteristics, which may be obtained through previous machine learning models (such as support vector machines, decision trees or neural networks) training, and are used to describe the performance behavior of the power supply under different states. Next, the computer device combines each power monitoring data in the debugging sample with the corresponding power performance characterization vector to generate integrated power monitoring data. This integration may be achieved through simple splicing, weighted averaging, or more complex mathematical operations (such as matrix multiplication), depending on the requirements of the machine learning model and data structure adopted. After the integration is completed, each debugging sample will no longer be a series of isolated power monitoring data points, but a set of integrated power monitoring data streams that incorporate prior knowledge. These data streams not only contain raw power state information, but also incorporate previously learned power performance characteristics, providing richer and more contextual information.

[0092] For example, if a certain power supply monitoring data represents a voltage value, and the corresponding power supply performance characterization vector describes the efficiency characteristics of the power supply at different voltage levels, then the integrated data will be able to simultaneously reflect the actual value of the voltage and the efficiency status of the power supply at that voltage level.

[0093] Finally, an updated set of active debugging samples is output, in which each debugging sample has been converted into an integrated power monitoring data stream. These integrated data streams will serve as input for subsequent neural network training or debugging, helping to improve the model's ability to diagnose and predict power performance.

[0094] Step S64: for each debugging sample in the passive debugging sample set, each power monitoring data in the debugging sample is respectively integrated with the power performance characterization vector of the corresponding prior information into integrated power monitoring data, and an integrated power monitoring data stream corresponding to each debugging sample in the passive debugging sample set is generated.

[0095] Similarly, the passive debugging sample set contains debugging samples of the target power supply on multiple non-intersecting past node sequences, which usually reflect the abnormal or faulty state of the power supply performance. Each debugging sample contains a series of power supply monitoring data, which records the status information of the power supply at each node.

[0096] The computer device accesses a database storing prior information and performs data integration operations. For each power monitoring data in the debugging sample, the computer device integrates it with the corresponding power performance characterization vector. The integration method can be simple data concatenation or can be achieved through complex mathematical operations (such as weighted average, matrix multiplication, etc.), depending on the requirements of the machine learning model and data structure adopted. The integrated data will contain both the original power monitoring data and the power performance characteristics associated with it, thereby providing richer and more contextual information. After the integration is completed, each negative debugging sample will be converted into a set of integrated power monitoring data streams. These data streams not only contain the actual state information of the power supply, but also incorporate the previously learned power performance characteristics, so that the neural network can better understand and process these samples. For example, suppose that the power monitoring data in a negative debugging sample represents an abnormal voltage fluctuation, and the corresponding power performance characterization vector describes the type of fault that may occur in the power supply under different voltage fluctuation conditions. By integrating these two information sources, the computer device can generate an integrated power monitoring data stream, which contains both the actual data of the voltage fluctuation and the fault characteristic information associated with the fluctuation. Finally, an updated set of negative debugging samples is output, in which each debugging sample has been converted into an integrated power monitoring data stream. These integrated data streams will serve as important inputs for subsequent neural network training or debugging, helping to improve the model's ability to identify and diagnose abnormal power performance conditions.

[0097] Step S65: Debugging the embedded mapping neural network based on the integrated power monitoring data stream corresponding to each debugging sample in the active debugging sample set and the integrated power monitoring data stream corresponding to each debugging sample in the passive debugging sample set.

[0098] Specifically, the computer device will use the integrated power monitoring data streams in the active debugging sample set and the passive debugging sample set to train the embedding mapping neural network. During the training process, the neural network will try to learn the mapping relationship from the input data (integrated power monitoring data stream) to the target output (power performance characterization). By continuously adjusting the weights and bias parameters of the neural network, the computer device can minimize the difference between the predicted output and the actual output, thereby optimizing the performance of the neural network.

[0099] For example, assume that the embedding mapping neural network is a deep convolutional neural network (DCNN), which can extract features from the input data through multiple convolutional layers and pooling layers, and map these features to the embedding space through a fully connected layer. During the training process, the computer device uses the backpropagation algorithm to calculate the gradient of the loss function and update the parameters of the neural network based on these gradients.

[0100] In this way, the embedding mapping neural network can gradually learn the effective mapping relationship from integrating power monitoring data streams to power performance characterization. Eventually, after enough training iterations, the neural network will be able to accurately predict and identify the performance under different power states.

[0101] In some implementations, step S65, based on the integrated power monitoring data stream corresponding to each debugging sample in the active debugging sample set and the integrated power monitoring data stream corresponding to each debugging sample in the passive debugging sample set, debugging the embedded mapping neural network may specifically include:

[0102] Step S651: Load the integrated power monitoring data stream corresponding to each debugging sample in the active debugging sample set into the embedding mapping neural network for embedding mapping to obtain multiple embedding mapping vectors, and obtain the embedding mapping vector stream corresponding to each debugging sample in the active debugging sample set.

[0103] Each debugging sample corresponds to an integrated power monitoring data stream, which contains the monitoring data of the target power supply on the past node sequence with intersection. These data reflect the performance of the power supply under specific conditions and are necessary information for debugging the neural network. In step S651, the computer device loads the integrated power monitoring data stream of each debugging sample into the embedded mapping neural network.

[0104] When the integrated power monitoring data stream is loaded into the neural network, the neural network will map the data into a set of embedding mapping vectors through a series of operations and processing. Each embedding mapping vector corresponds to a data point in the input data stream, which captures the position and characteristics of the data point in the embedding space.

[0105] Through this process, each debugging sample in the set of active debugging samples is converted into an embedding map vector stream. This vector stream contains the embedding map vectors of all data points in the debugging sample, which together describe the performance of the power supply on the past node sequences with intersection.

[0106] Step S652: Load the integrated power monitoring data stream corresponding to each debugging sample in the passive debugging sample set into the embedding mapping neural network for embedding mapping to obtain multiple embedding mapping vectors, and obtain the embedding mapping vector stream corresponding to each debugging sample in the passive debugging sample set.

[0107] Similar to step S651 , the input becomes the integrated power monitoring data stream corresponding to each debugging sample in the passive debugging sample set.

[0108] Step S653: determining the commonality measurement results between the embedded mapping vector flows corresponding to each debugging sample in the active debugging sample set, and obtaining the commonality measurement results corresponding to the active debugging sample set.

[0109] Step S654: determining the commonality measurement results between the embedded mapping vector flows corresponding to each debugging sample in the passive debugging sample set, and obtaining the commonality measurement results corresponding to the passive debugging sample set.

[0110] Step S653 and step S654 involve determining the commonality measurement results between the embedded mapping vector streams in the debugging sample set. It aims to quantify the similarity or consistency between different debugging samples in the embedding space. First, step S653 focuses on the active debugging sample set. In the active debugging sample set, each debugging sample corresponds to an integrated power monitoring data stream, and these data streams have been processed by the embedding mapping neural network and converted into an embedded mapping vector stream. Commonality measurement is a method for evaluating the degree of similarity between these vector streams. The computer device calculates the similarity between the embedded mapping vector streams of each pair of debugging samples in the active debugging sample set, which can be achieved through various distance metrics or similarity metrics, such as Euclidean distance, cosine similarity, etc. The results of these calculations are combined to form a commonality measurement result, which reflects the overall distribution of debugging samples in the active debugging sample set in the embedding space.

[0111] For example, if the embedding mapping vector flows of two debugging samples are close to each other in the embedding space, then their commonality measurement results will be high, indicating that the two samples have similarities in power supply performance. On the contrary, if the two vector flows are far apart, the commonality measurement results will be low, indicating that there are large differences in their power supply performance. Step S654 is similar to step S653, but processes a set of negative debugging samples. The debugging samples in the negative debugging sample set correspond to the performance of the target power supply on multiple past node sequences that do not have intersections. The computer device also calculates the commonality measurement results between the embedding mapping vector flows of each pair of debugging samples in this set. These calculations reveal the distribution characteristics of the debugging samples in the negative debugging sample set in the embedding space, thereby providing information about the performance differences of the power supply under different conditions.

[0112] The acquisition of commonality metrics is crucial for the subsequent steps. They not only provide quantitative indicators of similarities and differences within the debug sample set, but also provide a basis for debugging the embedding mapping neural network. If the commonality metrics of the positive debug sample set show high similarity between samples, this may mean that the neural network performs well in capturing the characteristics of normal power performance. On the contrary, if the commonality metrics of the negative debug sample set show large differences, this may reveal the ability of the neural network in distinguishing different abnormal or fault states.

[0113] Step S655: Debugging the embedding mapping neural network based on the commonality measurement results corresponding to the active debugging sample set and the commonality measurement results corresponding to the passive debugging sample set.

[0114] The goal of step S655 is to optimize the performance of the neural network by analyzing the distribution of the debugging sample set in the embedding space to improve its diagnostic accuracy for the power supply state. Specifically, the computer device first considers the commonality measurement results corresponding to the active debugging sample set. These commonality measurement results reflect the similarity and distribution of the active debugging samples in the embedding space. If the commonality measurement results show that there is a high degree of similarity between the active debugging samples, this means that the embedding mapping neural network performs well in capturing the normal power supply performance characteristics. However, if the commonality measurement results show a large difference, this may mean that the neural network needs further adjustment and optimization. Next, the computer device considers the commonality measurement results corresponding to the negative debugging sample set. The negative debugging sample set contains the performance of the target power supply on multiple past node sequences that do not have an intersection, and its commonality measurement results reveal the ability of the neural network to distinguish different abnormal or fault states. If the commonality measurement results of the negative debugging samples show that there is a large difference between the samples, this may mean that the neural network can effectively identify different fault modes. However, if the commonality measurement results are too low, it may indicate that the neural network has difficulty in distinguishing fault states.

[0115] Based on these commonality measurement results, the computer device will debug the embedding mapping neural network. The debugging process may involve multiple aspects, including adjusting the parameters of the neural network, optimizing the network structure, or improving the training strategy. For example, the sample distribution in the embedding space can be improved by adjusting the weight parameters of the neural network, or a higher level of feature representation can be extracted by increasing the number of network layers. The goal of debugging is to improve the diagnostic accuracy of the neural network on power performance so that it can better adapt to various normal and abnormal conditions.

[0116] It should be noted that debugging the embedding mapping neural network is an iterative process, which may require multiple attempts and adjustments to achieve the best performance. After each debugging, the computer equipment will re-evaluate the performance of the neural network and make further optimizations based on the evaluation results.

[0117] In some implementations, step S655, debugging the embedding mapping neural network based on the commonality measurement results corresponding to the active debugging sample set and the commonality measurement results corresponding to the passive debugging sample set, may specifically include the following steps:

[0118] Step S6551a: Perform a first numerical calculation on the commonality measurement results corresponding to the multiple active debugging sample sets to obtain a first commonality measurement result.

[0119] Step S6552a: Perform the first numerical calculation on the commonality measurement results corresponding to the plurality of passive debugging sample sets respectively to obtain a second commonality measurement result.

[0120] In the above step S6551a, the computer device performs a first numerical calculation on the commonality measurement results corresponding to the multiple sets of active debugging samples. The first numerical calculation here can be a mathematical operation such as weighted average, summation, maximum or minimum value of the commonality measurement results. The specific operation method depends on the design and implementation requirements of the technical solution. Through these numerical calculations, the computer device can obtain a comprehensive commonality measurement result, that is, the first commonality measurement result. This first commonality measurement result reflects the overall distribution and similarity of the active debugging sample set in the embedding space, and provides an important basis for subsequent neural network debugging.

[0121] Step S6552a is similar to step S6551a, but the object of processing is a set of passive debugging samples. In this step, the computer device performs the same first numerical calculation on the commonality measurement results corresponding to multiple sets of passive debugging samples. Through calculation, a comprehensive commonality measurement result, namely the second commonality measurement result, is obtained. This second commonality measurement result reveals the overall distribution and degree of difference of the passive debugging sample set in the embedding space. Compared with the first commonality measurement result, the second commonality measurement result may present different characteristics, because the passive debugging sample set contains the performance of the target power supply on multiple past node sequences that do not have intersections, reflecting the behavior of the power supply under abnormal or faulty conditions.

[0122] The purpose of these two steps is to comprehensively evaluate the performance of the embedding mapping neural network from both positive and negative perspectives. By comparing the first commonality measurement results with the second commonality measurement results, the computer device can understand the performance differences of the neural network under normal and abnormal conditions, thereby guiding the subsequent debugging process.

[0123] Step S6553a: Debugging the embedding mapping neural network based on the error between the first commonality measurement result and the second commonality measurement result.

[0124] The first commonality measurement result is calculated from the positive debugging sample set, which represents the common characteristics of the target power supply on multiple past node sequences with intersections. The second commonality measurement result is calculated from the negative debugging sample set, which reflects the differences of the target power supply on multiple past node sequences without intersections. Because the samples in the positive debugging sample set have intersecting past node sequences, their representation vectors in the embedding space should be relatively similar, which means that the neural network needs to learn to capture the common characteristics between these samples. On the contrary, because the samples in the negative debugging sample set have no intersection between past node sequences, their representation vectors in the embedding space should have large differences, and the neural network needs to be able to distinguish these different samples.

[0125] In step S6553a, the computer device calculates the error between the first commonality measurement result and the second commonality measurement result, which reflects the ability of the neural network to capture common features and distinguish different samples. By minimizing this error, the neural network can learn a more accurate embedding representation, so that samples with intersections are closer in the embedding space, while samples without intersections are more dispersed.

[0126] Specifically, during the training process, optimization algorithms such as gradient descent may be used to adjust the weights and parameters of the neural network to reduce the error. In this way, the neural network can better capture the common features of samples in the positive debugging sample set and distinguish different samples in the negative debugging sample set.

[0127] Through this process, the training effect of the embedding mapping neural network is improved because it can better process samples corresponding to past node sequences with different intersection characteristics. This training method helps to improve the diagnostic accuracy of the neural network for the power status, so that it can more accurately identify and distinguish different power performance modes in practical applications. In step S6553a, the computer device adjusts the internal parameters of the embedding mapping neural network, such as weights, biases, etc., and possible network structures, such as increasing or decreasing the number of hidden layers, the number of neurons, etc., according to this error. The goal of debugging is to minimize this error so that the neural network can more accurately capture and distinguish normal and abnormal power performance modes.

[0128] Alternatively, in some other implementations, step S655, debugging the embedding mapping neural network based on the commonality measurement results corresponding to the active debugging sample set and the commonality measurement results corresponding to the passive debugging sample set, includes:

[0129] Step S6551b: Analyze the statistical dispersion of the commonality measurement results corresponding to multiple sets of active debugging samples to obtain a first analysis result.

[0130] Step S6552b: Analyze the statistical dispersion of the commonality measurement results corresponding to the multiple passive debugging sample sets to obtain a second analysis result.

[0131] Among them, in step S6551b, the computer device pays attention to the commonality measurement results of multiple sets of active debugging samples. The active debugging sample set contains debugging samples of the target power supply on multiple past node sequences with intersections. The commonality measurement results reflect the similarity and correlation degree of these samples in the embedding space. The computer device analyzes the statistical dispersion of these commonality measurement results, that is, their range and distribution between different samples. This analysis can be achieved by calculating statistical indicators such as variance and standard deviation to quantify the dispersion of the commonality measurement results. The first analysis result provides specific information about the distribution characteristics of the active debugging sample set. If the commonality measurement results show a small statistical dispersion, it means that the active debugging samples have a more consistent representation in the embedding space, and the neural network performs well in capturing the common characteristics of these samples. On the contrary, if the dispersion is large, it may indicate that the neural network has difficulties in processing certain past node sequences with intersections and needs further optimization.

[0132] Next, in step S6552b, the computer device analyzes the commonality measurement results of multiple sets of negative debugging samples. The negative debugging sample set contains debugging samples of the target power supply on multiple past node sequences that do not have intersections. Similarly, the computer analyzes the statistical dispersion of these commonality measurement results to understand the distribution of negative debugging samples in the embedding space. The second analysis result reveals the distribution characteristics of the negative debugging sample set. Compared with the positive debugging sample set, the representation of negative debugging samples in the embedding space should have greater differences because there is no intersection between their past node sequences. If the dispersion of the commonality measurement results is small, this may mean that the neural network has difficulty in distinguishing these different examples. On the contrary, a larger dispersion indicates that the neural network can effectively distinguish negative debugging samples and capture the differences between them.

[0133] Through the analysis of steps S6551b and S6552b, the computer device obtains important insights into the distribution characteristics of the positive debugging sample set and the negative debugging sample set in the embedding space. These analysis results provide guidance for subsequent model optimization, helping the computer device to adjust the parameters and structure of the embedding mapping neural network to improve its diagnostic accuracy for the power state.

[0134] During the optimization process, the computer device may use various techniques, such as back-propagation algorithms, gradient descent optimizers, etc., to adjust the weights and parameters of the neural network to improve its representation ability and diagnostic performance in the embedding space.

[0135] Step S6553b: Based on the first analysis result and the second analysis result, debug the embedding mapping neural network.

[0136] As an implementation method, the first analysis result is the variance or standard deviation of the commonality measurement results corresponding to multiple sets of positive debugging samples, and the second analysis result is the variance or standard deviation of the commonality measurement results corresponding to multiple sets of negative debugging samples. The following are their calculation formulas:

[0137] Calculation formula for the first analysis result (variance or standard deviation):

[0138] For the active debugging sample set, let its commonality measurement result be ,in is the number of active debugging samples.

[0139] If the first analysis result is a variance ,but:

[0140]

[0141] in, is the mean of the commonality measurement results, that is:

[0142]

[0143] If the first analysis result is the standard deviation ,but:

[0144]

[0145] The calculation formula for the second analysis result (variance or standard deviation):

[0146] For the set of negative debugging samples, let its commonality measurement result be ,in is the number of negative debugging samples.

[0147] If the second analysis result is the variance ,but:

[0148]

[0149] in, is the mean of the commonality measurement results of the passive debugging sample set, that is:

[0150]

[0151] If the second analysis result is the standard deviation ,but:

[0152]

[0153] Based on this, step S6553b, based on the first analysis result and the second analysis result, debugging the embedding mapping neural network may specifically include:

[0154] Step S65531b, based on the sum of the first analysis result and the second analysis result, obtain the first network cost of the embedded mapping neural network, the first network cost is consistent with the changing trend of the first analysis result, and the first network cost is consistent with the changing trend of the second analysis result.

[0155] Specifically, the first analysis result and the second analysis result are the variance or standard deviation of the commonality measurement results of multiple sets of positive debugging samples and multiple sets of negative debugging samples, respectively. These two results provide feedback on the ability of the embedding mapping neural network to represent samples in the embedding space. A larger variance or standard deviation may mean that the neural network has difficulty in capturing the common characteristics of samples or distinguishing different samples.

[0156] In step S65531b, the computer device calculates the sum of the first analysis result and the second analysis result. This sum is used as the first network cost of the embedding mapping neural network. Network cost is a concept in machine learning that measures the performance of the model on the training data. Generally, the smaller the network cost, the better the performance of the model. The first network cost is consistent with the change trend of the first analysis result and the second analysis result, which means that if the variance or standard deviation of the commonality measurement result increases, the first network cost will also increase accordingly. This indicates that the representation ability of the neural network in the embedding space has decreased and needs to be adjusted. The computer device adjusts the network configuration variables of the embedding mapping neural network according to the first network cost. The network configuration variables are parameters that control the structure and behavior of the neural network, such as weights, biases, learning rates, etc. By adjusting these variables, the computer device can optimize the representation ability of the neural network in the embedding space, thereby improving its diagnostic accuracy for the power supply status. In addition, the calculation method of the first network cost is not limited to summation. In practical applications, the first analysis result and the second analysis result can also be weighted summed, maximum or minimum values ​​can be taken as needed to obtain a more representative network cost. These operations can be adjusted according to specific application scenarios and requirements.

[0157] Step S65532b, adjusting the network configuration variables of the embedded mapping neural network based on the first network cost.

[0158] Alternatively, in another embodiment, step S6553b, based on the first analysis result and the second analysis result, debugging the embedding mapping neural network includes:

[0159] Step S65531b', calculating the mean of the commonality measurement results corresponding to the multiple active debugging sample sets to obtain a first commonality measurement result mean.

[0160] Step S65532b', calculating the mean of the commonality measurement results corresponding to the multiple sets of passive debugging samples to obtain a second commonality measurement result mean.

[0161] In this embodiment, step S65531b' requires the computer device to calculate the average of the commonality measurement results corresponding to the multiple active debugging sample sets to obtain the first commonality measurement result average. The average calculation is to sum up the measurement results of these individual samples and then divide them by the number of samples to obtain an average value that can represent the overall characteristics of the entire active debugging sample set.

[0162] Step S65533b', subtract the first commonality measurement result mean from the second commonality measurement result mean to obtain the commonality measurement result mean difference.

[0163] Step S65534b', subtract the commonality measurement result mean difference from the preset commonality measurement result threshold to obtain a target value, and perform weighted summation on the target value, the first analysis result, and the second analysis result to obtain a second network cost.

[0164] Assume that the preset commonality measurement result threshold is T (for example, 1), and the commonality measurement result mean difference is (i.e. the mean of the first commonality measurement result minus the mean of the second commonality measurement result), the first analysis result is The second analysis result is The weights of the weighted sum are , , The second network cost is The calculation formula can be expressed as:

[0165]

[0166] , , are weight coefficients determined according to specific needs and experience, which reflect the relative importance of each component in the cost of the second network. These weight coefficients are usually non-negative, and their sum may be equal to 1 (if the cost needs to be normalized).

[0167] Step S65535b', adjusting the network configuration variables of the embedded mapping neural network based on the second network cost.

[0168] In one embodiment, in step S50, the inference power supply monitoring data stream is obtained by using a debugged data stream inference neural network. The method provided in the embodiment of the present application also includes a data stream inference neural network debugging step, and the data stream inference neural network debugging step may specifically include:

[0169] Step T1: Obtain a power monitoring data flow debugging sample; the power monitoring data flow debugging sample includes the power monitoring data of each past node of the target power in the first past node sequence, and each power monitoring data in the power monitoring data flow debugging sample is arranged according to the node order of each past node.

[0170] Step T2: Integrate each power monitoring data in the power monitoring data stream debugging sample with the corresponding power performance characterization vector of prior information into integrated power monitoring data to generate a debugging sample integrated power monitoring data stream.

[0171] Step T3: embed and map the debugging sample integrated power monitoring data stream based on the debugged embedding mapping neural network to generate a debugging sample embedding mapping vector stream.

[0172] Step T4: embed the debugging sample into the mapping vector stream and load it into the data flow inference neural network to be debugged, infer the power monitoring data of each past node in the second past node sequence of the target power supply, and generate an inference data stream.

[0173] Step T5: Debug the data flow inference neural network based on the inference data flow and the power monitoring data flow mark.

[0174] The first past node sequence and the second past node sequence are different sequences, and the second past node sequence occurs after the first past node sequence. Such prediction capabilities are of great significance for timely detection of power failures, optimization of energy management, and ensuring the stable operation of power systems. When the computer device performs the reasoning task of the power monitoring data stream, it uses a debugged data stream reasoning neural network. The key to this reasoning process is how to effectively debug this neural network to ensure that it can accurately process and analyze the power monitoring data. In step T1, the computer device obtains the debugging samples of the power monitoring data stream. These samples contain the power monitoring data of each node of the target power supply in the first past node sequence. Importantly, these data are arranged in the order of the nodes to which they belong, which ensures the timing and relevance of the data and provides a basis for subsequent neural network processing. In step T2, the computer device processes each power monitoring data, integrates these data with the corresponding power performance characterization vector of the prior information, and forms integrated power monitoring data. The purpose of this is to enhance the feature expression of the data so that the neural network can more easily capture the key information of the power performance. The integrated data constitutes the debugged sample integrated power monitoring data stream, which is ready for the subsequent embedding mapping operation. In step T3, the computer device processes the integrated power monitoring data stream using the debugged embedding mapping neural network. The role of this embedding mapping neural network is to map the original data into a low-dimensional space while retaining the main features of the data. After embedding mapping processing, the debug sample integrated power monitoring data stream is converted into a debug sample embedding mapping vector stream, which is more suitable as the input of the data stream reasoning neural network. In step T4, the computer device loads the debug sample embedding mapping vector stream into the data stream reasoning neural network to be debugged. The task of this reasoning neural network is to predict the power monitoring data of the target power supply at the future node (i.e., the second past node sequence) based on the known past node data. Through this reasoning process, the computer device generates a reasoning data stream, which reflects the prediction result of the neural network on the future state of the target power supply. In step T5, the computer device debugs the data stream reasoning neural network based on the generated reasoning data stream and the real label (i.e., the actually observed data) of the power monitoring data stream. This debugging process usually includes comparing the differences between the reasoning results and the real data, and adjusting the parameters and structure of the neural network according to these differences to improve the accuracy of its prediction.

[0175] Based on the foregoing embodiments, the embodiments of the present application provide a power supply diagnostic device based on big data. The units included in the device and the modules included in each unit can be implemented by a processor in a computer device; of course, they can also be implemented by a specific logic circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.

[0176] Figure 2 A schematic diagram of the composition structure of a power supply diagnosis device based on big data provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the power supply diagnosis device 200 based on big data includes:

[0177] The monitoring data acquisition module 210 is used to acquire a past power monitoring data stream of a target power source; the past power monitoring data stream includes power monitoring data of each past node of the target power source in a plurality of past nodes, and each of the power monitoring data is arranged according to the node order of the respective past node;

[0178] A priori information acquisition module 220 is used to acquire the priori information corresponding to each power monitoring data in the past power monitoring data stream, wherein the priori information is label information characterizing the performance of the target power supply, and the node describing the performance in the label information is located after the past node of the power monitoring data corresponding to the priori information;

[0179] A performance feature mining module 230, configured to mine a power performance characterization vector of each of the prior information to obtain a power performance characterization vector of each of the prior information;

[0180] The performance characteristic integration module 240 is used to integrate each power monitoring data in the past power monitoring data stream with the corresponding power performance characterization vector of the prior information into integrated power monitoring data to obtain the past integrated power monitoring data stream;

[0181] The power data inference module 250 is used to infer the power monitoring data of each monitoring node of the target power supply in multiple uninterrupted monitoring nodes based on the past integrated power monitoring data stream to obtain an inferred power monitoring data stream; the monitoring node is located after each of the past nodes.

[0182] The description of the above device embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. In some embodiments, the functions or modules included in the device provided in the embodiment of the present application can be used to execute the method described in the above method embodiment. For technical details not disclosed in the device embodiment of the present application, please refer to the description of the method embodiment of the present application for understanding.

[0183] It should be noted that in the embodiments of the present application, if the above-mentioned power supply diagnosis method based on big data is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application can be essentially or partly reflected in the form of a software product that contributes to the relevant technology. The software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a disk or an optical disk. In this way, the embodiments of the present application are not limited to any specific hardware, software or firmware, or any combination of hardware, software, and firmware.

[0184] An embodiment of the present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, some or all of the steps in the above method are implemented.

[0185] The embodiment of the present application provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, some or all of the steps in the above method are implemented. The computer-readable storage medium can be transient or non-transient.

[0186] An embodiment of the present application provides a computer program, including a computer-readable code. When the computer-readable code is run in a computer device, a processor in the computer device executes some or all of the steps for implementing the above method.

[0187] The embodiment of the present application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and when the computer program is read and executed by a computer, some or all of the steps in the above method are implemented. The computer program product can be implemented specifically by hardware, software or a combination thereof. In some embodiments, the computer program product is specifically embodied as a computer storage medium, and in other embodiments, the computer program product is specifically embodied as a software product, such as a software development kit (SDK) and the like.

[0188] It should be noted here that the description of the various embodiments above tends to emphasize the differences between the various embodiments, and the same or similar aspects can be referenced to each other. The description of the above device, storage medium, computer program and computer program product embodiments is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the embodiments of the device, storage medium, computer program and computer program product of this application, please refer to the description of the method embodiment of this application for understanding.

[0189] The embodiment of the present application also provides a computer device for executing the method, and the hardware entity of the computer device includes: a processor and a memory, wherein the memory stores a computer program that can be run on the processor, and the steps in the method of any of the above embodiments are implemented when the processor executes the program. The memory stores a computer program that can be run on the processor, and the memory is configured to store instructions and applications executable by the processor. It can also cache data to be processed or processed by the processor and each module in the computer device (for example, image data, audio data, voice communication data, and video communication data), which can be implemented through flash memory (FLASH) or random access memory (Random Access Memory, RAM). When the processor executes the program, the steps of any of the above-mentioned power supply diagnosis methods based on big data are implemented. The processor generally controls the overall operation of the computer device.

[0190] The embodiment of the present application provides a computer storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the power supply diagnosis method based on big data in any of the above embodiments. It should be pointed out here that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium and device embodiments of the present application, please refer to the description of the method embodiments of the present application for understanding. The above processor can be at least one of a target application integrated circuit (Application Specific Integrated Circuit, ASIC), a digital signal processor (Digital Signal Processor, DSP), a digital signal processing device (Digital Signal Processing Device, DSPD), a programmable logic device (Programmable Logic Device, PLD), a field programmable gate array (Field Programmable Gate Array, FPGA), a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, and a microprocessor. It can be understood that the electronic device that implements the above processor function can also be other, and the embodiment of the present application is not specifically limited.

[0191] The above-mentioned computer storage medium / memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory (Flash Memory), a magnetic surface memory, an optical disk, or a compact disc read-only memory (CD-ROM) and the like; it can also be various terminals including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

Claims

1. A power supply diagnosis method based on big data, characterized in that: The method comprises: Acquire a past power monitoring data stream of a target power source; the past power monitoring data stream includes power monitoring data of each past node of the target power source in a plurality of past nodes, and each of the power monitoring data is arranged according to the node order of the respective past node; Acquire prior information corresponding to each power monitoring data in the past power monitoring data stream, wherein the prior information is label information characterizing the performance of the target power supply, and the node describing the performance in the label information is located after the past node of the power monitoring data corresponding to the prior information; wherein the power supply performance refers to the state description of the power supply at the node; Mining a power performance characterization vector of each of the prior information to obtain a power performance characterization vector of each of the prior information; Integrate each power monitoring data in the past power monitoring data stream with the corresponding power performance characterization vector of prior information into integrated power monitoring data to obtain the past integrated power monitoring data stream; Based on the past integrated power monitoring data stream, the power monitoring data of each monitoring node of the target power supply in a plurality of uninterrupted monitoring nodes is inferred to obtain an inferred power monitoring data stream; the monitoring node is located after each of the past nodes.

2. The method according to claim 1, characterized in that The step of integrating each power monitoring data in the past power monitoring data stream with the corresponding power performance characterization vector of prior information into integrated power monitoring data to obtain the past integrated power monitoring data stream includes: For each power monitoring data in the past power monitoring data stream, combining the power monitoring data with a power performance characterization vector of corresponding prior information to obtain integrated power monitoring data corresponding to each power monitoring data; According to the arrangement position of the power monitoring data in the past power monitoring data stream, the integrated power monitoring data corresponding to each power monitoring data is arranged to generate a past integrated power monitoring data stream.

3. The method according to claim 2, characterized in that The method further comprises: Arranging the power performance characterization vectors of the prior information corresponding to each of the power monitoring data according to the arrangement positions of the power monitoring data in the past power monitoring data stream, to generate a power performance characterization vector stream; For each power monitoring data in the past power monitoring data stream, the power monitoring data is combined with a power performance characterization vector of corresponding prior information to obtain integrated power monitoring data corresponding to each power monitoring data, including: The power monitoring data and the power performance characterization vector at the same position in the past power monitoring data stream and the power performance characterization vector stream are combined to obtain integrated power monitoring data corresponding to each power monitoring data.

4. The method according to claim 1, characterized in that: The inference of the power monitoring data of each monitoring node of the target power supply in a plurality of uninterrupted monitoring nodes based on the past integrated power monitoring data stream to obtain the inferred power monitoring data stream includes: According to the power performance characterization vector included in each integrated power monitoring data in the past integrated power monitoring data stream, the past integrated power monitoring data stream is segmented to obtain a plurality of data segments; The power performance characterization vectors included in the integrated power monitoring data in the same data segment are the same, and the power performance characterization vectors included in the integrated power monitoring data in the two preceding and succeeding data segments are different; Based on the multiple data segments, the power monitoring data of the target power source at each monitoring node in the multiple uninterrupted monitoring nodes is inferred to obtain an inferred power monitoring data stream.

5. The method according to claim 4, characterized in that The inference of the power supply monitoring data of each monitoring node of the target power supply in the plurality of uninterrupted monitoring nodes based on the plurality of data segments to obtain an inference power supply monitoring data stream includes: Performing embedding mapping on the multiple data segments respectively to obtain an embedding mapping vector corresponding to each of the data segments; Based on the embedded mapping vectors corresponding to each of the data segments, generating an embedded mapping vector stream corresponding to the past power monitoring data stream; Based on the embedded mapping vector flow corresponding to the past power monitoring data flow, the power monitoring data of each monitoring node of the target power supply in a plurality of uninterrupted monitoring nodes is inferred to obtain an inferred power monitoring data flow.

6. The method according to claim 5, characterized in that The embedding mapping vector is generated by using a debugged embedding mapping neural network. The method further includes a debugging process of the embedding mapping neural network. The debugging process of the embedding mapping neural network includes: Acquire an active debugging sample set, the active debugging sample set including debugging samples corresponding to the target power supply in a plurality of past node sequences having intersections, the debugging samples corresponding to the past node sequences including power monitoring data of each past node of the target power supply in the past node sequences, and the power monitoring data in the debugging samples are arranged according to the order of the respective nodes; Acquire a passive debugging sample set, wherein the passive debugging sample set includes debugging samples corresponding to a plurality of past node sequences having no intersection respectively of the target power source; For each debugging sample in the active debugging sample set, each power monitoring data in the debugging sample is respectively integrated with the power performance characterization vector of the corresponding prior information into integrated power monitoring data, and an integrated power monitoring data stream corresponding to each debugging sample in the active debugging sample set is generated; For each debugging sample in the passive debugging sample set, each power monitoring data in the debugging sample is respectively integrated with the power performance characterization vector of the corresponding prior information into integrated power monitoring data, and an integrated power monitoring data stream corresponding to each debugging sample in the passive debugging sample set is generated; Loading the integrated power monitoring data streams corresponding to each debugging sample in the active debugging sample set into the embedding mapping neural network for embedding and mapping to obtain a plurality of embedding mapping vectors, and obtaining the embedding mapping vector streams corresponding to each debugging sample in the active debugging sample set; Loading the integrated power monitoring data streams corresponding to each debugging sample in the passive debugging sample set into the embedding mapping neural network for embedding mapping to obtain a plurality of embedding mapping vectors, and obtaining the embedding mapping vector streams corresponding to each debugging sample in the passive debugging sample set; Determine the commonality measurement results between the embedded mapping vector flows respectively corresponding to each debugging sample in the active debugging sample set, and obtain the commonality measurement results corresponding to the active debugging sample set; Determine the commonality measurement results between the embedded mapping vector flows respectively corresponding to each debugging sample in the passive debugging sample set, and obtain the commonality measurement results corresponding to the passive debugging sample set; The embedding mapping neural network is debugged based on the commonality measurement results corresponding to the positive debugging sample set and the commonality measurement results corresponding to the negative debugging sample set.

7. The method according to claim 6, characterized in that The step of debugging the embedding mapping neural network based on the commonality measurement results corresponding to the active debugging sample set and the commonality measurement results corresponding to the passive debugging sample set includes: Performing a first numerical calculation on the commonality measurement results respectively corresponding to the plurality of active debugging sample sets to obtain a first commonality measurement result; Performing the first numerical calculation on the commonality measurement results respectively corresponding to the plurality of passive debugging sample sets to obtain a second commonality measurement result; Debugging the embedding mapping neural network based on an error between the first commonality measurement result and the second commonality measurement result; Alternatively, debugging the embedding mapping neural network based on the commonality measurement results corresponding to the active debugging sample set and the commonality measurement results corresponding to the passive debugging sample set includes: Analyze the statistical dispersion of the commonality measurement results corresponding to the multiple active debugging sample sets to obtain a first analysis result; Analyzing the statistical dispersion of the commonality measurement results corresponding to the plurality of negative debugging sample sets, to obtain a second analysis result; Based on the first analysis result and the second analysis result, the embedding mapping neural network is debugged.

8. The method according to claim 7, characterized in that The first analysis result is the variance or standard deviation of the commonality measurement results corresponding to the multiple sets of positive debugging samples; wherein the second analysis result is the variance or standard deviation of the commonality measurement results corresponding to the multiple sets of negative debugging samples; The debugging of the embedding mapping neural network based on the first analysis result and the second analysis result includes: Based on the sum of the first analysis result and the second analysis result, a first network cost of the embedding mapping neural network is obtained, wherein the first network cost is consistent with a change trend of the first analysis result, and the first network cost is consistent with a change trend of the second analysis result; adjusting a network configuration variable of the embedded mapping neural network based on the first network cost; or, The debugging of the embedding mapping neural network based on the first analysis result and the second analysis result includes: Calculating the average of the commonality measurement results corresponding to the multiple active debugging sample sets to obtain a first commonality measurement result average; Calculating the average of the commonality measurement results corresponding to the plurality of passive debugging sample sets to obtain a second commonality measurement result average; Subtract the mean of the first commonality measurement result from the mean of the second commonality measurement result to obtain a commonality measurement result mean difference; Subtracting the commonality measurement result mean difference from a preset commonality measurement result threshold to obtain a target value, and performing weighted summation of the target value, the first analysis result, and the second analysis result to obtain a second network cost; A network configuration variable of the embedding mapping neural network is adjusted based on the second network cost.

9. The method according to claim 1, characterized in that: The inference power supply monitoring data stream is obtained by using a debugged data stream inference neural network. The method further includes a data stream inference neural network debugging step, and the data stream inference neural network debugging step includes: Obtain a power monitoring data stream debugging sample; the power monitoring data stream debugging sample includes power monitoring data of each past node of the target power supply in the first past node sequence, and each power monitoring data in the power monitoring data stream debugging sample is arranged according to the node sequence of each past node; Integrate each power monitoring data in the power monitoring data stream debugging sample with the corresponding power performance characterization vector of prior information into integrated power monitoring data to generate a debugging sample integrated power monitoring data stream; Based on the debugged embedded mapping neural network, embedding and mapping the debug sample integrated power monitoring data stream is performed to generate a debug sample embedded mapping vector stream; The debugging sample embedding mapping vector stream is loaded into the data flow inference neural network to be debugged, and the power supply monitoring data of each past node of the target power supply in the second past node sequence is inferred to generate an inference data stream; The data flow inference neural network is debugged based on the inference data flow and the power monitoring data flow marker.

10. A power supply diagnostic device based on big data, characterized in that: include: A monitoring data acquisition module, used to acquire past power monitoring data streams of a target power source; The past power monitoring data stream includes power monitoring data of the target power source at each past node among a plurality of past nodes, and each of the power monitoring data is arranged according to the node sequence of the respective past node; A priori information acquisition module, used to acquire the priori information corresponding to each power monitoring data in the past power monitoring data stream, wherein the priori information is label information characterizing the performance of the target power supply, and the node describing the performance in the label information is located after the past node of the power monitoring data corresponding to the priori information; A performance feature mining module, used for mining a power performance characterization vector of each of the prior information to obtain a power performance characterization vector of each of the prior information; A performance characteristic integration module, used to integrate each power monitoring data in the past power monitoring data stream with the corresponding power performance characterization vector of prior information into integrated power monitoring data to obtain the past integrated power monitoring data stream; A power data inference module, used to infer the power monitoring data of each monitoring node of the target power supply in a plurality of uninterrupted monitoring nodes based on the past integrated power monitoring data stream, to obtain an inferred power monitoring data stream; The monitoring node is located behind each of the passing nodes.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps in the method according to any one of claims 1 to 9 are implemented.

Images