An AI-based power supply anomaly diagnosis and alarm method and system

Abstract

The present application provides a method and system for power anomaly diagnosis and alarm based on artificial intelligence, which matches the real-time power data with the pre-prepared abnormal power data to determine whether there is an abnormal situation. By matching with the data corresponding to a more comprehensive abnormal situation, the early warning of the abnormal power data in a complex environment can be improved. When inferring the degree of coincidence between the real-time power data and the abnormal power data, the characterization information of the data items in the abnormal power data, the characterization information of the data items in the environmental parameters, and the characterization information of the data items in the operating parameters are analyzed, and the implicit relationship between the characterization information of the data items in the abnormal power data and the characterization information of the data items in the environmental parameters and the characterization information of the data items in the operating parameters is considered in two dimensions of the operating parameters and the environmental parameters, so as to infer the degree of coincidence between the real-time power data and the abnormal power data, and improve the reliability and accuracy of the inferred degree of coincidence.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular, to a power supply anomaly diagnosis and alarm method and system based on artificial intelligence. Background Art

[0002] The power supply is a key component of various computer systems and systems. It is responsible for storing electrical energy in other forms and converting it into electrical energy to supply the system when needed. Therefore, the stability and reliability of the power supply are crucial for the normal operation of the entire system. Power supply anomalies can manifest in various forms, such as unstable output voltage, no output, or abnormal output voltage, etc. These anomalies may be caused by various reasons, including but not limited to excessive current, high voltage, component aging, poor contact, or equipment damage, etc. Traditional power supply anomaly diagnosis methods mainly rely on manual inspection, experience judgment, and simple threshold comparison. These methods often fail to detect problems in a timely and accurate manner when faced with complex or novel power supply anomalies, resulting in low diagnostic efficiency or even misdiagnosis. With the development of artificial intelligence technology, anomaly detection methods based on machine learning and deep learning have been introduced into power supply anomaly diagnosis. These methods can identify abnormal situations by learning a large amount of historical data, improving the accuracy and efficiency of diagnosis. Modern power supply anomaly diagnosis methods begin to comprehensively consider the internal relationships and mutual influences among multiple parameters and data streams. For example, in addition to considering the operating parameters of the power supply (such as voltage, current, etc.), environmental parameters (such as temperature, humidity, etc.) and the status information of the equipment will also be combined for comprehensive judgment. This multi-parameter fusion diagnosis method helps to improve the accuracy and reliability of diagnosis. With the development of technologies such as the Internet of Things, big data, and cloud computing, power supply anomaly diagnosis is developing towards the direction of intelligence and automation. How to ensure the accuracy of power supply anomaly diagnosis is a technical problem that needs to be solved. Summary of the Invention

[0003] The purpose of the present invention is to provide a power supply anomaly diagnosis and alarm method and system based on artificial intelligence. The technical solution of the embodiment of this application is realized as follows:

[0004] In a first aspect, the embodiment of this application provides a power supply anomaly diagnosis and alarm method based on artificial intelligence, and the method includes:

[0005] Obtain a preset power supply anomaly data stream and a real-time power supply monitoring data stream. The real-time power supply monitoring data stream includes an environmental parameter monitoring data stream and an operating parameter monitoring data stream of real-time power supply data. The preset power supply anomaly data stream represents data items in abnormal power supply data. The environmental parameter monitoring data stream represents data items in the environmental parameters of the real-time power supply data. The operating parameter monitoring data stream represents data items in the operating parameters of the real-time power supply data;

[0006] Extract characterization information from the preset power anomaly data stream, the environmental parameter monitoring data stream, and the operating parameter monitoring data stream respectively to obtain an abnormal power data characterization vector, an environmental parameter data characterization vector, and an operating parameter data characterization vector;

[0007] Interact the abnormal power data characterization vector, the environmental parameter data characterization vector, and the operating parameter data characterization vector to obtain an interaction data characterization vector;

[0008] Perform abnormal diagnosis on the interaction data characterization vector to obtain a power anomaly diagnosis result, where the power anomaly diagnosis result represents the degree of coincidence between the real-time power data and the abnormal power data;

[0009] When the power anomaly diagnosis result indicates that the degree of coincidence between the real-time power data and the abnormal power data reaches a preset requirement, perform a power anomaly alarm.

[0010] As an implementation manner, the power anomaly diagnosis result is obtained based on a coincidence degree evaluation network, and the coincidence degree evaluation network includes an abnormal data embedding mapping layer, an environmental parameter embedding mapping layer, and an operating parameter embedding mapping layer; the extracting characterization information from the preset power anomaly data stream, the environmental parameter monitoring data stream, and the operating parameter monitoring data stream respectively to obtain an abnormal power data characterization vector, an environmental parameter data characterization vector, and an operating parameter data characterization vector includes:

[0011] Extract characterization information from the preset power anomaly data stream according to the abnormal data embedding mapping layer to obtain the abnormal power data characterization vector;

[0012] Extract characterization information from the environmental parameter monitoring data stream according to the environmental parameter embedding mapping layer to obtain the environmental parameter data characterization vector;

[0013] Extract characterization information from the operating parameter monitoring data stream according to the operating parameter embedding mapping layer to obtain the operating parameter data characterization vector.

[0014] As an implementation manner, the extracting characterization information from the preset power anomaly data stream according to the abnormal data embedding mapping layer to obtain the abnormal power data characterization vector includes:

[0015] Determine the first power state characterization information and the first distribution characterization information of the preset power anomaly data stream according to the abnormal data embedding mapping layer, where the first power state characterization information represents the characterization information of multiple data items in the preset power anomaly data stream, and the first distribution characterization information represents the characterization information of the distribution of multiple data items in the preset power anomaly data stream;

[0016] Extract characterization information from the first power status characterization information and the first distribution characterization information to obtain the abnormal power data characterization vector.

[0017] As an implementation, the extracting characterization information from the first power status characterization information and the first distribution characterization information to obtain the abnormal power data characterization vector includes:

[0018] Extract characterization information from the first power status characterization information and the first distribution characterization information to obtain a first anchor characterization vector, a first result characterization vector, and a first search characterization vector;

[0019] Interact the first anchor characterization vector, the first result characterization vector, and the first search characterization vector to obtain a candidate abnormal power data characterization vector;

[0020] Perform multiple rounds of non-linear transformation on the candidate abnormal power data characterization vector to obtain the abnormal power data characterization vector.

[0021] As an implementation, the extracting characterization information from the environmental parameter monitoring data stream according to the environmental parameter embedding mapping layer to obtain the environmental parameter data characterization vector includes:

[0022] According to the environmental parameter embedding mapping layer, determine the second power status characterization information and the second distribution characterization information of the environmental parameter monitoring data stream, where the second power status characterization information represents the characterization information of multiple data items in the environmental parameter monitoring data stream, and the second distribution characterization information represents the characterization information of the distribution of multiple data items in the environmental parameter monitoring data stream;

[0023] Extract characterization information from the second power status characterization information and the second distribution characterization information to obtain the environmental parameter data characterization vector.

[0024] As an implementation, the extracting characterization information from the second power status characterization information and the second distribution characterization information to obtain the environmental parameter data characterization vector includes:

[0025] Extract characterization information from the second power status characterization information and the second distribution characterization information to obtain a second anchor characterization vector, a second result characterization vector, and a second search characterization vector;

[0026] Interact the second anchor characterization vector, the second result characterization vector, and the second search characterization vector to obtain a candidate environmental parameter data characterization vector;

[0027] Perform multiple rounds of non - linear transformation on the candidate environmental parameter data characterization vector to obtain the environmental parameter data characterization vector.

[0028] As an implementation manner, the obtaining of the operating parameter data characterization vector by extracting characterization information from the operating parameter monitoring data stream according to the operating parameter embedding mapping layer includes:

[0029] According to the operating parameter embedding mapping layer, determine the third power state characterization information and the third distribution characterization information of the operating parameter monitoring data stream. The third power state characterization information represents the characterization information of multiple data items in the operating parameter monitoring data stream, and the third distribution characterization information represents the characterization information of the distribution of multiple data items in the operating parameter monitoring data stream; extract characterization information from the third power state characterization information and the third distribution characterization information to obtain the operating parameter data characterization vector.

[0030] As an implementation manner, the extracting characterization information from the third power state characterization information and the third distribution characterization information to obtain the operating parameter data characterization vector includes:

[0031] Extract characterization information from the third power state characterization information and the third distribution characterization information to obtain a third anchor characterization vector, a third result characterization vector, and a third search characterization vector;

[0032] Interact the third anchor characterization vector, the third result characterization vector, and the third search characterization vector to obtain a candidate operating parameter data characterization vector;

[0033] Perform multiple rounds of non - linear transformation on the candidate operating parameter data characterization vector to obtain the operating parameter data characterization vector.

[0034] As an implementation manner, the fitness evaluation network is debugged through the following steps:

[0035] Obtain an example preset power anomaly data stream, an example real - time power monitoring data stream, and an actual power anomaly diagnosis result. The example real - time power monitoring data stream includes an example environmental parameter monitoring data stream and an example operating parameter monitoring data stream, and the actual power anomaly diagnosis result represents the actual fitness of the example real - time power data to the example abnormal power data;

[0036] According to the consistency evaluation network, perform characterization information extraction on the example preset power anomaly data stream, the example environmental parameter monitoring data stream, and the example operating parameter monitoring data stream respectively, to obtain an example abnormal power data characterization vector, an example environmental parameter data characterization vector, and an example operating parameter data characterization vector; perform interaction on the example abnormal power data characterization vector, the example environmental parameter data characterization vector, and the example operating parameter data characterization vector to obtain an example interaction data characterization vector; perform abnormal diagnosis on the example interaction data characterization vector to obtain an example power anomaly diagnosis result;

[0037] Debug the consistency evaluation network according to the example power anomaly diagnosis result and the actual power anomaly diagnosis result.

[0038] As an implementation manner, the environmental parameter embedding mapping layer is a debugged network layer, and the environmental parameter embedding mapping layer is obtained through debugging by the following steps:

[0039] Obtain a first environmental parameter monitoring data stream, mask the data item information in a local area of the first environmental parameter monitoring data stream to obtain a second environmental parameter monitoring data stream;

[0040] According to the environmental parameter embedding mapping layer, perform reasoning on the data item information in the masked area of the second environmental parameter monitoring data stream to obtain a second reasoning confidence level, where the second reasoning confidence level represents the confidence level that the data items in the masked area of the first environmental parameter monitoring data stream obtained by reasoning belong to each type of data item;

[0041] Debug the environmental parameter embedding mapping layer according to the second reasoning confidence level and the second actual confidence level, where the second actual confidence level represents the actual confidence level that the data items in the masked area of the first environmental parameter monitoring data stream belong to each type of data item;

[0042] The operating parameter embedding mapping layer is a debugged network layer, and the operating parameter embedding mapping layer is obtained through debugging by the following steps:

[0043] Obtain a first operating parameter monitoring data stream, mask the data item information in a local area of the first operating parameter monitoring data stream to obtain a second operating parameter monitoring data stream;

[0044] According to the operating parameter embedding mapping layer, perform reasoning on the data item information in the masked area of the second operating parameter monitoring data stream to obtain a third reasoning confidence level, where the third reasoning confidence level represents the confidence level that the data items in the masked area of the first operating parameter monitoring data stream obtained by reasoning belong to each type of data item;

[0045] Adjust the running parameter embedding mapping layer according to the third inference confidence level and the third actual confidence level, where the third actual confidence level represents the actual confidence level of the data items in the masked area of the first running parameter monitoring data stream belonging to each type of data item.

[0046] In a second aspect, the present application provides a computer system, including a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor executes the computer program, the method described above is implemented.

[0047] Advantages of the present application: The present application provides an artificial intelligence-based power anomaly diagnosis and alarm method and system, which matches real-time power data with pre-prepared abnormal power data to help monitor whether the implemented power data has abnormal conditions. By matching with data corresponding to more comprehensive abnormal situations, the early warning of abnormal power data in a complex environment can be improved. When inferring the matching degree between real-time power data and abnormal power data based on a neural network, the characterization information of the data items in the abnormal power data, the characterization information of the data items in the environmental parameters of the real-time power data, and the characterization information of the data items in the running parameters of the real-time power data are analyzed. The implicit relationship between the characterization information of the data items in the abnormal power data, the characterization information of the data items in the environmental parameters, and the characterization information of the data items in the running parameters is considered in two dimensions of running parameters and environmental parameters to infer the matching degree between real-time power data and abnormal power data, improving the reliability and accuracy of the inferred matching degree.

[0048] In the following description, other features will be partially stated. When examining the following content and the drawings, those skilled in the art will partially discover these features, or these features can be learned through production or application. Through practicing or using various aspects of the methods, tools, and combinations listed in the detailed examples described later, the features in the current application can be implemented and obtained. Description of the Drawings

[0049] The drawings here are incorporated into the specification and form a part of this specification. These drawings show embodiments that conform to the present application and are used together with the specification to illustrate the technical solutions of the present application.

[0050] Figure 1 It is a schematic diagram of the application scenario of the artificial intelligence-based power anomaly diagnosis and alarm method provided by the embodiment of the present application.

[0051] Figure 2 It is a flowchart of an artificial intelligence-based power anomaly diagnosis and alarm method provided by the embodiment of the present application.

[0052] Figure 3It is a schematic diagram of the composition of a computer system provided by an embodiment of the present application. Detailed implementation manners

[0053] In order to make the objectives, technical solutions, and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be construed as limitations on the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present application.

[0054] In the following descriptions, reference is made to "some embodiments", "as an implementation manner / scheme", "in an implementation manner". These describe subsets of all possible embodiments. However, it can be understood that "some embodiments", "as an implementation manner / scheme", "in an implementation manner" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0055] In the following descriptions, the terms "first / second / third" and other similar terms are only used to distinguish similar objects and do not represent a specific order for the objects. It can be understood that "first / second / third" can be interchanged with a specific order or sequence when permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0056] The power anomaly diagnosis and alarm method based on artificial intelligence provided by the embodiments of the present application can be executed by a computer system. The computer system can be various types of terminals such as a laptop computer, a tablet computer, a desktop computer, a mobile device (for example, a mobile phone, a portable music player, a personal digital assistant, a dedicated messaging device, a portable game device), etc., or can be implemented as a server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or can also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms.

[0057] Figure 1FIG. 0 is a schematic diagram of an application scenario of the power supply anomaly diagnosis and alarm method based on artificial intelligence provided by an embodiment of the present application. It includes a plurality of sensors 100, a network 200, and a computer system 300. Communication connection is achieved between the plurality of sensors 100 and the computer system 300 through the network 200. The sensor 100 is used to detect data of the target power supply and send the acquired data to the computer system 300 through the network 200. The computer system 300 is used to execute the method provided by the embodiment of the present application. Specifically, the embodiment of the present application provides a power supply anomaly diagnosis and alarm method based on artificial intelligence, as Figure 2 shown, this method includes:

[0058] Step S110, obtain a preset power supply anomaly data stream and a real-time power supply monitoring data stream. The real-time power supply monitoring data stream includes an environmental parameter monitoring data stream and an operating parameter monitoring data stream of real-time power supply data. The preset power supply anomaly data stream represents data items in abnormal power supply data. The environmental parameter monitoring data stream represents data items in the environmental parameters of real-time power supply data. The operating parameter monitoring data stream represents data items in the operating parameters of real-time power supply data.

[0059] In the embodiment of the present application, the preset power supply anomaly data stream is a set of pre-defined data, usually based on historical data or expert knowledge, used to represent the data characteristics that the power supply may exhibit in an abnormal state. These data streams contain various power supply data items considered to be abnormal. For example, in a data center environment, the preset power supply anomaly data stream may include a series of power supply temperature, voltage, and current values when historical faults occurred, and these values exceeded the normal operating range and were thus considered abnormal.

[0060] The real-time power supply monitoring data stream is a data stream of the real-time working state and environmental conditions of the power supply, used to reflect the current actual situation of the power supply. These data streams will be continuously updated to provide the latest information on power supply performance. For example, the real-time working voltage, current, temperature, etc. data of a power supply unit, and these data will be continuously collected and transmitted to the monitoring system to form a real-time power supply monitoring data stream.

[0061] The environmental parameter monitoring data stream is a part of the real-time power supply monitoring data stream, specifically focusing on the state of the environment around the power supply, such as data of environmental factors such as temperature and humidity. These factors may affect the performance and lifespan of the power supply. For example, temperature sensors in the data center will continuously record the temperature of the computer room and send these data to the monitoring system to form an environmental parameter monitoring data stream. If the temperature exceeds the recommended operating range of the device, an alarm may be triggered.

[0062] The operating parameter monitoring data stream is also part of the real-time power supply monitoring data stream, mainly focusing on the operating status of the power supply itself, including operating parameters such as voltage, current, and power factor. These data reflect the actual working efficiency and stability of the power supply. For example, the output voltage and current of the power supply are monitored in real time and recorded as part of the operating parameter monitoring data stream. If the output voltage of the power supply fluctuates abnormally, this may indicate a problem inside the power supply and further inspection is required.

[0063] A data item refers to a single, specific data element in a data stream. In the preset power supply anomaly data stream, environmental parameter monitoring data stream, and operating parameter monitoring data stream, each specific reading or measurement value is a data item.

[0064] Step S120: Extract characterization information from the preset power supply anomaly data stream, environmental parameter monitoring data stream, and operating parameter monitoring data stream respectively to obtain an abnormal power supply data characterization vector, an environmental parameter data characterization vector, and an operating parameter data characterization vector.

[0065] After obtaining the preset power supply anomaly data stream, environmental parameter monitoring data stream, and operating parameter monitoring data stream, extract characterization information from the preset power supply anomaly data stream to obtain an abnormal power supply data characterization vector, which represents the types of each data item and the numerical magnitudes of multiple data items in the abnormal power supply data. Extract characterization information from the environmental parameter monitoring data stream to obtain an environmental parameter data characterization vector, which represents the types of each data item and the numerical magnitudes of multiple data items in the environmental parameters of the real-time power supply data. Extract characterization information from the operating parameter monitoring data stream to obtain an operating parameter data characterization vector, which represents the types of each data item and the numerical magnitudes of multiple data items in the operating parameters of the real-time power supply data.

[0066] For example, for the preset power supply anomaly data stream, it contains records of power supply anomaly situations that have occurred in the past. The computer system analyzes these data and extracts the characteristics of the abnormal power supply data, such as abnormal voltage fluctuations, abnormal temperature increases, etc. These characteristics will be converted into an abnormal power supply data characterization vector. For example, an abnormal power supply data characterization vector may contain a set of numbers, each number representing a specific abnormal characteristic, such as [0.8, 0.3, 0.5], where 0.8 may represent the degree of abnormal voltage fluctuation, 0.3 represents the degree of abnormal temperature increase, and 0.5 represents the degree of some other abnormal characteristic.

[0067] For the environmental parameter monitoring data stream, it reflects the real-time status of the environment where the power supply device is located, such as temperature, humidity, etc. The computer extracts key environmental parameters from these data streams, such as the current temperature, humidity value, etc., and converts them into environmental parameter data characterization vectors. For example, an environmental parameter data characterization vector may be in the form of [25, 60], where 25 represents the current environmental temperature (in degrees Celsius) and 60 represents the current humidity percentage. For the operating parameter monitoring data stream, it records the real-time operating status of the power supply device, such as voltage, current, etc. The computer analyzes these data, extracts key operating parameters, such as input voltage, output current, etc., and converts them into operating parameter data characterization vectors. For example, an operating parameter data characterization vector may be in the form of [220, 5], where 220 represents the current input voltage (in volts) and 5 represents the current output current (in amperes).

[0068] Step S130, interact with the abnormal power data characterization vector, the environmental parameter data characterization vector, and the operating parameter data characterization vector to obtain an interactive data characterization vector.

[0069] After obtaining the abnormal power data characterization vector, the environmental parameter data characterization vector, and the operating parameter data characterization vector, interact with the abnormal power data characterization vector, the environmental parameter data characterization vector, and the operating parameter data characterization vector, that is, perform feature fusion, to obtain an interactive data characterization vector. The interactive data characterization vector contains the feature information of abnormal power data, environmental parameters, and operating parameters, making the feature information of the interactive data characterization vector richer.

[0070] Step S140, perform abnormal diagnosis on the interactive data characterization vector to obtain a power supply abnormal diagnosis result, and the power supply abnormal diagnosis result represents the degree of coincidence between the real-time power data and the abnormal power data.

[0071] Since the obtained interactive data characterization vector includes the features of abnormal power data, environmental parameters, and operating parameters, and the abnormal power data, environmental parameters, and operating parameters contain the core information required for the degree of coincidence of the characterization data matching, the power supply abnormal diagnosis result of the abnormal power data and the real-time power data can be obtained based on the abnormal diagnosis of the interactive data characterization vector. Among them, the process of performing abnormal diagnosis on the interactive data characterization vector is a process of inferring the degree of coincidence based on the interactive data characterization vector. In other words, regression is performed based on this interactive data characterization vector to obtain the power supply abnormal diagnosis result of the abnormal power data and the real-time power data.

[0072] When the above embodiments of the present application infer the degree of coincidence between real-time power data and abnormal power data, they comprehensively consider the characterization information of data items in the abnormal power data, the characterization information of data items in the environmental parameters of the real-time power data, and the characterization information of data items in the operating parameters of the real-time power data. From the two dimensions of operating parameters and environmental parameters, the implicit relationship between the characterization information of data items in the abnormal power data and the characterization information of data items in the environmental parameters and the characterization information of data items in the operating parameters is analyzed, so as to infer the degree of coincidence between the real-time power data and the abnormal power data. Not only is the analysis dimension comprehensive, but the two dimensions of operating parameters and environmental parameters are used as the matching directions, increasing the accuracy of the inferred degree of coincidence.

[0073] In one embodiment, the present application provides a coincidence degree evaluation network, which is used to infer the coincidence degree between real-time power data and abnormal power data. As an example of the structure of this coincidence degree evaluation network, the coincidence degree evaluation network may include: an abnormal data embedding mapping layer, an environmental parameter embedding mapping layer, an operating parameter embedding mapping layer, an interaction layer, and a dense layer.

[0074] Among them, the abnormal data embedding mapping layer, the environmental parameter embedding mapping layer, and the operating parameter embedding mapping layer are respectively connected to the interaction layer, and the interaction layer is connected to the dense layer. The abnormal data embedding mapping layer is used to extract the feature information of the preset power abnormal data stream, the environmental parameter embedding mapping layer is used to extract the feature information of the environmental parameter monitoring data stream, and the operating parameter embedding mapping layer is used to extract the feature information of the operating parameter monitoring data stream. The interaction layer is used to interact the feature information extracted by the abnormal data embedding mapping layer, the environmental parameter embedding mapping layer, and the operating parameter embedding mapping layer (such as fusion methods such as addition, splicing, and stacking). The dense layer is used to perform abnormal diagnosis on the feature information to infer the power abnormal diagnosis result. The dense layer (Densely Connected layer) is also called the fully connected layer. The process of abnormal diagnosis and the process of full connection include feature integration, dimensionality reduction, or conversion to a more specific representation for subsequent classification, regression, or other tasks. In the embodiments of the present application, the dense layer is used to convert the interacted features into one or more numerical values, and these numerical values represent the degree of coincidence of the real-time power data with the abnormal power data. If a specific degree of coincidence value is predicted (regression task), then the output of the dense layer can be a continuous value representing the size of the degree of coincidence. If it is only to determine whether the real-time power data and the abnormal power data have a high degree of coincidence or a low degree of coincidence (classification task), then a softmax layer can be set after the dense layer to convert the output into a probability distribution for classification.

[0075] As an implementation design, the interaction layer may include a first filter, a second filter, and a third filter. Filters and network components that use convolution kernels for filtering can be referred to as convolutional layers. The first filter is connected to the abnormal data embedding mapping layer. The first filter is used to filter (i.e., convolve) the abnormal power data characterization vector output by the abnormal data embedding mapping layer to obtain a deep abnormal power data characterization vector; the second filter is connected to the environmental parameter embedding mapping layer. The second filter is used to filter the environmental parameter data characterization vector output by the environmental parameter embedding mapping layer to obtain a deep environmental parameter data characterization vector; the third filter is connected to the operating parameter embedding mapping layer. The third filter is used to filter the operating parameter data characterization vector output by the operating parameter embedding mapping layer to obtain a deep operating parameter data characterization vector.

[0076] For example, the execution data of the abnormal data embedding mapping layer is a preset power abnormal data stream, the execution data of the environmental parameter embedding mapping layer is an environmental parameter monitoring data stream, and the execution data of the operating parameter embedding mapping layer is an operating parameter monitoring data stream. The first filter, the second filter, and the third filter respectively output the deep abnormal power data characterization vector, environmental parameter data characterization vector, and operating parameter data characterization vector obtained by filtering, and then interact the deep abnormal power data characterization vector, environmental parameter data characterization vector, and operating parameter data characterization vector to obtain an interaction data characterization vector. The execution data of the dense layer is the interaction data characterization vector, and the output of the dense layer is the power abnormal diagnosis result.

[0077] Step S150, when the power abnormal diagnosis result indicates that the degree of coincidence between the real-time power data and the abnormal power data reaches a preset requirement, perform a power abnormal alarm.

[0078] It can be understood that the abnormal power data prepared in advance for the coincidence comparison in the embodiments of the present application may include multiple. When the coincidence comparison in one time does not reach the preset requirement (such as the coincidence value is less than the preset value, or it shows non-coincidence), the remaining abnormal power data is sequentially used to determine the coincidence with the real-time power data, and when the abnormal power data with the coincidence reaching the preset requirement is determined, a power abnormal alarm is performed. The warning method is, for example, to execute the warning strategy corresponding to the abnormal power data that reaches the preset requirement, such as broadcasting the abnormal content, warning in a preset manner, and so on.

[0079] As described above, the coincidence evaluation network can infer the power abnormal diagnosis result based on it. Among them, the coincidence evaluation network includes: an abnormal data embedding mapping layer, an environmental parameter embedding mapping layer, an operating parameter embedding mapping layer, an interaction layer, and a dense layer. Then, the method provided by the embodiments of the present application includes:

[0080] Step S210: Obtain a preset power anomaly data stream and a real-time power monitoring data stream. The real-time power monitoring data stream includes an environmental parameter monitoring data stream and an operating parameter monitoring data stream of real-time power data. The preset power anomaly data stream represents data items in the abnormal power data. The environmental parameter monitoring data stream represents data items in the environmental parameters of the real-time power data. The operating parameter monitoring data stream represents data items in the operating parameters of the real-time power data.

[0081] The process of step S210 can refer to the above step S110.

[0082] Step S220: According to the abnormal data embedding mapping layer in the conformity evaluation network, extract characterization information from the preset power anomaly data stream to obtain an abnormal power data characterization vector.

[0083] Input the preset power anomaly data stream into the abnormal data embedding mapping layer in the conformity evaluation network. The abnormal data embedding mapping layer extracts characterization information from the preset power anomaly data stream and outputs an abnormal power data characterization vector.

[0084] As an implementation design, the network architecture of the abnormal data embedding mapping layer can be a convolutional neural network (CNN), a recurrent neural network (RNN), and its variants, such as a long short-term memory network (LSTM) or a gated recurrent unit (GRU). These network structures are suitable for processing sequence data, can capture the temporal dependence in the data, and are very suitable for processing power anomaly data streams, environmental parameter monitoring data streams, and operating parameter monitoring data streams with temporal characteristics.

[0085] As an implementation design, the process of extracting characterization information from the preset power anomaly data stream according to the abnormal data embedding mapping layer in the conformity evaluation network to obtain an abnormal power data characterization vector includes: determining the first power state characterization information and the first distribution characterization information of the preset power anomaly data stream according to the abnormal data embedding mapping layer, and extracting characterization information from the first power state characterization information and the first distribution characterization information to obtain an abnormal power data characterization vector. Among them, the first power state characterization information represents the characterization information of multiple data items in the preset power anomaly data stream, and the first distribution characterization information represents the characterization information of the distribution of multiple data items in the preset power anomaly data stream.

[0086] The preset power anomaly data stream represents multiple data items in the abnormal power data. The preset power anomaly data stream can include the data item features of multiple data items in the abnormal power data. In this application, a data item feature can be regarded as a token, and a token represents a data item. The first token in the preset power anomaly data stream is <start> ， <start>Represents the first token, and the last token in the preset power anomaly data stream is <end> ， <end>Represents the last token.

[0087] Based on the characteristics of multiple data items in the preset power anomaly data stream, the first power state characterization information and the first distribution characterization information can be determined. The first power state characterization information represents the characteristic information of multiple data items in the preset power anomaly data stream, and the first distribution characterization information represents the distribution information of multiple data items in the preset power anomaly data stream. According to the anomaly data embedding mapping layer, the first power state characterization information and the first distribution characterization information are subjected to characterization information extraction to obtain an abnormal power data characterization vector.

[0088] As an implementation manner, the characterization information extraction of the first power state characterization information and the first distribution characterization information to obtain an abnormal power data characterization vector may include the following steps S221 to S223:

[0089] Step S221, perform characterization information extraction on the first power state characterization information and the first distribution characterization information to obtain a first anchor characterization vector, a first result characterization vector, and a first search characterization vector.

[0090] The anchor characterization vector is a key vector, the result characterization vector is a value vector, and the search characterization vector is a query vector. They belong to different vector domains. Each data item in the abnormal power data corresponds to its own first anchor characterization vector, first result characterization vector, and first search characterization vector. The first result characterization vector represents the characteristic information of the data item itself, and the first search characterization vector and the first anchor characterization vector are used to determine the attention score.

[0091] As an implementation manner, obtaining the first search characterization vector, the first anchor characterization vector, and the first result characterization vector may specifically include: taking the product of the first power state characterization information and the first parameter array as a semantic array, and obtaining the first search characterization vector, the first anchor characterization vector, and the first result characterization vector corresponding to the first power state characterization information according to the semantic array. Taking the product of the first distribution characterization information and the second parameter array to obtain a position array, and obtaining the first search characterization vector, the first anchor characterization vector, and the first result characterization vector corresponding to the first distribution characterization information according to the position array. The parameter array usually exists in the form of a matrix. In the attention mechanism, the matrix representations of search, key, and value are usually obtained from the input sequence through a linear transformation first. These linear transformations are implemented by learnable parameter arrays (search array, key array, and value array). These parameter arrays are optimized during the training process so that the neural network can better capture the relevant information in the input data stream and generate effective outputs accordingly.

[0092] The first parameter array and the second parameter array are used for spatial mapping, and the first parameter array and the second parameter array are network parameter variables obtained through training. After obtaining the semantic array, the semantic array is split into a first search representation vector, a first anchor representation vector, and a first result representation vector corresponding to the first power state representation information. After obtaining the position array, the position array is split into a first search representation vector, a first anchor representation vector, and a first result representation vector corresponding to the first distribution representation information. For example, for the first parameter array, which is a parameter array with three dimensions, the semantic array obtained by multiplying the first power state representation information and the first parameter array is a semantic array with three dimensions. Each dimension of the semantic array is respectively used as the first search representation vector, the first anchor representation vector, and the first result representation vector corresponding to the first power state representation information.

[0093] Step S222: Interact the first anchor representation vector, the first result representation vector, and the first search representation vector to obtain a candidate abnormal power data representation vector.

[0094] After obtaining the first anchor representation vector, the first result representation vector, and the first search representation vector, interact the first anchor representation vector, the first result representation vector, and the first search representation vector to obtain a candidate abnormal power data representation vector.

[0095] As an implementation manner, interacting the first anchor representation vector, the first result representation vector, and the first search representation vector to obtain a candidate abnormal power data representation vector may include, for example: normalizing the multiplication result of the first search representation vector, the first anchor representation vector, and the normalization coefficient to obtain a normalized representation vector, and using the multiplication result of the normalized representation vector and the first result representation vector as the candidate abnormal power data representation vector. Obtain the normalization coefficient, where the normalization coefficient represents the scaling ratio of normalization. In the embodiment of the present application, based on the normalization coefficient as a parameter for normalization, the multiplication result is normalized to obtain a normalized representation vector. The normalized representation vector represents the correlation degree between the first search representation vector and the first anchor representation vector. Then, the normalized representation vector can be used as the influence coefficient (i.e., weight) of the first result representation vector, and the multiplication result between the normalized representation vector and the first result representation vector is determined as the candidate abnormal power data representation vector.

[0096] The process of interacting the first anchor representation vector, the first result representation vector, and the first search representation vector to obtain a candidate abnormal power data representation vector can refer to the following formula:

[0097] V1 = f(S·M T ·α)·G;

[0098] Specifically, V1 is the candidate abnormal power supply data representation vector, S is the first search representation vector, M is the first anchor representation vector, T is the transpose, α is the normalization coefficient, G is the first result representation vector, and f is the normalization function.

[0099] Step S223: Perform multiple rounds of non-linear transformation on the candidate abnormal power supply data representation vector to obtain the abnormal power supply data representation vector.

[0100] As an implementation, based on a multi-layer perceptron, perform multiple rounds of non-linear transformation on the candidate abnormal power supply data representation vector to obtain the abnormal power supply data representation vector. Among them, the multi-layer perceptron is expressed as: V1' = MLP(V1).

[0101] Specifically, V1' is the abnormal power supply data representation vector, MLP is the multi-layer perceptron, and V1 is the candidate abnormal power supply data representation vector.

[0102] When extracting the abnormal power supply data representation vector, for example, the input data is a token in the preset power supply abnormal data stream. For example, the tokens in the preset power supply abnormal data stream include <start> 、<2>、<3>、<4>、… <end>Determine the power status characterization information and distribution characterization information of each data item according to the tokens in the preset power anomaly data stream. Based on the anomaly data embedding mapping layer, extract the characterization information from the power status characterization information and distribution characterization information of each data item to obtain an output, which is the anomaly power data characterization vector. The dimension of the anomaly power data characterization vector is the same as that of the preset power anomaly data stream. The anomaly power data characterization vector includes the deep features of each data item in the anomaly power data.

[0103] Step S230, based on the environmental parameter embedding mapping layer in the conformity evaluation network, extract the characterization information from the environmental parameter monitoring data stream to obtain the environmental parameter data characterization vector.

[0104] Input the environmental parameter monitoring data stream into the environmental parameter embedding mapping layer in the conformity evaluation network. The environmental parameter embedding mapping layer extracts the characterization information from the environmental parameter monitoring data stream and outputs the environmental parameter data characterization vector.

[0105] As an implementation design, the architecture of the environmental parameter embedding mapping layer can be a convolutional neural network (CNN), a recurrent neural network (RNN) and its variants, such as long short-term memory network (LSTM) or gated recurrent unit (GRU). These network structures are suitable for processing sequence data, can capture the temporal dependence in the data, and are very suitable for processing power anomaly data streams, environmental parameter monitoring data streams and operating parameter monitoring data streams with temporal characteristics.

[0106] As an implementation design, based on the environmental parameter embedding mapping layer in the conformity evaluation network, extract the characterization information from the environmental parameter monitoring data stream to obtain the environmental parameter data characterization vector, which can specifically include: determine the second power status characterization information and the second distribution characterization information of the environmental parameter monitoring data stream according to the environmental parameter embedding mapping layer, and extract the characterization information from the second power status characterization information and the second distribution characterization information to obtain the environmental parameter data characterization vector. The second power status characterization information represents the characterization information of multiple data items in the environmental parameter monitoring data stream, and the second distribution characterization information represents the characterization information of the distribution of multiple data items in the environmental parameter monitoring data stream.

[0107] The environmental parameter monitoring data stream represents multiple data items in the environmental parameter sequence. The environmental parameter monitoring data stream can include the data item features of multiple data items in the environmental parameter data stream. As mentioned above, a data item feature can be regarded as a token, that is, one token represents one data item. The first token in the environmental parameter monitoring data stream is <start> ， <start>Represents the first token, and the last token in the environmental parameter monitoring data stream is <end> ， <end>Represents the last token. By monitoring the characteristics of multiple data items in the data stream according to the environmental parameters, the second power state characterization information and the second distribution characterization information can be determined. The second power state characterization information represents the characteristics of multiple data items in the environmental parameter monitoring data stream itself, and the second distribution characterization information is the characterization information of the distribution of multiple data items in the environmental parameter monitoring data stream. Based on the environmental parameter embedding mapping layer, the second power state characterization information and the second distribution characterization information are subjected to characterization information extraction to obtain the environmental parameter data characterization vector.

[0108] As an implementation manner, the manner of performing characterization information extraction on the second power state characterization information and the second distribution characterization information to obtain the environmental parameter data characterization vector may include the following steps S231 to S233:

[0109] Step S231, perform characterization information extraction on the second power state characterization information and the second distribution characterization information to obtain the second anchor characterization vector, the second result characterization vector, and the second search characterization vector.

[0110] Each data item in the environmental parameters includes its own second anchor characterization vector, second result characterization vector, and second search characterization vector. The second result characterization vector represents the characteristics of the data item itself, and the second search characterization vector and the second anchor characterization vector are used to determine the attention score.

[0111] As an implementation manner, obtaining the second search characterization vector, the second anchor characterization vector, and the second result characterization vector includes: obtaining the semantic array by multiplying the second power state characterization information by the third parameter array, and obtaining the second search characterization vector, the second anchor characterization vector, and the second result characterization vector corresponding to the second power state characterization information according to the semantic array. Obtain the position array by multiplying the second distribution characterization information by the fourth parameter array, and obtain the second search characterization vector, the second anchor characterization vector, and the second result characterization vector corresponding to the second distribution characterization information according to the position array.

[0112] The third parameter array and the fourth parameter array are used for spatial mapping, and the third parameter array and the fourth parameter array are network parameter variables obtained by training.

[0113] Step S232, interact the second anchor characterization vector, the second result characterization vector, and the second search characterization vector to obtain the candidate environmental parameter data characterization vector.

[0114] After obtaining the second anchor characterization vector, the second result characterization vector, and the second search characterization vector, interact the second anchor characterization vector, the second result characterization vector, and the second search characterization vector to obtain the candidate environmental parameter data characterization vector.

[0115] As an implementation manner, the process of interacting the second anchor representation vector, the second result representation vector, and the second search representation vector to obtain the candidate environmental parameter data representation vector includes: standardizing the multiplication result of the second search representation vector, the second anchor representation vector, and the normalization coefficient to obtain a standardized representation vector, and determining the multiplication result of the standardized representation vector and the second result representation vector as the candidate environmental parameter data representation vector.

[0116] Step S233: Perform multiple rounds of non-linear transformation on the candidate environmental parameter data representation vector to obtain the environmental parameter data representation vector.

[0117] As an implementation manner, use a multi-layer perceptron to perform multiple rounds of non-linear transformation on the candidate environmental parameter data representation vector to obtain the environmental parameter data representation vector.

[0118] Step S240: Extract the representation information from the operation parameter monitoring data stream according to the operation parameter embedding mapping layer in the compliance evaluation network to obtain the operation parameter data representation vector.

[0119] Input the operation parameter monitoring data stream into the operation parameter embedding mapping layer in the compliance evaluation network. The operation parameter embedding mapping layer extracts the representation information from the operation parameter monitoring data stream and outputs the operation parameter data representation vector.

[0120] As an implementation design, the network architecture of the operation parameter embedding mapping layer can be a convolutional neural network (CNN), a recurrent neural network (RNN), and its variants, such as long short-term memory network (LSTM) or gated recurrent unit (GRU). These network structures are suitable for processing sequence data, can capture the time dependence in the data, and are very suitable for processing power anomaly data streams, environmental parameter monitoring data streams, and operation parameter monitoring data streams with time series characteristics.

[0121] As an implementation design, according to the operating parameters in the matching degree evaluation network, the embedding mapping layer is embedded, and the characterization information extraction is performed on the operating parameter monitoring data stream to obtain the operating parameter data characterization vector. Specifically, it may include: according to the operating parameter embedding mapping layer, determining the third power state characterization information and the third distribution characterization information of the operating parameter monitoring data stream, and performing characterization information extraction on the third power state characterization information and the third distribution characterization information to obtain the operating parameter data characterization vector. The third power state characterization information represents the characterization information of multiple data items in the operating parameter monitoring data stream, and the third distribution characterization information represents the characterization information of the distribution of multiple data items in the operating parameter monitoring data stream. The operating parameter monitoring data stream represents multiple data items in the operating parameter sequence. The operating parameter monitoring data stream may include the data item features of multiple data items in the operating parameter data stream. As mentioned above, one data item feature can be regarded as a token, that is, one token represents one data item. The first token in the operating parameter monitoring data stream is <start> ， <start>Represents the first token, and the last token in the monitored data stream of the running parameter is <end> ， <end>Represents the last token. By monitoring the characteristics of multiple data items in the data stream according to the running parameters, the third power state characterization information and the third distribution characterization information can be determined. The third power state characterization information represents the characteristics of multiple data items in the running parameter monitoring data stream itself, and the third distribution characterization information is the characterization information of the distribution of multiple data items in the running parameter monitoring data stream. Based on the running parameter embedding mapping layer, the third power state characterization information and the third distribution characterization information are subjected to characterization information extraction to obtain the running parameter data characterization vector.

[0122] As an implementation manner, performing characterization information extraction on the third power state characterization information and the third distribution characterization information to obtain the running parameter data characterization vector may specifically include the following steps S241 to step S243:

[0123] Step S241, performing characterization information extraction on the third power state characterization information and the third distribution characterization information to obtain the third anchor characterization vector, the third result characterization vector, and the third search characterization vector.

[0124] Each data item in the running parameters includes its own third anchor characterization vector, third result characterization vector, and third search characterization vector. The third result characterization vector is the characteristic of the data item itself, and the third search characterization vector and the third anchor characterization vector are configured to determine the attention score.

[0125] As an implementation manner, obtaining the third search characterization vector, the third anchor characterization vector, and the third result characterization vector may specifically include: obtaining the semantic array by multiplying the third power state characterization information by the fifth parameter array, and obtaining the third search characterization vector, the third anchor characterization vector, and the third result characterization vector corresponding to the third power state characterization information according to the semantic array. Multiplying the third distribution characterization information by the sixth parameter array to obtain the position array, and obtaining the third search characterization vector, the third anchor characterization vector, and the third result characterization vector corresponding to the third distribution characterization information according to the position array. The fifth parameter array and the sixth parameter array are used for spatial mapping, and the fifth parameter array and the sixth parameter array are network parameter variables obtained by training.

[0126] Step S242, performing interaction on the third anchor characterization vector, the third result characterization vector, and the third search characterization vector to obtain the candidate running parameter data characterization vector.

[0127] After obtaining the third anchor characterization vector, the third result characterization vector, and the third search characterization vector, perform interaction on the third anchor characterization vector, the third result characterization vector, and the third search characterization vector to obtain the candidate running parameter data characterization vector.

[0128] As an implementation manner, the process of interacting the third anchor representation vector, the third result representation vector, and the third search representation vector to obtain the candidate operating parameter data representation vector includes: standardizing the multiplication result of the third search representation vector, the third anchor representation vector, and the normalization coefficient to obtain a standardized representation vector, and determining the multiplication result of the standardized representation vector and the third result representation vector as the candidate operating parameter data representation vector.

[0129] Step S243: Perform multiple rounds of non-linear transformation on the candidate operating parameter data representation vector to obtain the operating parameter data representation vector.

[0130] As an implementation manner, use a multi-layer perceptron to perform multiple rounds of non-linear transformation on the candidate operating parameter data representation vector to obtain the operating parameter data representation vector.

[0131] Step S250: According to the interaction layer in the fitness evaluation network, interact the abnormal power supply data representation vector, the environmental parameter data representation vector, and the operating parameter data representation vector to obtain the interaction data representation vector.

[0132] As an implementation design, the interaction layer includes a first filter, a second filter, and a third filter. Input the abnormal power supply data representation vector into the first filter, and the first filter performs multiple filtering operations on the abnormal power supply data representation vector to obtain a deep abnormal power supply data representation vector. Input the environmental parameter data representation vector into the second filter, and based on the second filter, perform multiple filtering operations on the environmental parameter data representation vector to obtain a deep environmental parameter data representation vector. Input the operating parameter data representation vector into the third filter, and based on the third filter, perform multiple filtering operations on the operating parameter data representation vector to obtain a deep operating parameter data representation vector. Interact (such as concatenate) the deep abnormal power supply data representation vector, the environmental parameter data representation vector, and the operating parameter data representation vector to obtain the interaction data representation vector.

[0133] Step S260: According to the dense layer in the fitness evaluation network, perform abnormal diagnosis on the interaction data representation vector to obtain the power supply abnormal diagnosis result, and the power supply abnormal diagnosis result represents the fitness of the real-time power supply data to the abnormal power supply data.

[0134] Input the interaction data representation vector into the dense layer, and the dense layer performs abnormal diagnosis on the interaction data representation vector and outputs the power supply abnormal diagnosis result.

[0135] As an implementation design, the power supply anomaly diagnosis result is a matching degree value. The larger the matching degree value is, the greater the matching degree of the real-time power supply data to the abnormal power supply data. The smaller the matching degree value is, the smaller the matching degree of the real-time power supply data to the abnormal power supply data. As mentioned above, in other implementation designs, the power supply anomaly diagnosis result is a binary classification result. When the power supply anomaly diagnosis result is the first result, the real-time power supply data matches the abnormal power supply data. When the power supply anomaly diagnosis result is the second result, the real-time power supply data does not match the abnormal power supply data.

[0136] When this application embodiment infers the matching degree between the real-time power supply data and the abnormal power supply data, it comprehensively measures the characterization information of the data items in the abnormal power supply data, the characterization information of the data items in the environmental parameters of the real-time power supply data, and the characterization information of the data items in the operating parameters of the real-time power supply data. From the two aspects of operating parameters and environmental parameters, it considers the implicit connection between the characterization information of the data items in the abnormal power supply data and the characterization information of the data items in the environmental parameters and operating parameters, so as to infer the matching degree between the real-time power supply data and the abnormal power supply data. The inferred matching degree has high accuracy and reliability.

[0137] The following introduces the debugging process of the matching degree evaluation network, which can specifically include the following steps:

[0138] Step S310, obtain the example preset power supply anomaly data stream, the example real-time power supply monitoring data stream, and the actual power supply anomaly diagnosis result. The example real-time power supply monitoring data stream includes the example environmental parameter monitoring data stream and the example operating parameter monitoring data stream. The actual power supply anomaly diagnosis result represents the actual matching degree of the example real-time power supply data to the example abnormal power supply data.

[0139] The example preset power supply anomaly data stream and the example real-time power supply monitoring data stream are training samples. The example real-time power supply monitoring data stream includes the example environmental parameter monitoring data stream of the example real-time power supply data and the example operating parameter monitoring data stream. The example preset power supply anomaly data stream represents the data items in the example abnormal power supply data. The example environmental parameter monitoring data stream represents the data items in the environmental parameters of the example real-time power supply data. The example operating parameter monitoring data stream represents the data items in the operating parameters of the example real-time power supply data.

[0140] The process of step S310 can refer to the aforementioned step S110.

[0141] Step S320, according to the matching degree evaluation network, extract the characterization information of the example preset power supply anomaly data stream, the example environmental parameter monitoring data stream, and the example operating parameter monitoring data stream respectively, and obtain the example abnormal power supply data characterization vector, the example environmental parameter data characterization vector, and the example operating parameter data characterization vector.

[0142] Input the example preset power anomaly data stream, example environmental parameter monitoring data stream, and example operating parameter monitoring data stream into the consistency evaluation network. The consistency evaluation network respectively extracts characterization information from the example preset power anomaly data stream, example environmental parameter monitoring data stream, and example operating parameter monitoring data stream, and outputs an example abnormal power data characterization vector, an example environmental parameter data characterization vector, and an example operating parameter data characterization vector.

[0143] As an implementation design, the consistency evaluation network includes an abnormal data embedding mapping layer, an environmental parameter embedding mapping layer, and an operating parameter embedding mapping layer. Obtaining the example abnormal power data characterization vector, example environmental parameter data characterization vector, and example operating parameter data characterization vector can specifically include the following steps S321 to S323.

[0144] Step S321: According to the abnormal data embedding mapping layer, extract characterization information from the example preset power anomaly data stream to obtain an example abnormal power data characterization vector.

[0145] As an implementation manner, according to the abnormal data embedding mapping layer, determine the first example power state characterization information and the first example distribution characterization information of the example preset power anomaly data stream, and extract characterization information from the first example power state characterization information and the first example distribution characterization information to obtain an example abnormal power data characterization vector. Among them, the first example power state characterization information represents the characterization information of multiple data items in the example preset power anomaly data stream, and the first example distribution characterization information represents the characterization information of the distribution of multiple data items in the example preset power anomaly data stream.

[0146] As an implementation manner, extracting characterization information from the first example power state characterization information and the first example distribution characterization information to obtain an example abnormal power data characterization vector includes: extracting characterization information from the first example power state characterization information and the first example distribution characterization information to obtain a first example anchor characterization vector, a first example result characterization vector, and a first example search characterization vector. Interact the first example anchor characterization vector, the first example result characterization vector, and the first example search characterization vector to obtain a candidate example abnormal power data characterization vector. Perform multiple rounds of non-linear transformation on the candidate example abnormal power data characterization vector to obtain an example abnormal power data characterization vector.

[0147] Step S322: According to the environmental parameter embedding mapping layer, extract characterization information from the example environmental parameter monitoring data stream to obtain an example environmental parameter data characterization vector.

[0148] As an implementation manner, according to the environmental parameters, the mapping layer is embedded to determine the second example power state characterization information and the second example distribution characterization information of the example environmental parameter monitoring data stream, and the characterization information extraction is performed on the second example power state characterization information and the second example distribution characterization information to obtain the example environmental parameter data characterization vector. The second example power state characterization information represents the characterization information of multiple data items in the example environmental parameter monitoring data stream, and the second example distribution characterization information represents the characterization information of the distribution of multiple data items in the example environmental parameter monitoring data stream.

[0149] As an implementation manner, the process of performing characterization information extraction on the second example power state characterization information and the second example distribution characterization information to obtain the example environmental parameter data characterization vector includes: performing characterization information extraction on the second example power state characterization information and the second example distribution characterization information to obtain the second example anchor characterization vector, the second example result characterization vector, and the second example search characterization vector. Interact the second example anchor characterization vector, the second example result characterization vector, and the second example search characterization vector to obtain the candidate example environmental parameter data characterization vector. Perform multiple rounds of non-linear transformation on the candidate example environmental parameter data characterization vector to obtain the example environmental parameter data characterization vector.

[0150] Step S323, according to the operating parameters, the mapping layer is embedded to perform characterization information extraction on the example operating parameter monitoring data stream to obtain the example operating parameter data characterization vector.

[0151] As an implementation manner, according to the operating parameters, the mapping layer is embedded to determine the third example power state characterization information and the third example distribution characterization information of the example operating parameter monitoring data stream, and the characterization information extraction is performed on the third example power state characterization information and the third example distribution characterization information to obtain the example operating parameter data characterization vector. The third example power state characterization information represents the characterization information of multiple data items in the example operating parameter monitoring data stream, and the third example distribution characterization information represents the characterization information of the distribution of multiple data items in the example operating parameter monitoring data stream.

[0152] As an implementation manner, the process of performing characterization information extraction on the third example power state characterization information and the third example distribution characterization information to obtain the example operating parameter data characterization vector includes: performing characterization information extraction on the third example power state characterization information and the third example distribution characterization information to obtain the third example anchor characterization vector, the third example result characterization vector, and the third example search characterization vector. Interact the third example anchor characterization vector, the third example result characterization vector, and the third example search characterization vector to obtain the example candidate operating parameter data characterization vector. Perform multiple rounds of non-linear transformation on the example candidate operating parameter data characterization vector to obtain the example operating parameter data characterization vector.

[0153] Step S330: According to the matching degree evaluation network, interact with the sample abnormal power data representation vector, the sample environmental parameter data representation vector, and the sample operating parameter data representation vector to obtain the sample interaction data representation vector.

[0154] As an implementation design, the matching degree evaluation network includes an interaction layer. According to the interaction layer, interact with the sample abnormal power data representation vector, the sample environmental parameter data representation vector, and the sample operating parameter data representation vector to obtain the sample interaction data representation vector.

[0155] As an implementation manner, the interaction layer includes a first filter, a second filter, and a third filter. Input the sample abnormal power data representation vector into the first filter, and perform multiple filtrations on the sample abnormal power data representation vector based on the first filter to obtain a deep sample abnormal power data representation vector. Input the sample environmental parameter data representation vector into the second filter, and perform multiple filtrations on the sample environmental parameter data representation vector based on the second filter to obtain a deep sample environmental parameter data representation vector. Input the sample operating parameter data representation vector into the third filter, and perform multiple filtrations on the sample operating parameter data representation vector based on the third filter to obtain a deep sample operating parameter data representation vector. Interact the deep sample abnormal power data representation vector, the sample environmental parameter data representation vector, and the sample operating parameter data representation vector to obtain the sample interaction data representation vector.

[0156] Step S340: According to the matching degree evaluation network, perform abnormal diagnosis on the sample interaction data representation vector to obtain the sample power abnormal diagnosis result.

[0157] As an implementation design, the matching degree evaluation network includes a dense layer. According to the dense layer, perform abnormal diagnosis on the sample interaction data representation vector to obtain the sample power abnormal diagnosis result.

[0158] Step S350: Debug the matching degree evaluation network according to the sample power abnormal diagnosis result and the actual power abnormal diagnosis result.

[0159] Obtain the actual power abnormal diagnosis result of the sample abnormal power data and the sample real-time power data. This actual power abnormal diagnosis result represents the actual matching degree of the sample real-time power data to the sample abnormal power data. The obtained sample power abnormal diagnosis result is the matching degree inferred based on the matching degree evaluation network. Debug the matching degree evaluation network according to the difference between the sample power abnormal diagnosis result and the actual power abnormal diagnosis result.

[0160] Since the goal of the consistency evaluation network is to infer the actual consistency of the example real-time power data with the example abnormal power data, the more closely the example power anomaly diagnosis result matches the actual power anomaly diagnosis result, the more accurate the consistency evaluation network is. According to the error between the example power anomaly diagnosis result and the actual power anomaly diagnosis result, the consistency evaluation network is debugged to reduce the error between the example power anomaly diagnosis result obtained based on the debugged consistency evaluation network and the actual power anomaly diagnosis result, enhance the inference effect of the consistency evaluation network, and improve the accuracy of the consistency evaluation network.

[0161] As an implementation design, the above steps S310 to S350 are repeated to complete the iterative debugging of the consistency evaluation network until a preset debugging cut-off condition is reached, such as the number of debuggings reaches a preset maximum number or the network loss is less than a preset loss value.

[0162] When debugging the consistency evaluation network for inferring the consistency between real-time power data and abnormal power data in the embodiments of this application, the characterization information of data items in the abnormal power data, the characterization information of data items in the environmental parameters of the real-time power data, and the characterization information of data items in the operating parameters of the real-time power data are comprehensively measured. From the two dimensions of operating parameters and environmental parameters, the consistency evaluation network is made to learn the implicit relationship between the characterization information of data items in the abnormal power data and the characterization information of data items in the environmental parameters and operating parameters, and infer the consistency between the real-time power data and the abnormal power data. Not only is the consideration dimension comprehensive, but the analysis is carried out from two perspectives of operating parameters and environmental parameters, which improves the inference effect of the consistency evaluation network.

[0163] The above consistency evaluation network includes an abnormal data embedding mapping layer, an environmental parameter embedding mapping layer, and an operating parameter embedding mapping layer. Before debugging the consistency evaluation network, the abnormal data embedding mapping layer, the environmental parameter embedding mapping layer, and the operating parameter embedding mapping layer are either untrained embedding mapping layers or embedding mapping layers after pre-training. After pre-training is completed, based on the abnormal data embedding mapping layer, environmental parameter embedding mapping layer, and operating parameter embedding mapping layer obtained from pre-training, a consistency evaluation network is established, and then the consistency evaluation network is debugged.

[0164] The pre-training process of the abnormal data embedding mapping layer, the environmental parameter embedding mapping layer, and the operating parameter embedding mapping layer is as follows.

[0165] Step S410, obtain a first preset power anomaly data stream, and mask the data item information in a local area of the first preset power anomaly data stream to obtain a second preset power anomaly data stream.

[0166] The first preset power supply abnormal data stream is the preset power supply abnormal data stream of any abnormal power supply data. This first preset power supply abnormal data stream represents the data items in the abnormal power supply data, and the abnormal power supply data includes multiple data items. The first preset power supply abnormal data stream includes data item information of multiple regions, and the data items of each region represent the data items of the corresponding region. Arbitrarily mask the data item information of the local region in the first preset power supply abnormal data stream to obtain the second preset power supply abnormal data stream. Then, the data items of the local region in the second preset power supply abnormal data stream are unknowable data items.

[0167] As an implementation design, set the data items of the local region in the first preset power supply abnormal data stream as preset marks. For example, the preset mark is <h>。

[0168] Step S420: Based on the anomaly data embedding mapping layer, reason about the data item information in the masked area of the second preset power anomaly data stream to obtain a first reasoning confidence level. The first reasoning confidence level represents the confidence level that the data items in the masked area of the first preset power anomaly data stream obtained by reasoning belong to each type of data item.

[0169] Input the second preset power anomaly data stream into the anomaly data embedding mapping layer. The anomaly data embedding mapping layer reasons about the data item information in the masked area of the second preset power anomaly data stream and outputs the first reasoning confidence level.

[0170] As an implementation design, during pre-debugging, compared with the matching degree evaluation network, the anomaly data embedding mapping layer further includes an output structure. After the pre-debugging of the anomaly data embedding mapping layer is completed, the output structure in the anomaly data embedding mapping layer is deleted, and a matching degree evaluation network is established based on the anomaly data embedding mapping layer after deleting the output structure. Based on the anomaly data embedding mapping layer, determine the power state characterization information and distribution characterization information of the second preset power anomaly data stream, perform characterization information extraction on the power state characterization information and distribution characterization information to obtain an abnormal power data characterization vector. Among them, the power state characterization information represents the characterization information of multiple data items in the preset power anomaly data stream, and the distribution characterization information represents the characterization information of the distribution of multiple data items in the preset power anomaly data stream. Then use the output structure to reason about the abnormal power data characterization vector to obtain the first reasoning confidence level.

[0171] The method for the anomaly data embedding mapping layer to extract the abnormal power data characterization vector of the second preset power anomaly data stream is the same as the method for extracting the abnormal power data characterization vector in the above embodiments.

[0172] Step S430: Debug the anomaly data embedding mapping layer according to the first reasoning confidence level and the first actual confidence level. The first actual confidence level represents the actual confidence level that the data items in the masked area of the first preset power anomaly data stream belong to each type of data item.

[0173] Obtain the first actual confidence level, which represents the actual confidence level that the data items in the masked area of the first preset power anomaly data stream belong to each type of data item. The obtained first reasoning confidence level is reasoned by the anomaly data embedding mapping layer. Debug the anomaly data embedding mapping layer according to the difference between the first reasoning confidence level and the first actual confidence level.

[0174] Because the goal of the abnormal data embedding mapping layer is to infer the actual confidence that the data items in the masked area in the first preset power supply abnormal data stream belong to each type of data item, the more similar the first inference confidence is to the first actual confidence, the better the effect of the abnormal data embedding mapping layer in extracting the abnormal power supply data representation vector. According to the error between the first inference confidence and the first actual confidence, the first inference confidence and the first actual confidence are debugged to reduce the error between the first inference confidence obtained according to the debugged abnormal data embedding mapping layer and the first actual confidence, thereby increasing the representation information extraction capability of the abnormal data embedding mapping layer, so that the abnormal data embedding mapping layer can roughly learn to extract the abnormal power supply data representation vector.

[0175] In the above implementation mode, the abnormal data embedding mapping layer is pre-debugged based on self-supervision according to the example preset power abnormality data stream, which increases the effect of the abnormal data embedding mapping layer on extracting characterization information of the preset power abnormality data stream. Subsequently, a consistency evaluation network is established based on the pre-debugged abnormal data embedding mapping layer, which helps to reduce the debugging complexity of the consistency evaluation network and ensure the accuracy and reliability of the consistency evaluation network.

[0176] The pre-debugging process of embedding environment parameters into the mapping layer may include:

[0177] Step S510, obtaining a first environmental parameter monitoring data stream, shielding data item information of a local area in the first environmental parameter monitoring data stream, and obtaining a second environmental parameter monitoring data stream.

[0178] The first environmental parameter monitoring data stream can be an environmental parameter monitoring data stream of any type of environmental parameter. The first environmental parameter monitoring data stream represents data items in the environmental parameters, and the environmental parameters include multiple data items. The first environmental parameter monitoring data stream includes data item information of multiple regions, and the data item information of each region represents the data items of the region. The data item information of the local region in the first environmental parameter monitoring data stream is arbitrarily shielded to obtain the second environmental parameter monitoring data stream, and then the data item information of the local region in the second environmental parameter monitoring data stream is unknown. The shielding method of the first environmental parameter monitoring data stream can refer to the shielding method of the aforementioned first preset power supply abnormality data stream.

[0179] Step S520, based on the environmental parameter embedded mapping layer, infer the data item information of the shielded area in the second environmental parameter monitoring data stream to obtain a second inference confidence, and the second inference confidence represents the confidence that the data items in the shielded area in the first environmental parameter monitoring data stream obtained by inference belong to each category of data items.

[0180] Input the second environmental parameter monitoring data stream into the environmental parameter embedding and mapping layer. The environmental parameter embedding and mapping layer infers the data item information in the masked area of the second environmental parameter monitoring data stream and outputs the second inference confidence level.

[0181] As an implementation design, during pre-debugging, compared with the matching degree evaluation network, the environmental parameter embedding and mapping layer further includes an output structure. After the pre-debugging of the environmental parameter embedding and mapping layer is completed, the output structure in the environmental parameter embedding and mapping layer is deleted, and a matching degree evaluation network is established based on the environmental parameter embedding and mapping layer after deleting the output structure. Then, based on the environmental parameter embedding and mapping layer, the power state characterization information and distribution characterization information of the second environmental parameter monitoring data stream are determined, and characterization information extraction is performed on the power state characterization information and distribution characterization information to obtain an environmental parameter data characterization vector. Among them, the power state characterization information represents the characterization information of multiple data items in the environmental parameter monitoring data stream, and the distribution characterization information represents the characterization information of the distribution of multiple data items in the environmental parameter monitoring data stream. Then, based on the output structure, an inference is made on the environmental parameter data characterization vector to obtain the second inference confidence level. The method for the environmental parameter embedding and mapping layer to extract the environmental parameter data characterization vector of the second environmental parameter monitoring data stream can refer to the method for extracting the environmental parameter data characterization vector in the foregoing embodiments.

[0182] Step S530: Debug the environmental parameter embedding and mapping layer according to the second inference confidence level and the second actual confidence level. The second actual confidence level represents the actual confidence level of the data items in the masked area of the first environmental parameter monitoring data stream belonging to each type of data item.

[0183] Obtain the second actual confidence level, which represents the actual confidence level of the data items in the masked area of the first environmental parameter monitoring data stream belonging to each type of data item. The obtained second inference confidence level is based on the inference of the environmental parameter embedding and mapping layer. Debug the environmental parameter embedding and mapping layer according to the error between the second inference confidence level and the second actual confidence level.

[0184] Since the goal of the environmental parameter embedding and mapping layer is to infer the actual confidence level of the data items in the masked area of the first environmental parameter monitoring data stream belonging to each type of data item, the more similar the second inference confidence level and the second actual confidence level are, the better the effect of the environmental parameter embedding and mapping layer in extracting the environmental parameter data characterization vector. Then, according to the error between the second inference confidence level and the second actual confidence level, debug the second inference confidence level and the second actual confidence level to reduce the error between the second inference confidence level obtained based on the debugged environmental parameter embedding and mapping layer and the second actual confidence level, so as to increase the characterization information extraction effect of the environmental parameter embedding and mapping layer, thereby making the environmental parameter embedding and mapping layer familiar with extracting the environmental parameter data characterization vector.

[0185] Based on the example environmental parameter monitoring data stream, the embodiment of this application pre - debugs the environmental parameter embedding mapping layer through a self - supervision mechanism, improves the effect of the environmental parameter embedding mapping layer in extracting characterization information from the environmental parameter monitoring data stream, and then establishes a matching degree evaluation network based on the pre - debugged environmental parameter embedding mapping layer, which helps to reduce the debugging complexity of the matching degree evaluation network and ensure the reliability and accuracy of the matching degree evaluation network.

[0186] The following introduces the pre - debugging process of the environmental parameter embedding mapping layer, including the following steps:

[0187] Step S610: Obtain the first operating parameter monitoring data stream, and mask the data item information in the local area of the first operating parameter monitoring data stream to obtain the second operating parameter monitoring data stream.

[0188] The first operating parameter monitoring data stream can be the operating parameter monitoring data stream of any type of operating parameter. The first operating parameter monitoring data stream represents the data items in the operating parameter, and the operating parameter includes multiple data items. The first operating parameter monitoring data stream includes data item information in multiple regions, and the data item information in each region represents the data items in the region. Arbitrarily mask the data item information in the local area of the first operating parameter monitoring data stream to obtain the second operating parameter monitoring data stream. Then, the data item information in the local area of the second operating parameter monitoring data stream is unknown. The masking method of the first operating parameter monitoring data stream can refer to the masking method of the first preset power supply abnormal data stream in the above - mentioned embodiment.

[0189] Step S620: According to the operating parameter embedding mapping layer, infer the data item information in the masked area of the second operating parameter monitoring data stream to obtain the third inference confidence level. The third inference confidence level represents the confidence level that the data items in the masked area of the first operating parameter monitoring data stream obtained by inference belong to each type of data item.

[0190] Load the second operating parameter monitoring data stream into the operating parameter embedding mapping layer, reason about the data item information in the masked area of the second operating parameter monitoring data stream based on the operating parameter embedding mapping layer, and output the third reasoning confidence. As an implementation design, during pre-debugging, compared with the fitness evaluation network, the operating parameter embedding mapping layer further includes an output structure. After completing the pre-debugging of the operating parameter embedding mapping layer, delete the output structure in the operating parameter embedding mapping layer, and construct the fitness evaluation network based on the operating parameter embedding mapping layer after deleting the output structure. Then, based on the operating parameter embedding mapping layer, determine the power state representation information and distribution representation information of the second operating parameter monitoring data stream, perform representation information extraction on the power state representation information and distribution representation information, and obtain the operating parameter data representation vector. Among them, the power state representation information represents the representation information of multiple data items in the operating parameter monitoring data stream, and the distribution representation information represents the representation information of the distribution of multiple data items in the operating parameter monitoring data stream. Then, reason about the operating parameter data representation vector based on the output structure to obtain the third reasoning confidence.

[0191] The method for the operating parameter embedding mapping layer to extract the operating parameter data representation vector of the second operating parameter monitoring data stream can refer to the above method for extracting the operating parameter data representation vector.

[0192] Step S630: Debug the operating parameter embedding mapping layer according to the third reasoning confidence and the third actual confidence. The third actual confidence represents the actual confidence that the data items in the masked area of the first operating parameter monitoring data stream belong to each type of data item.

[0193] Obtain the third actual confidence, which represents the actual confidence that the data items in the masked area of the first operating parameter monitoring data stream belong to each type of data item. The obtained third reasoning confidence is based on the reasoning of the operating parameter embedding mapping layer. Debug the operating parameter embedding mapping layer according to the error between the third reasoning confidence and the third actual confidence.

[0194] Since the goal of the operating parameter embedding mapping layer is to reason about the actual confidence that the data items in the masked area of the first operating parameter monitoring data stream belong to each type of data item, the more similar the third reasoning confidence and the third actual confidence are, the better the effect of the operating parameter embedding mapping layer in extracting the operating parameter data representation vector. Then, based on the error between the third reasoning confidence and the third actual confidence, debug the third reasoning confidence and the third actual confidence to reduce the error between the third reasoning confidence obtained based on the debugged operating parameter embedding mapping layer and the third actual confidence, and increase the representation information extraction effect of the operating parameter embedding mapping layer, so that the operating parameter embedding mapping layer becomes familiar with the process of extracting the operating parameter data representation vector.

[0195] This application performs pre - debugging on the operation parameter embedding mapping layer through a self - supervision mechanism based on the example operation parameter monitoring data stream, enhancing the effect of the operation parameter embedding mapping layer in extracting characterization information from the operation parameter monitoring data stream. Then, a matching degree evaluation network is established based on the pre - debugged operation parameter embedding mapping layer, helping to reduce the debugging complexity of the matching degree evaluation network and ensuring the accuracy and reliability of the matching degree evaluation network.

[0196] It should be noted that in the embodiments of this application, if the above - mentioned method is implemented in the form of software function modules 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 embodiments of this application, in essence, or the part that contributes to the related technology, can be embodied in the form of a software product. The software product is stored in a storage medium and includes several instructions for causing a computer system (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read - only memories (ROMs), magnetic disks, or optical discs, etc., which can store program codes. In this way, the embodiments of this application are not limited to any specific combination of hardware and software.

[0197] The embodiments of this application provide a computer system, including a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor executes the computer program, the above - mentioned method is implemented.

[0198] The embodiments of this application provide a computer - readable storage medium, on which a computer program is stored. When the computer program is executed by the processor, the above - mentioned method is implemented. The computer - readable storage medium can be transient or non - transient.

[0199] The embodiments of this application provide a computer program product. The computer program product includes a non - transient computer - readable storage medium storing a computer program. When the computer program is read and executed by a computer, some or all of the steps in the above - mentioned method are implemented. The computer program product can be specifically implemented in a manner of hardware, software, or a combination thereof. In an optional embodiment, the computer program product is specifically embodied as a computer storage medium. In another optional embodiment, the computer program product is specifically embodied as a software product, such as a software development kit (SDK), etc.

[0200] It should be noted that Figure 3 is a schematic diagram of the hardware entity of a computer system 300 provided by the embodiments of this application, as Figure 3 As shown, the hardware entities of the computer system 300 include: a processor 310, a communication interface 320, and a memory 330, where: The processor 310 generally controls the overall operation of the computer system 300. The communication interface 320 enables the computer system to communicate with other terminals or servers via a network. The memory 330 is configured to store instructions and applications executable by the processor 310, and can also cache data to be processed or already processed by the processor 310 and each module in the computer system 300 (for example, image data, audio data, voice communication data, and video communication data), and can be implemented by flash memory (FLASH) or random access memory (Random Access Memory, RAM). Data transmission can be performed between the processor 310, the communication interface 320, and the memory 330 via a bus 340. It should be noted here that: The descriptions of the above storage medium and device embodiments are similar to those of the above method embodiments and have similar beneficial effects to the method embodiments. For the technical details not disclosed in the storage medium and device embodiments of the present application, please refer to the descriptions of the method embodiments of the present application for understanding.

[0201] It should be understood that the "one embodiment" or "an embodiment" mentioned throughout the specification means that the specific features, structures, or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, the "in one embodiment" or "in an embodiment" that appears throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present application, the order numbers of the above processes do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application. The serial numbers of the embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0202] It should be noted that in this article, the term "comprising", "including" or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0203] In several embodiments provided by this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed with each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be electrical, mechanical, or other forms.

[0204] The units described above as separate components may or may not be physically separated. The components shown as units may or may not be physical units; they can be located in one place or distributed to multiple network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0205] In addition, each functional unit in the embodiments of this application can be all integrated in a processing unit, or each unit can be separately used as a unit, or two or more units can be integrated in one unit; the above integrated units can be implemented in the form of hardware, or in the form of hardware plus software functional units.

[0206] Those of ordinary skill in the art can understand that all or part of the steps to implement the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: removable storage devices, read-only memory (ROM), magnetic disks, or optical disks and other various media that can store program codes.

[0207] Alternatively, if the above integrated units of this application are implemented in the form of software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the related technology can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer system (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the various embodiments of this application. And the foregoing storage medium includes: removable storage devices, ROM, magnetic disks, or optical disks and other various media that can store program codes.

[0208] As described above, it is only the implementation mode of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application.< / h> < / end> < / end> < / start> < / start> < / end> < / end> < / start> < / start> < / end> < / start> < / end> < / end> < / start> < / start>

Claims

1. A power anomaly diagnosis and alarm method based on artificial intelligence, characterized in that, The method includes: Obtaining a preset power anomaly data stream and a real-time power monitoring data stream. The real-time power monitoring data stream includes an environmental parameter monitoring data stream and an operating parameter monitoring data stream of real-time power data. The preset power anomaly data stream represents data items in abnormal power data. The environmental parameter monitoring data stream represents data items in the environmental parameters of the real-time power data. The operating parameter monitoring data stream represents data items in the operating parameters of the real-time power data; Respectively performing characterization information extraction on the preset power anomaly data stream, the environmental parameter monitoring data stream, and the operating parameter monitoring data stream to obtain an abnormal power data characterization vector, an environmental parameter data characterization vector, and an operating parameter data characterization vector; Performing interaction on the abnormal power data characterization vector, the environmental parameter data characterization vector, and the operating parameter data characterization vector to obtain an interaction data characterization vector; Performing abnormal diagnosis on the interaction data characterization vector to obtain a power anomaly diagnosis result, where the power anomaly diagnosis result represents the degree of coincidence of the real-time power data with the abnormal power data; When the power anomaly diagnosis result indicates that the degree of coincidence between the real-time power data and the abnormal power data reaches a preset requirement, perform a power anomaly alarm; Among them, the power anomaly diagnosis result is obtained based on a coincidence degree evaluation network, and the coincidence degree evaluation network includes an abnormal data embedding mapping layer, an environmental parameter embedding mapping layer, and an operating parameter embedding mapping layer; the respectively performing characterization information extraction on the preset power anomaly data stream, the environmental parameter monitoring data stream, and the operating parameter monitoring data stream to obtain an abnormal power data characterization vector, an environmental parameter data characterization vector, and an operating parameter data characterization vector includes: Based on the abnormal data embedding mapping layer, determining first power state characterization information and first distribution characterization information of the preset power anomaly data stream. The first power state characterization information represents the characterization information of multiple data items in the preset power anomaly data stream. The first distribution characterization information represents the characterization information of the distribution of multiple data items in the preset power anomaly data stream; Performing characterization information extraction on the first power state characterization information and the first distribution characterization information to obtain the abnormal power data characterization vector; Based on the environmental parameter embedding mapping layer, performing characterization information extraction on the environmental parameter monitoring data stream to obtain the environmental parameter data characterization vector; Based on the operating parameter embedding mapping layer, performing characterization information extraction on the operating parameter monitoring data stream to obtain the operating parameter data characterization vector.

2. The method according to claim 1, wherein The performing characterization information extraction on the first power state characterization information and the first distribution characterization information to obtain the abnormal power data characterization vector includes: Performing characterization information extraction on the first power state characterization information and the first distribution characterization information to obtain a first anchor characterization vector, a first result characterization vector, and a first search characterization vector; Interact with the first anchor representation vector, the first result representation vector, and the first search representation vector to obtain a candidate abnormal power supply data representation vector; Perform multiple rounds of non-linear transformation on the candidate abnormal power supply data representation vector to obtain the abnormal power supply data representation vector.

3. The method according to claim 1, wherein The extracting the representation information from the environmental parameter monitoring data stream according to the environmental parameter embedding mapping layer to obtain the environmental parameter data representation vector includes: According to the environmental parameter embedding mapping layer, determine the second power supply state representation information and the second distribution representation information of the environmental parameter monitoring data stream, where the second power supply state representation information represents the representation information of multiple data items in the environmental parameter monitoring data stream, and the second distribution representation information represents the representation information of the distribution of multiple data items in the environmental parameter monitoring data stream; Extract the representation information from the second power supply state representation information and the second distribution representation information to obtain the environmental parameter data representation vector.

4. The method according to claim 3, wherein The extracting the representation information from the second power supply state representation information and the second distribution representation information to obtain the environmental parameter data representation vector includes: Extract the representation information from the second power supply state representation information and the second distribution representation information to obtain a second anchor representation vector, a second result representation vector, and a second search representation vector; Interact with the second anchor representation vector, the second result representation vector, and the second search representation vector to obtain a candidate environmental parameter data representation vector; Perform multiple rounds of non-linear transformation on the candidate environmental parameter data representation vector to obtain the environmental parameter data representation vector.

5. The method according to claim 1, characterized in that The extracting the representation information from the operating parameter monitoring data stream according to the operating parameter embedding mapping layer to obtain the operating parameter data representation vector includes: According to the operating parameter embedding mapping layer, determine the third power supply state representation information and the third distribution representation information of the operating parameter monitoring data stream, where the third power supply state representation information represents the representation information of multiple data items in the operating parameter monitoring data stream, and the third distribution representation information represents the representation information of the distribution of multiple data items in the operating parameter monitoring data stream; extract the representation information from the third power supply state representation information and the third distribution representation information to obtain the operating parameter data representation vector.

6. The method according to claim 5, characterized in that, The extracting the representation information from the third power supply state representation information and the third distribution representation information to obtain the operating parameter data representation vector includes: Extract the representation information from the third power supply state representation information and the third distribution representation information to obtain a third anchor representation vector, a third result representation vector, and a third search representation vector; Interact with the third anchor representation vector, the third result representation vector, and the third search representation vector to obtain a candidate operating parameter data representation vector; Perform multiple rounds of non-linear transformation on the candidate operating parameter data representation vector to obtain the operating parameter data representation vector.

7. The method according to any one of claims 1 to 6, characterized in that The fitness evaluation network is debugged through the following steps: Obtain an example preset power anomaly data stream, an example real-time power monitoring data stream, and an actual power anomaly diagnosis result. The example real-time power monitoring data stream includes an example environmental parameter monitoring data stream and an example operating parameter monitoring data stream. The actual power anomaly diagnosis result represents the actual degree of coincidence of the example real-time power monitoring data stream with the example preset power anomaly data stream; According to the coincidence degree evaluation network, perform characterization information extraction on the example preset power anomaly data stream, the example environmental parameter monitoring data stream, and the example operating parameter monitoring data stream respectively to obtain an example abnormal power data characterization vector, an example environmental parameter data characterization vector, and an example operating parameter data characterization vector; perform interaction on the example abnormal power data characterization vector, the example environmental parameter data characterization vector, and the example operating parameter data characterization vector to obtain an example interaction data characterization vector; perform abnormal diagnosis on the example interaction data characterization vector to obtain an example power anomaly diagnosis result; Debug the coincidence degree evaluation network according to the example power anomaly diagnosis result and the actual power anomaly diagnosis result.

8. The method according to claim 7, wherein The environmental parameter embedding mapping layer is a debugged network layer, and the environmental parameter embedding mapping layer is obtained through the following steps: Obtain a first environmental parameter monitoring data stream, and mask the data item information in a local area of the first environmental parameter monitoring data stream to obtain a second environmental parameter monitoring data stream; According to the environmental parameter embedding mapping layer, perform reasoning on the data item information in the masked area of the second environmental parameter monitoring data stream to obtain a second reasoning confidence level. The second reasoning confidence level represents the confidence level that the data items in the masked area of the first environmental parameter monitoring data stream obtained by reasoning belong to each type of data item; Debug the environmental parameter embedding mapping layer according to the second reasoning confidence level and the second actual confidence level. The second actual confidence level represents the actual confidence level that the data items in the masked area of the first environmental parameter monitoring data stream belong to each type of data item; The operating parameter embedding mapping layer is a debugged network layer, and the operating parameter embedding mapping layer is obtained through the following steps: Obtain a first operating parameter monitoring data stream, and mask the data item information in a local area of the first operating parameter monitoring data stream to obtain a second operating parameter monitoring data stream; According to the operating parameter embedding mapping layer, perform reasoning on the data item information in the masked area of the second operating parameter monitoring data stream to obtain a third reasoning confidence level. The third reasoning confidence level represents the confidence level that the data items in the masked area of the first operating parameter monitoring data stream obtained by reasoning belong to each type of data item; Debug the operating parameter embedding mapping layer according to the third reasoning confidence level and the third actual confidence level. The third actual confidence level represents the actual confidence level that the data items in the masked area of the first operating parameter monitoring data stream belong to each type of data item.

9. A computer system, comprising a memory and a processor, characterized in that, The memory stores a computer program that can run on the processor, and when the processor executes the computer program, the method described in any one of claims 1 to 8 is implemented.

Images