Switch cabinet state monitoring method and system based on artificial intelligence

Abstract

The embodiment of the invention provides a switch cabinet state monitoring method and system based on artificial intelligence, and the method comprises the steps: obtaining a state updating activity combination and a classification index combination of a target switch cabinet control system, distinguishing a dominant state activity from a reactive state activity, and determining an activity session type of each state updating activity combination, the operation and maintenance personnel can be helped to manage and monitor the switch cabinet control task more effectively, and the abnormity can be reported in time when the control unit instance is abnormal, so that the possible loss is avoided or reduced. In addition, an anomaly detection mechanism based on an activity session category can more accurately position anomalies, and anomalies in a time domain or an operation domain can be accurately captured and reported to a switch cabinet system operation and maintenance center. Therefore, operation and maintenance personnel can quickly and accurately find problems and take necessary repair measures to ensure stable operation of the switch cabinet control system.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a switch cabinet status monitoring method and system based on artificial intelligence. Background Art

[0002] In traditional switchgear control systems, various status update activities usually require manual monitoring and management. For example, operation and maintenance personnel need to regularly check the status of the switchgear and analyze various status update activities to ensure the normal operation of the switchgear control system. However, this approach has some significant disadvantages.

[0003] First, manual monitoring and management is a time-consuming and inefficient process. In large or complex switchgear control systems, there may be a large number of status update activities, and manual monitoring and management will take up a lot of human resources.

[0004] Secondly, the accuracy of manual monitoring and management is affected by human factors. For example, it may be difficult for operation and maintenance personnel to effectively manage dominant state activities and reactive state activities, or the understanding and processing of certain complex state update activities may not be accurate enough. Therefore, how to improve the intelligence level of switchgear control systems, achieve more efficient and accurate state update activity monitoring and management, and effective anomaly detection, is an urgent problem to be solved in the current field of switchgear control technology. Summary of the invention

[0005] In view of this, the purpose of this application is to provide a switch cabinet status monitoring method and system based on artificial intelligence.

[0006] According to a first aspect of the present application, there is provided a switch cabinet state monitoring method based on artificial intelligence, which is applied to a state monitoring system, and the method comprises: Acquire X state update activity combinations and X classification index combinations of a target switch cabinet control system, where X is a positive integer greater than 1, each of the X state update activity combinations includes a dominant state activity and a reactive state activity of a pending activity session category, the X classification index combinations include activity classification indexes of each dominant state activity and each reactive state activity in the X state update activity combinations, each state update activity combination includes a dominant state activity with an active trigger state and a reactive state activity with a passive trigger state, the dominant state activity includes a dominant element in a target switch cabinet control task, the dominant element includes a preset operation rule or an instruction sequence, the reactive state activity includes a reactive element in the target switch cabinet control task, the reactive element includes a feedback mechanism or an adaptive strategy; Determine X activity session categories of the X state update activity combinations based on X dominant state description features, X first index category description features, X reactive state description features, and X second index category description features, wherein the X dominant state description features are description vectors of the X dominant state activities in the X state update activity combinations, the X first index category description features are index category features of the X dominant state activities, the X reactive state description features are description vectors of the X reactive state activities in the X state update activity combinations, the X first index category description features are index category features of the X dominant state activities, and the X second index category description features are index category features of the X reactive state activities; When an abnormality occurs in a control unit instance in a target node based on any one of the X active session categories, the target instance abnormality information is reported to the switch cabinet system operation and maintenance center. The abnormality occurs in the control unit instance in the target node includes an abnormality in the time domain and / or operation domain of the control unit instance response, and the target instance abnormality information reflects that an abnormality occurs in the control unit instance in the target node.

[0007] In a possible implementation of the first aspect, determining X activity session categories of the X status update activity combinations based on X dominant status description features, X first index category description features, X reactive status description features, and X second index category description features includes: Obtaining description vectors of X dominant state activities in the X state update activity combinations, generating X dominant state description features, obtaining description vectors of X reactive state activities in the X state update activity combinations, generating X reactive state description features, obtaining index category features of activity classification indexes of the X dominant state activities in the X classification index combinations, generating X first index category description features, and obtaining index category features of activity classification indexes of the X reactive state activities in the X classification index combinations, generating X second index category description features; The X dominant state description features, the X first index category description features, the X reactive state description features, and the X second index category description features are merged to generate X target feature representations corresponding to the X state update activity combinations; Based on the X target feature representations, X activity session categories of the X state update activity combinations are determined, each of the X activity session categories reflecting a session relationship between the dominant state activity and the reactive state activity in a corresponding one of the X state update activity combinations.

[0008] In a possible implementation manner of the first aspect, the acquiring description vectors of X dominant state activities in the X state update activity combinations to generate X dominant state description features includes: The description vector of the rth dominant state activity among the X dominant state activities is obtained based on the following operation, and the rth dominant state description feature is generated, where r is a positive integer not less than 1 and not greater than X: When the r-th dominant state activity includes Yr state behaviors, obtaining a behavior feature representation of each of the Yr state behaviors to generate Yr behavior feature representations, where Yr is a positive integer; The Yr behavior feature representations are subjected to equalization processing to generate the r-th dominant state description feature; or, the Yr behavior feature representations are subjected to weight fusion processing to generate the r-th dominant state description feature.

[0009] In a possible implementation manner of the first aspect, the acquiring description vectors of X reactive state activities in the X state update activity combinations and generating X reactive state description features includes: The description vector of the rth reactive state activity among the X reactive state activities is obtained based on the following operation, and the rth reactive state description feature is generated, where r is a positive integer not less than 1 and not greater than X: When the rth reactive state activity includes Zr state behaviors, obtaining a behavior feature representation of each of the Zr state behaviors, generating Zr behavior feature representations, where Zr is a positive integer; The Zr behavior feature representations are subjected to equalization processing to generate the rth reactive state description feature; or, weight fusion is performed on the Zr behavior feature representations to generate the rth reactive state description feature.

[0010] In a possible implementation manner of the first aspect, acquiring index category features of the activity classification indexes of the X dominant state activities in the X classification index combinations to generate X first index category description features includes: The index category feature of the activity classification index of the rth dominant state activity among the X dominant state activities is obtained based on the following operation, and the rth first index category description feature is generated, where r is a positive integer not less than 1 and not greater than X: When the activity classification index of the rth dominant state activity includes the rth index description field, the field feature representation of the rth index description field is obtained, and the rth first index category description feature is the field feature representation of the rth index description field.

[0011] In a possible implementation of the first aspect, the acquiring index category features of the activity classification indexes of the X reactive state activities in the X classification index combinations to generate X second index category description features includes: The index category feature of the activity classification index of the rth reactive state activity among the X reactive state activities is obtained based on the following operation, and the rth second index category description feature is generated, where r is a positive integer not less than 1 and not greater than X: When the activity classification index of the rth reactive state activity includes the rth index description field, the field feature representation of the rth index description field is obtained, and the rth second index category description feature is the field feature representation of the rth index description field.

[0012] In a possible implementation of the first aspect, the fusing the X dominant state description features, the X first index category description features, the X reactive state description features, and the X second index category description features to generate X target feature representations corresponding to the X state update activity combinations includes: The rth target feature representation corresponding to the rth state update activity combination among the X state update activity combinations is obtained based on the following operation, wherein the rth state update activity combination includes the rth dominant state activity and the rth reactive state activity, and r is a positive integer not less than 1 and not greater than X: The task trigger description vector, the rth dominant state description feature, the rth first index category description feature, the rth reactive state description feature, and the rth second index category description feature are merged to generate the rth target feature representation, wherein the task trigger description vector reflects the trigger node of the target switch cabinet control task, the state update activity in the X state update activity combinations is the state update activity parsed from the target switch cabinet control task, the rth dominant state description feature is the description vector of the rth dominant state activity, the rth first index category description feature is the index category feature of the activity classification index of the rth dominant state activity, the rth reactive state description feature is the description vector of the rth reactive state activity, and the rth second index category description feature is the index category feature of the activity classification index of the rth reactive state activity; Among them, when the dimension of the task trigger description vector, the dimension of the r-th dominant state description feature, the dimension of the r-th first index category description feature, the dimension of the r-th reactive state description feature and the dimension of the r-th second index category description feature are all 1*K, the task trigger description vector, the r-th dominant state description feature, the r-th first index category description feature, the r-th reactive state description feature and the r-th second index category description feature are fused to generate the r-th target feature representation with a dimension of 5*K, where K is a positive integer greater than 1.

[0013] In a possible implementation manner of the first aspect, determining, based on the X target feature representations, X activity session categories of the X status update activity combinations includes: Determine an rth activity session category of an rth status update activity combination among the X status update activity combinations based on an rth target feature representation among the X target feature representations based on the following operations, where r is a positive integer not less than 1 and not greater than X: Loading the r-th target feature representation into the AI ​​neural network, generating M session category classification data corresponding to the M active session categories, the M session category classification data reflecting the possibility that the r-th active session category is each candidate session category among the M set active session categories, where M is a positive integer greater than 1; The rth activity session category is determined to be equal to a target activity session category among the M set activity session categories, and among the M session category classification data, the rth activity session category has the highest probability of being the target activity session category.

[0014] In a possible implementation manner of the first aspect, determining, based on the X target feature representations, X activity session categories of the X status update activity combinations includes: The X target feature representations are loaded into the AI ​​neural network in rounds to generate a conversation category classification data sequence corresponding to the target feature representations of each round, wherein the target feature representations of each round include J target feature representations among the X target feature representations, and the conversation category classification data sequence corresponding to the target feature representations of each round includes J groups of conversation category classification data, and each group of conversation category classification data among the J groups of conversation category classification data includes M conversation category classification data, and the M conversation category classification data covered by the s-th group of conversation category classification data reflect the possibility that the s-th active conversation category corresponding to the s-th target feature representation is each candidate conversation category among the M set active conversation categories, and the s-th group of conversation category classification data corresponds to the s-th target feature representation among the J target feature representations, where M is a positive integer greater than 1, J is a positive integer not less than 2 and less than X, and s is a positive integer not less than 1 and not greater than J; Determine the sth active session category among J active session categories based on the sth group of session category classification data in the sequence of session category classification data by using the following operations: The sth active session category is determined to be equal to the target active session category among the M set active session categories, and among the M session category classification data covered by the sth group of session category classification data, the sth active session category is most likely to be the target active session category.

[0015] For example, in a possible implementation of the first aspect, obtaining X state update activity combinations and X classification index combinations of the target switch cabinet control system includes: Parsing H state update activities and H activity classification indexes having unique mapping relationships in the target switch cabinet control task, wherein the H activity classification indexes include the activity classification index of each of the H state update activities, and each of the H state update activities includes an element in the target switch cabinet control task; The two state update activities in X groups of the H state update activities form the X state update activity combinations, wherein each state update activity combination in the X state update activity combinations includes two state update activities in one group of the H state update activities, the state update activity in the two state update activities in the one group that has an active trigger state is the dominant state activity, and the state update activity in the passive trigger state is the reactive state activity; Determine, from the H activity classification indexes, the activity classification indexes of the dominant state activity and the reactive state activity covered in each state update activity combination in the X state update activity combinations, to generate the X classification index combinations; Among them, the session relationship between the dominant state activity and the reactive state activity represented by each of the X active session categories is one of the set M session relationships, M is a positive integer greater than 1, and the M session relationships include M-1 set session relationships and no session relationship.

[0016] According to a second aspect of the present application, a state monitoring system is provided, which includes a processor and a readable storage medium, wherein the readable storage medium stores a program, and when the program is executed by the processor, the aforementioned switch cabinet state monitoring method based on artificial intelligence is implemented.

[0017] According to a third aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer-executable instructions, and when it is monitored that the computer-executable instructions are executed, the aforementioned switch cabinet status monitoring method based on artificial intelligence is implemented.

[0018] According to any of the above aspects, in the present application, by obtaining the state update activity combination and its classification index combination of the target switch cabinet control system, and distinguishing between dominant state activities and reactive state activities, the activity session category of each state update activity combination is determined, which can not only help the operation and maintenance personnel to manage and monitor the switch cabinet control tasks more effectively, but also report the abnormality in time when the control unit instance is abnormal, thereby avoiding or reducing the possible losses. In addition, the abnormality detection mechanism based on the activity session category can locate the abnormality more accurately, and the abnormality in both the time domain and the operation domain can be accurately captured and reported to the switch cabinet system operation and maintenance center. In this way, the operation and maintenance personnel can quickly and accurately find the problem, take necessary repair measures, and ensure the stable operation of the switch cabinet control system. That is, the present application significantly improves the intelligence and operation efficiency of the switch cabinet control system, and also provides a powerful tool for accurate abnormality detection and rapid fault repair. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 A schematic diagram of a flow chart of a switch cabinet state monitoring method based on artificial intelligence provided in an embodiment of the present application; Figure 2A schematic diagram of the component structure of a state monitoring system for implementing the above-mentioned switch cabinet state monitoring method based on artificial intelligence provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present application.

[0022] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] Figure 1 The flowchart of the switch cabinet state monitoring method based on artificial intelligence provided by the embodiment of the present application is shown. It should be understood that in other embodiments, the order of some steps of the switch cabinet state monitoring method based on artificial intelligence in this embodiment can be interchanged according to actual needs, or some steps can be omitted or deleted. The detailed steps of the switch cabinet state monitoring method based on artificial intelligence are introduced as follows.

[0024] Step S110, obtaining X state update activity combinations and X classification index combinations of the target switch cabinet control system.

[0025] In this embodiment, X is a positive integer greater than 1, each of the X state update activity combinations includes a dominant state activity and a reactive state activity of a pending activity session category, the X classification index combinations include activity classification indexes of each dominant state activity and each reactive state activity in the X state update activity combinations, each state update activity combination includes a dominant state activity in which an active trigger state exists and a reactive state activity in which a passive trigger state exists, the dominant state activity includes a dominant element in a target switch cabinet control task, the dominant element includes a preset operation rule or an instruction sequence, the reactive state activity includes a reactive element in the target switch cabinet control task, and the reactive element includes a feedback mechanism or an adaptive strategy.

[0026] For example, suppose in a switch cabinet control system, X=3, and three state update activity combinations need to be processed. Each state update activity combination contains a dominant state activity and a reactive state activity.

[0027] For example, the first state update activity combination may involve load dispatch. The dominant state activities include grid load balancing operations performed according to preset operation rules or instruction sequences, such as power dispatch plans arranged in advance according to prediction algorithms; reactive state activities may be real-time power adjustments based on feedback mechanisms or adaptive strategies when there is a large gap between the actual grid load and the prediction. Both dominant state activities and reactive state activities will have their own classification indexes to identify the specific categories or states of the dominant state activities and reactive state activities.

[0028] For example, in a power switch cabinet control system, there may be a target task of "maintaining grid load balance during peak hours". In order to complete this task, a series of state update activities may be required.

[0029] The dominant state activity involves performing tasks according to a preset operation rule or instruction sequence. In this example, the dominant element may be "pre-set the switchgear opening and closing plan based on the power demand forecast and grid capacity." This preset operation rule or instruction sequence will determine which switchgear needs to be opened and which needs to be closed during peak hours, and when to perform these operations.

[0030] Reactive state activities include making adjustments based on feedback mechanisms or adaptive strategies when the actual situation deviates from the preset plan. In this example, the reactive element may be "If the actual power demand exceeds the forecast, or a switch cabinet fails, the system needs to immediately adjust the status of other switch cabinets to maintain the grid load balance." This feedback mechanism or adaptive strategy will ensure that the system can flexibly respond to changes in the actual situation and always maintain the stable operation of the grid.

[0031] In this process, the dominant state activities and reactive state activities work together to achieve the target tasks of the switchgear control system.

[0032] Step S120: determining X activity session categories of the X state update activity combinations based on the X dominant state description features, the X first index category description features, the X reactive state description features, and the X second index category description features.

[0033] In this embodiment, the X dominant state description features are description vectors of the X dominant state activities in the X state update activity combinations, the X first index category description features are index category features of the X dominant state activities, the X reactive state description features are description vectors of the X reactive state activities in the X state update activity combinations, the X first index category description features are index category features of the X dominant state activities, and the X second index category description features are index category features of the X reactive state activities.

[0034] For example, when determining the activity session category, it is necessary to determine the activity session category of each state update activity combination based on X dominant state description features, X first index category description features, X reactive state description features, and X second index category description features.

[0035] For example, for the state update activity combination of load scheduling, the following characteristics need to be considered: Dominant state description feature: This is a description of the dominant state activity. Each dominant state activity has an associated description vector, which contains various information describing the dominant state activity. For example, for the aforementioned dominant state activity "pre-set the switchgear opening and closing plan according to the power demand forecast and grid capacity", its description features may include preset operation rules (how to set the switchgear opening and closing plan), instruction sequence (specific opening and closing plan), execution time, etc.

[0036] The first index category describes the characteristics: This is the classification of the dominant state activity. It describes the category or state to which the dominant state activity belongs. For example, the dominant state activity may be classified into different types such as "load scheduling" and "equipment inspection", and may also be marked as different execution states such as "in progress" and "completed".

[0037] Reactive state description feature: This is a description of the reactive state activity. Each reactive state activity has an associated description vector, which contains various information describing the reactive state activity. For example, for the aforementioned reactive state activity "If the actual power demand exceeds the forecast, or a switch cabinet fails, it is necessary to immediately adjust the state of other switch cabinets to maintain the load balance of the power grid", its description features may include feedback mechanism (how to adjust the state of the switch cabinet according to the actual situation), adaptive strategy (when to start the adaptive strategy, how to adjust), execution results, etc.

[0038] The second index category describes the characteristics: This is the classification of reactive state activities. It describes the category or state to which the reactive state activity belongs. For example, reactive state activities may be classified into different types such as "load adjustment" and "fault handling", and may also be marked as different execution states such as "in progress" and "completed".

[0039] Through these characteristics, each combination of status update activities can be more accurately understood and managed for subsequent monitoring and management.

[0040] Step S130: When an abnormality occurs in a control unit instance in a target node based on any one of the X active session categories, abnormal information of the target instance is reported to a switch cabinet system operation and maintenance center.

[0041] In this embodiment, the abnormality of the control unit instance in the target node includes an abnormality in the time domain and / or operation domain of the control unit instance response, and the target instance abnormality information reflects that the abnormality of the control unit instance in the target node occurs.

[0042] For example, in step S120, the activity session category is determined by the dominant state description feature, the first index category description feature, the reactive state description feature, and the second index category description feature.

[0043] The dominant state description features and the first index category description features are used to describe and classify dominant state activities, while the reactive state description features and the second index category description features are used to describe and classify reactive state activities. Each state update activity combination (including a dominant state activity and a reactive state activity) will be assigned to a specific activity session category based on these features.

[0044] For example, if the target feature representation of a state update activity combination indicates that its dominant state activity is in progress and the reactive state activity has been completed, it may be assigned to an activity session category called "normal operation". If the dominant state activity fails and needs to be handled immediately, it may be assigned to the activity session category of "abnormal repair". If an emergency situation is detected, such as the grid load exceeding a threshold, and the reactive state activities fail to respond in time, it may be assigned to the activity session category of "emergency handling".

[0045] When an abnormality occurs in the control unit instance in the target node characterized based on any one of the X active session categories (such as the occurrence of an "abnormal repair" or "emergency processing" active session category), this means that if an abnormality occurs in the control unit instance when executing a state update activity combination corresponding to a certain active session category, for example, it fails to dispatch electricity according to preset operating rules, or cannot correctly respond to real-time changes in grid load, then this abnormality is regarded as an abnormality under this active session category.

[0046] Exemplarily, the target node may refer to a specific switch cabinet or a group of switch cabinets, while the control unit instance is a device or system responsible for performing specific tasks (such as load scheduling, power grid monitoring, etc.). Time domain anomalies are usually related to time. For example, if a control unit instance needs to complete a specific state update activity within a specified time, but the actual execution time exceeds expectations, then it may be considered that a time domain anomaly has occurred. Another example is that the reaction time of the control unit instance is significantly slower, or the execution frequency of its state update activities has changed abnormally. Operational domain anomalies are usually related to the function or operation of the control unit instance. For example, if a control unit instance responsible for load scheduling fails to distribute power correctly, or a control unit instance responsible for power grid monitoring cannot accurately read the power grid state, then it may be considered that an operational domain anomaly has occurred. The target instance anomaly information reflects that the control unit instance in the target node has an anomaly, which means that when any of the above anomalies occur in the control unit instance, corresponding anomaly information will be generated. These anomaly information may include the time when the anomaly occurred, the specific anomaly type (time domain or operation domain), the state information of the control unit instance, the combination of state update activities being executed when the anomaly occurred, etc.

[0047] By analyzing these abnormal information, operation and maintenance personnel can locate the problem more accurately and take appropriate measures to repair it in time, thereby ensuring the normal operation of the switch cabinet control system.

[0048] Therefore, under any active session category, as long as the control unit instance has an exception when executing the state update activity combination corresponding to the active session category, the exception information needs to be reported to the switch cabinet system operation and maintenance center. This exception information may include the response time of the control unit instance, operation details, etc., which can help the operation and maintenance personnel understand and handle the exception more accurately.

[0049] Based on the above steps, by obtaining the state update activity combination and its classification index combination of the target switch cabinet control system, and distinguishing between dominant state activities and reactive state activities, the activity session category of each state update activity combination is determined, which can not only help the operation and maintenance personnel to manage and monitor the switch cabinet control tasks more effectively, but also report the abnormalities in time when the control unit instance is abnormal, thereby avoiding or reducing the possible losses. In addition, the abnormality detection mechanism based on the activity session category can locate the abnormality more accurately, and the abnormalities in both the time domain and the operation domain can be accurately captured and reported to the switch cabinet system operation and maintenance center. In this way, the operation and maintenance personnel can quickly and accurately find the problem, take necessary repair measures, and ensure the stable operation of the switch cabinet control system. That is, the present application significantly improves the intelligence and operation efficiency of the switch cabinet control system, and also provides a powerful tool for accurate abnormality detection and rapid fault repair.

[0050] In a possible implementation, step S120 may include: Step S121, obtaining description vectors of X dominant state activities in the X state update activity combinations, generating X dominant state description features, obtaining description vectors of X reactive state activities in the X state update activity combinations, generating X reactive state description features, obtaining index category features of activity classification indexes of the X dominant state activities in the X classification index combinations, generating X first index category description features, and obtaining index category features of activity classification indexes of the X reactive state activities in the X classification index combinations, generating X second index category description features.

[0051] Assume that a power switch cabinet control system is running, in which there are two state update activity combinations (ie, X=2). Each state update activity combination includes a dominant state activity and a reactive state activity.

[0052] First, description vectors of the dominant state activities and reactive state activities in the two state update activity combinations can be obtained. For example, the first state update activity combination may involve the dominant state activity "dispatching electricity according to preset operation rules" and the reactive state activity "real-time power adjustment according to actual grid load changes"; the second state update activity combination may involve the dominant state activity "regularly checking the switch cabinet status" and the reactive state activity "starting the repair program immediately after the fault is found". Then, the dominant state description features and reactive state description features can be generated based on these description vectors. In addition, the index category features of the activity classification index corresponding to each state activity can also be obtained, and the first index category description features and the second index category description features can be generated.

[0053] Step S122: fusing the X dominant state description features, the X first index category description features, the X reactive state description features, and the X second index category description features to generate X target feature representations corresponding to the X state update activity combinations.

[0054] This fusion process may involve concatenating all of these feature vectors together, or using more complex methods such as neural networks to produce a comprehensive feature representation that contains all the relevant information.

[0055] Step S123: determining X activity session categories of the X state update activity combinations based on the X target feature representations, each of the X activity session categories reflecting a session relationship between the dominant state activity and the reactive state activity in a corresponding state update activity combination of the X state update activity combinations.

[0056] For example, if the target feature representation of a state update activity combination indicates that its dominant state activity is in progress and the reactive state activity has been completed, it may be assigned to an activity session category called "normal operation". If the dominant state activity fails and needs to be handled immediately, it may be assigned to the activity session category of "abnormal repair". If the system detects an emergency, such as the grid load exceeding a threshold, and the reactive state activities fail to respond in time, it may be assigned to the activity session category of "emergency handling".

[0057] Therefore, through the above steps, the activity session category of each state update activity combination can be determined based on the dominant state description feature, the first index category description feature, the reactive state description feature and the second index category description feature.

[0058] In a possible implementation, in step S121, obtaining description vectors of X dominant state activities in the X state update activity combinations and generating X dominant state description features may include: The description vector of the rth dominant state activity among the X dominant state activities is obtained based on the following operation, and the rth dominant state description feature is generated, where r is a positive integer not less than 1 and not greater than X: Step S1211, when the r-th dominant state activity includes Yr state behaviors, obtain a behavior feature representation of each of the Yr state behaviors, and generate Yr behavior feature representations, where Yr is a positive integer.

[0059] For example, suppose that a power switch cabinet control system is running, in which there are two state update activity combinations (i.e., X=2). Each state update activity combination includes a dominant state activity and a reactive state activity. For example, the first dominant state activity may be "dispatching electricity according to preset operating rules", which includes Y1=3 state behaviors: calculating preset operating rules, setting power dispatch parameters, and starting power dispatch. Each state behavior has a behavior feature representation. For example, the behavior feature representation of calculating preset operating rules may include calculation time, type of algorithm used, etc. After obtaining the behavior feature representations of these three state behaviors, three behavior feature representations are obtained.

[0060] Step S1212: Perform equalization processing on the Yr behavior feature representations to generate the rth dominant state description feature. Alternatively, perform weight fusion on the Yr behavior feature representations to generate the rth dominant state description feature.

[0061] Next, these three behavioral feature representations are balanced or weighted. Balanced processing may mean normalizing these three feature representations so that each feature representation has the same influence in the entire description feature; while weighted fusion may mean assigning different weights according to the importance of each state behavior, and then fusing the weighted feature representations together. Regardless of which method is used, the first dominant state description feature can be generated in the end.

[0062] Similarly, the second dominant state activity may be "regularly check the switch cabinet status", which includes Y2=2 state behaviors: start the inspection program and analyze the inspection results. After obtaining the behavioral feature representations of these two state behaviors, two behavioral feature representations are obtained. Then, the same equalization processing or weight fusion is performed to finally generate the second dominant state description feature.

[0063] Therefore, through the above steps, the corresponding dominant state description features can be generated based on the behavior feature representation of the state behavior in each dominant state activity.

[0064] In a possible implementation, in step S121, the description vectors of X reactive state activities in the X state update activity combinations are obtained to generate X reactive state description features, including: The description vector of the rth reactive state activity among the X reactive state activities is obtained based on the following operation, and the rth reactive state description feature is generated, where r is a positive integer not less than 1 and not greater than X: Step S1213, when the rth reactive state activity includes Zr state behaviors, obtain a behavior feature representation of each of the Zr state behaviors, and generate Zr behavior feature representations, where Zr is a positive integer.

[0065] For example, continue to use the example of the power switch cabinet control system to explain these steps. Similarly, assume that there are two state update activity combinations (ie, X=2), each of which includes a dominant state activity and a reactive state activity.

[0066] First, it is necessary to obtain the description vector of each reactive state activity. For example, the first reactive state activity may be "real-time power adjustment according to actual grid load changes", which includes Z1=3 state behaviors: reading grid load data, calculating load adjustment value, and executing load adjustment. Each state behavior has a behavior characteristic representation. For example, the behavior characteristic representation of reading grid load data may include data reading time, data volume read, etc. After obtaining the behavior characteristic representations of these three state behaviors, three behavior characteristic representations are obtained.

[0067] Step S1214: Perform equalization processing on the Zr behavior feature representations to generate the rth reactive state description feature. Alternatively, perform weight fusion on the Zr behavior feature representations to generate the rth reactive state description feature.

[0068] Next, these three behavioral feature representations are balanced or weighted. Balanced processing may mean normalizing these three feature representations so that each feature representation has the same influence in the entire description feature; while weighted fusion may mean assigning different weights according to the importance of each state behavior, and then fusing the weighted feature representations together. Regardless of which method is used, the first reactive state description feature can be generated in the end.

[0069] Similarly, the second reactive state activity is processed**: The second reactive state activity may be "start the repair program immediately after the fault is found", which contains Z2=2 state behaviors: detecting the switch cabinet fault and starting the repair program. After obtaining the behavioral feature representations of these two state behaviors, two behavioral feature representations are obtained. Then, the same balancing process or weight fusion is performed, and finally the second reactive state description feature is generated.

[0070] Therefore, through the above steps, the corresponding reactive state description features can be generated based on the behavioral feature representation of the state behavior in each reactive state activity.

[0071] In a possible implementation, in step S121, obtaining index category features of the activity classification indexes of the X dominant state activities in the X classification index combinations to generate X first index category description features includes: The index category feature of the activity classification index of the rth dominant state activity among the X dominant state activities is obtained based on the following operation, and the rth first index category description feature is generated, where r is a positive integer not less than 1 and not greater than X: Step S1215, when the activity classification index of the rth dominant state activity includes the rth index description field, obtain the field feature representation of the rth index description field, and the rth first index category description feature is the field feature representation of the rth index description field.

[0072] For example, let's use the previous power switch cabinet control system as an example. Assume that there are two state update activity combinations (ie, X=2), each of which includes a dominant state activity and a reactive state activity.

[0073] First, it is necessary to obtain the index category features of the activity classification index of each dominant state activity. For example, the first dominant state activity may be "dispatching electricity according to preset operation rules", and its activity classification index may include such an index description field: "load dispatch". At this time, the field feature representation may include the semantic information of the field, the correlation with other fields, etc. In this example, the first index category description feature of the first dominant state activity is the field feature representation of "load dispatch".

[0074] Similarly, the second dominant status activity is processed: the second dominant status activity may be "regularly check the switch cabinet status", and its activity classification index may include such an index description field: "status check". Similarly, the feature representation of this field is obtained as the first index category description feature of the second dominant status activity.

[0075] Therefore, through the above steps, the corresponding first index category description feature can be generated based on the index description field of the activity classification index of each dominant state activity.

[0076] In a possible implementation, in step S121, obtaining index category features of the activity classification indexes of the X reactive state activities in the X classification index combinations and generating X second index category description features include: The index category feature of the activity classification index of the rth reactive state activity among the X reactive state activities is obtained based on the following operation, and the rth second index category description feature is generated, where r is a positive integer not less than 1 and not greater than X: Step S1216, when the activity classification index of the rth reactive state activity includes the rth index description field, obtain the field feature representation of the rth index description field, and the rth second index category description feature is the field feature representation of the rth index description field.

[0077] For example, continue to use the example of the power switch cabinet control system to explain these steps. Assume that there are two state update activity combinations (ie, X=2), each of which includes a dominant state activity and a reactive state activity.

[0078] First, it is necessary to obtain the index category features of the activity classification index of each reactive state activity. For example, the first reactive state activity may be "real-time power adjustment according to actual grid load changes", and its activity classification index may include such an index description field: "real-time adjustment". At this time, the field feature representation may include the semantic information of the field, the correlation with other fields, etc. In this example, the second index category description feature of the first reactive state activity is the field feature representation of "real-time adjustment".

[0079] Similarly, the second reactive state activity is processed: the second reactive state activity may be "starting the repair program immediately after discovering the fault", and its activity classification index may include such an index description field: "fault repair". Similarly, the feature representation of this field is obtained as the second index category description feature of the second reactive state activity.

[0080] Therefore, through the above steps, the corresponding second index category description feature can be generated based on the index description field of the activity classification index of each reactive state activity.

[0081] In a possible implementation, step S122 may include: The rth target feature representation corresponding to the rth state update activity combination among the X state update activity combinations is obtained based on the following operation, wherein the rth state update activity combination includes the rth dominant state activity and the rth reactive state activity, and r is a positive integer not less than 1 and not greater than X: Step S1221, the task trigger description vector, the rth dominant state description feature, the rth first index category description feature, the rth reactive state description feature, and the rth second index category description feature are merged to generate the rth target feature representation, wherein the task trigger description vector reflects the trigger node of the target switch cabinet control task, the state update activity in the X state update activity combinations is the state update activity parsed from the target switch cabinet control task, the rth dominant state description feature is the description vector of the rth dominant state activity, the rth first index category description feature is the index category feature of the activity classification index of the rth dominant state activity, the rth reactive state description feature is the description vector of the rth reactive state activity, and the rth second index category description feature is the index category feature of the activity classification index of the rth reactive state activity.

[0082] Among them, when the dimension of the task trigger description vector, the dimension of the r-th dominant state description feature, the dimension of the r-th first index category description feature, the dimension of the r-th reactive state description feature and the dimension of the r-th second index category description feature are all 1*K, the task trigger description vector, the r-th dominant state description feature, the r-th first index category description feature, the r-th reactive state description feature and the r-th second index category description feature are fused to generate the r-th target feature representation with a dimension of 5*K, where K is a positive integer greater than 1.

[0083] For example, continue to use the example of the power switch cabinet control system to explain these steps. Assume that there are two state update activity combinations (ie, X=2), each of which includes a dominant state activity and a reactive state activity.

[0084] First, it is necessary to obtain the task trigger description vector, which reflects the trigger node of the target switch cabinet control task. For example, if a target switch cabinet control task is triggered by a timer, the task trigger description vector may contain information such as the trigger time and the trigger condition.

[0085] Then, the task trigger description vector, the rth dominant state description feature, the rth first index category description feature, the rth reactive state description feature, and the rth second index category description feature are fused to generate the rth target feature representation. In this example, assuming that all these features are 1*K vectors, the fused target feature representation is a 5*K matrix. For example, if K=3, the target feature representation may be represented in the following format: [Task trigger description vector] [Dominant state description characteristics] [First Index Category Description Features] [Reactive state description features] [Second index category description features] Each row is a 1*3 vector, representing the corresponding feature. In this way, the target feature representation of each state update activity combination is obtained.

[0086] Therefore, through the above steps, the target feature representation of each state update activity combination can be generated based on the task trigger description vector, the dominant state description feature, the first index category description feature, the reactive state description feature and the second index category description feature.

[0087] In a possible implementation, step S123 may include: Determine an rth activity session category of an rth status update activity combination among the X status update activity combinations based on an rth target feature representation among the X target feature representations based on the following operations, where r is a positive integer not less than 1 and not greater than X: Step S1231: Load the rth target feature representation into the AI ​​neural network to generate M session category classification data corresponding to the M active session categories, wherein the M session category classification data reflect the possibility that the rth active session category is each candidate session category among the M set active session categories, where M is a positive integer greater than 1.

[0088] For example, in the example of a power switchgear control system, it is possible to imagine several pre-set activity session categories (M), such as "normal operation", "abnormal repair" and "emergency processing". These categories describe the possible execution situations of a state update activity combination.

[0089] First, the target feature representation is loaded into the AI ​​neural network. For example, the target feature representation of the first state update activity combination is input into a pre-trained AI neural network, which can output the probabilities of M activity session categories based on the input feature representation. This means that the AI ​​neural network will predict which activity session category the state update activity combination is most likely to belong to.

[0090] For example, suppose the probability output by the AI ​​neural network is [0.7, 0.2, 0.1], which means that the AI ​​neural network predicts that the first status update activity combination has a 70% probability of being "normal operation", a 20% probability of being "abnormal repair", and a 10% probability of being "emergency processing".

[0091] Step S1232: Determine the rth active session category to be equal to the target active session category among the M set active session categories, and among the M session category classification data, the rth active session category has the greatest possibility of being the target active session category.

[0092] Next, the activity session category with the highest probability is selected as the activity session category of the state update activity combination. In this example, because the probability of "normal operation" is the highest (70%), the activity session category of the first state update activity combination is determined as "normal operation".

[0093] The same process is applied to the second status update activity combination. For example, if the second status update activity combination is predicted to have an 80% probability of being "abnormal repair", then its activity session category is determined as "abnormal repair".

[0094] Therefore, through the above steps, the activity session category of each status update activity combination can be determined based on the target feature representation and the prediction of the AI ​​neural network.

[0095] In a possible implementation, step S123 may further include: Step S1233: Load the X target feature representations into the AI ​​neural network in rounds to generate a conversation category classification data sequence corresponding to the target feature representations of each round, wherein the target feature representations of each round include J target feature representations among the X target feature representations, and the conversation category classification data sequence corresponding to the target feature representations of each round includes J groups of conversation category classification data, each group of conversation category classification data in the J groups of conversation category classification data includes M conversation category classification data, and the M conversation category classification data covered by the s-th group of conversation category classification data reflect the possibility that the s-th active conversation category corresponding to the s-th target feature representation is each candidate conversation category among the M set active conversation categories, and the s-th group of conversation category classification data corresponds to the s-th target feature representation among the J target feature representations, where M is a positive integer greater than 1, J is a positive integer not less than 2 and less than X, and s is a positive integer not less than 1 and not greater than J.

[0096] Determine the sth active session category among J active session categories based on the sth group of session category classification data in the sequence of session category classification data by using the following operations: Step S1234, determining the sth active session category to be equal to the target active session category among the M set active session categories, and among the M session category classification data covered by the sth group of session category classification data, the sth active session category is most likely to be the target active session category.

[0097] In the example of the power switch cabinet control system, it is assumed that there are three state update activity combinations (i.e., X=3), each of which includes a dominant state activity and a reactive state activity. In addition, three preset activity session categories (M) are set, such as "normal operation", "abnormal repair" and "emergency processing". Now let's explain it step by step: Assume that each round only processes the target feature representation of two status update activity combinations (i.e., J=2). In the first round, the target feature representation of the first two status update activity combinations is input into the AI ​​neural network to obtain two sets of session category classification data, each of which contains three elements (corresponding to three preset activity session categories). For example, the first set of data may be [0.6, 0.3, 0.1], indicating that the probability of the first status update activity combination being "normal operation" is 60%, the probability of being "abnormal repair" is 30%, and the probability of being "emergency processing" is 10%.

[0098] In the second round, the target feature representation of the third state update activity combination is input into the AI ​​neural network to obtain another set of session category classification data. In this way, a sequence of session category classification data is obtained.

[0099] Next, the corresponding activity session category is determined based on each set of session category classification data. The activity session category with the highest probability in each set of data is selected as the activity session category of the state update activity combination. In the first round, because the probability of "normal operation" is the highest (60%), the activity session category of the first state update activity combination is determined as "normal operation".

[0100] Similarly, in the second round, the activity session category with the highest probability is selected as the activity session category of the third state update activity combination.

[0101] In this way, the activity session category of each state update activity combination is determined in turns based on the target feature representation and the prediction of the AI ​​neural network.

[0102] It is worth noting that AI neural networks can be trained through supervised learning. First, a training sample set containing a large amount of historical data is required, and each training sample contains a feature representation of a state update activity combination and a corresponding activity session category label.

[0103] Here are some specific steps of the training process: First, the raw data needs to be preprocessed, which may include steps such as cleaning the data, handling missing values, encoding categorical variables, etc. For example, the descriptions of the dominant state activity and the reactive state activity may need to be converted into numerical features so that the neural network can process them.

[0104] Next, we need to select or extract features that are useful for predicting the activity session category. These features may include the type of dominant state activity, execution state, response time of reactive state activity, etc.

[0105] Then, you need to design the structure of the neural network, including choosing the number of hidden layers, the number of nodes in each layer, the activation function, etc. In addition, you also need to choose a suitable loss function and optimizer.

[0106] In the training phase, the training samples of the training sample set are input into the neural network, which predicts the active session category based on the target feature representation of each training sample, and then calculates the loss based on the predicted results and the actual active session category label. Then, the optimizer is used to adjust the network parameters to minimize the loss.

[0107] In addition, it is necessary to evaluate the performance of the model on a separate validation set and make necessary parameter adjustments. Finally, the final performance of the neural network is verified on the test set.

[0108] After the training is completed, an AI neural network is obtained that can predict the activity session category of the status update activity combination. When a new status update activity combination appears, its target feature representation can be input into the neural network to obtain the corresponding activity session category.

[0109] For example, in a possible implementation, step S110 may include: Step S111, parsing H state update activities and H activity classification indexes that have unique mapping relationships in the target switch cabinet control task, the H activity classification indexes including the activity classification index of each of the H state update activities, and each of the H state update activities including an element in the target switch cabinet control task.

[0110] Step S112, the two status update activities in X groups of the H status update activities are combined into the X status update activity combinations, each of the X status update activity combinations includes two status update activities in one group of the H status update activities, the status update activity in the one group that has an active trigger state is the dominant status activity, and the status update activity in the passive trigger state is the reactive status activity.

[0111] Step S113: determining the activity classification index of the dominant state activity and the reactive state activity covered in each state update activity combination in the X state update activity combinations from the H activity classification indexes, and generating the X classification index combinations.

[0112] Among them, the session relationship between the dominant state activity and the reactive state activity represented by each of the X active session categories is one of the set M session relationships, M is a positive integer greater than 1, and the M session relationships include M-1 set session relationships and no session relationship.

[0113] For example, it can be assumed that a target switchgear control task contains H state update activities and corresponding H activity classification indexes. For example, these state update activities may include load dispatching, power grid monitoring, fault repair, etc., and each activity has a corresponding activity classification index.

[0114] First, H state update activities and corresponding H activity classification indexes are parsed from the target switchgear control task. For example, if one element in the target switchgear control task is "execute load scheduling", then the corresponding state update activity is "load scheduling", and its activity classification index may be a code or label to indicate whether the activity is actively triggered or passively triggered, as well as its completion status and other information.

[0115] Then, the H status update activities are divided into X groups (each group contains two status update activities), and these groups are constructed into X status update activity combinations. For example, the first status update activity combination may include "load dispatch" (dominant status activity) and "grid monitoring" (reactive status activity). These two activities are combined to form a status update activity combination.

[0116] Next, the activity classification indexes of the dominant state activities and reactive state activities covered by each state update activity combination are determined from the H activity classification indexes to generate X classification index combinations. For example, for the first state update activity combination mentioned above, the activity classification indexes corresponding to "load dispatch" and "grid monitoring" are selected to form a classification index combination.

[0117] In this process, the conversation relationship between the dominant state activity and the reactive state activity represented by each activity conversation category is one of the preset M conversation relationships, such as "normal operation", "abnormal repair" and "emergency processing". Of course, there may also be a situation where there is no conversation relationship, such as two activities are completely independent and have no interaction.

[0118] Further, Figure 2 FIG. 1 is a schematic diagram of the hardware structure of a state monitoring system 100 for implementing the method provided in an embodiment of the present application. Figure 2 As shown, the status monitoring system 100 includes a memory 111 , a storage controller 112 , a processor 113 , a peripheral interface 114 , an input and output unit 115 , an audio unit 116 , a display unit 117 and a radio frequency unit 118 .

[0119] The memory 111, the storage controller 112, the processor 113, the peripheral interface 114, the input / output unit 115, the audio unit 116, the display unit 117, and the radio frequency unit 118 are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines.

[0120] The memory 111 may be, but not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electric erasable programmable read-only memory (EEPROM), etc. The memory 111 is used to store programs, and the processor 113 executes the programs after receiving the execution instruction. The processor 113 and other possible components may access the memory 111 under the control of the storage controller 112.

[0121] The processor 113 may be an integrated circuit chip with signal processing capabilities. The above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present invention may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0122] The peripheral interface 114 couples various input / output devices to the processor 113 and the memory 111. In some embodiments, the peripheral interface 114, the processor 113 and the memory controller 112 can be implemented in a single chip. In other embodiments, they can be implemented by separate chips.

[0123] The input-output unit 115 is used to provide input data to the user to implement the interaction between the user and the state monitoring system 100. The input-output unit 115 can be, but is not limited to, a mouse and a keyboard.

[0124] The audio unit 116 provides an audio interface to the user and may include one or more microphones, one or more speakers, and an audio circuit.

[0125] The display unit 117 provides an interactive interface (e.g., a user operation interface) between the state monitoring system 100 and the user or is used to display image data. In this embodiment, the display unit 117 may be a liquid crystal display or a touch display. If it is a touch display, it may be a capacitive touch screen or a resistive touch screen that supports single-point and multi-point touch operations. Supporting single-point and multi-point touch operations means that the touch display can sense touch operations generated from one or more positions on the touch display, and the sensed touch operations are handed over to the processor for calculation and processing.

[0126] The radio frequency unit 118 is used to receive and send radio wave signals (such as electromagnetic waves) to achieve mutual conversion between radio waves and electrical signals, thereby achieving communication between the status monitoring system 100 and the network 300 or other communication devices.

[0127] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of the embodiments of the present application are described. Other embodiments are within the scope of the attached claims. In some cases, the actions or steps recorded in the claims can be performed according to an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0128] Each embodiment in the embodiments of the present application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the above different embodiments, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0129] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program. The above program may be stored in a computer-readable storage medium, and the above storage medium may be a read-only memory, a disk or an optical disk, etc.

Claims

1. A switch cabinet state monitoring method based on artificial intelligence, characterized in that: The method comprises: Acquire X state update activity combinations and X classification index combinations of a target switch cabinet control system, where X is a positive integer greater than 1, each of the X state update activity combinations includes a dominant state activity and a reactive state activity of a pending activity session category, the X classification index combinations include activity classification indexes of each dominant state activity and each reactive state activity in the X state update activity combinations, each state update activity combination includes a dominant state activity with an active trigger state and a reactive state activity with a passive trigger state, the dominant state activity includes a dominant element in a target switch cabinet control task, the dominant element includes a preset operation rule or an instruction sequence, the reactive state activity includes a reactive element in the target switch cabinet control task, the reactive element includes a feedback mechanism or an adaptive strategy; Determine X activity session categories of the X state update activity combinations based on X dominant state description features, X first index category description features, X reactive state description features, and X second index category description features, wherein the X dominant state description features are description vectors of the X dominant state activities in the X state update activity combinations, the X first index category description features are index category features of the X dominant state activities, the X reactive state description features are description vectors of the X reactive state activities in the X state update activity combinations, the X first index category description features are index category features of the X dominant state activities, and the X second index category description features are index category features of the X reactive state activities; When an abnormality occurs in a control unit instance in a target node based on any one of the X active session categories, the target instance abnormality information is reported to the switch cabinet system operation and maintenance center. The abnormality occurs in the control unit instance in the target node includes an abnormality in the time domain and / or operation domain of the control unit instance response, and the target instance abnormality information reflects that an abnormality occurs in the control unit instance in the target node.

2. The switch cabinet state monitoring method based on artificial intelligence according to claim 1 is characterized in that: The determining of X activity session categories of the X state update activity combinations based on the X dominant state description features, the X first index category description features, the X reactive state description features, and the X second index category description features includes: Obtain description vectors of X dominant state activities in the X state update activity combinations, generate X dominant state description features, obtain description vectors of X reactive state activities in the X state update activity combinations, generate X reactive state description features, obtain index category features of activity classification indexes of the X dominant state activities in the X classification index combinations, generate X first index category description features, and obtain index category features of activity classification indexes of the X reactive state activities in the X classification index combinations, generate X second index category description features; The X dominant state description features, the X first index category description features, the X reactive state description features, and the X second index category description features are merged to generate X target feature representations corresponding to the X state update activity combinations; Based on the X target feature representations, X activity session categories of the X state update activity combinations are determined, each of the X activity session categories reflecting a session relationship between the dominant state activity and the reactive state activity in a corresponding one of the X state update activity combinations.

3. The switch cabinet state monitoring method based on artificial intelligence according to claim 2 is characterized in that: The obtaining description vectors of X dominant state activities in the X state update activity combinations to generate X dominant state description features includes: The description vector of the rth dominant state activity among the X dominant state activities is obtained based on the following operation, and the rth dominant state description feature is generated, where r is a positive integer not less than 1 and not greater than X: When the r-th dominant state activity includes Yr state behaviors, obtaining a behavior feature representation of each of the Yr state behaviors to generate Yr behavior feature representations, where Yr is a positive integer; The Yr behavior feature representations are subjected to equalization processing to generate the r-th dominant state description feature; or, the Yr behavior feature representations are subjected to weight fusion processing to generate the r-th dominant state description feature.

4. The switch cabinet state monitoring method based on artificial intelligence according to claim 2 is characterized in that: The obtaining description vectors of X reactive state activities in the X state update activity combinations and generating X reactive state description features includes: The description vector of the rth reactive state activity among the X reactive state activities is obtained based on the following operation, and the rth reactive state description feature is generated, where r is a positive integer not less than 1 and not greater than X: When the rth reactive state activity includes Zr state behaviors, obtaining a behavior feature representation of each state behavior in the Zr state behaviors, generating Zr behavior feature representations, where Zr is a positive integer; The Zr behavior feature representations are subjected to equalization processing to generate the rth reactive state description feature; or, weight fusion is performed on the Zr behavior feature representations to generate the rth reactive state description feature.

5. The switch cabinet state monitoring method based on artificial intelligence according to claim 2 is characterized in that: The step of obtaining index category features of the activity classification indexes of the X dominant state activities in the X classification index combinations to generate X first index category description features includes: The index category feature of the activity classification index of the rth dominant state activity among the X dominant state activities is obtained based on the following operation, and the rth first index category description feature is generated, where r is a positive integer not less than 1 and not greater than X: When the activity classification index of the rth dominant state activity includes the rth index description field, the field feature representation of the rth index description field is obtained, and the rth first index category description feature is the field feature representation of the rth index description field.

6. The switch cabinet state monitoring method based on artificial intelligence according to claim 2 is characterized in that: The step of obtaining index category features of the activity classification indexes of the X reactive state activities in the X classification index combinations to generate X second index category description features includes: The index category feature of the activity classification index of the rth reactive state activity among the X reactive state activities is obtained based on the following operation, and the rth second index category description feature is generated, where r is a positive integer not less than 1 and not greater than X: When the activity classification index of the rth reactive state activity includes the rth index description field, the field feature representation of the rth index description field is obtained, and the rth second index category description feature is the field feature representation of the rth index description field.

7. The switch cabinet state monitoring method based on artificial intelligence according to claim 2 is characterized in that: The fusing the X dominant state description features, the X first index category description features, the X reactive state description features, and the X second index category description features to generate X target feature representations corresponding to the X state update activity combinations includes: The rth target feature representation corresponding to the rth state update activity combination among the X state update activity combinations is obtained based on the following operation, wherein the rth state update activity combination includes the rth dominant state activity and the rth reactive state activity, and r is a positive integer not less than 1 and not greater than X: The task trigger description vector, the rth dominant state description feature, the rth first index category description feature, the rth reactive state description feature, and the rth second index category description feature are merged to generate the rth target feature representation, wherein the task trigger description vector reflects the trigger node of the target switch cabinet control task, the state update activity in the X state update activity combinations is the state update activity parsed from the target switch cabinet control task, the rth dominant state description feature is the description vector of the rth dominant state activity, the rth first index category description feature is the index category feature of the activity classification index of the rth dominant state activity, the rth reactive state description feature is the description vector of the rth reactive state activity, and the rth second index category description feature is the index category feature of the activity classification index of the rth reactive state activity; Among them, when the dimension of the task trigger description vector, the dimension of the r-th dominant state description feature, the dimension of the r-th first index category description feature, the dimension of the r-th reactive state description feature and the dimension of the r-th second index category description feature are all 1*K, the task trigger description vector, the r-th dominant state description feature, the r-th first index category description feature, the r-th reactive state description feature and the r-th second index category description feature are fused to generate the r-th target feature representation with a dimension of 5*K, where K is a positive integer greater than 1.

8. The switch cabinet state monitoring method based on artificial intelligence according to claim 2 is characterized in that: The determining, based on the X target feature representations, X activity session categories of the X status update activity combinations comprises: Determine an rth activity session category of an rth status update activity combination among the X status update activity combinations based on an rth target feature representation among the X target feature representations based on the following operations, where r is a positive integer not less than 1 and not greater than X: Loading the r-th target feature representation into the AI ​​neural network, generating M session category classification data corresponding to the M active session categories, the M session category classification data reflecting the possibility that the r-th active session category is each candidate session category among the M set active session categories, where M is a positive integer greater than 1; The rth activity session category is determined to be equal to a target activity session category among the M set activity session categories, and among the M session category classification data, the rth activity session category has the highest probability of being the target activity session category.

9. The switch cabinet state monitoring method based on artificial intelligence according to claim 2 is characterized in that: The determining, based on the X target feature representations, X activity session categories of the X status update activity combinations comprises: The X target feature representations are loaded into the AI ​​neural network in rounds to generate a conversation category classification data sequence corresponding to the target feature representations of each round, wherein the target feature representations of each round include J target feature representations among the X target feature representations, and the conversation category classification data sequence corresponding to the target feature representations of each round includes J groups of conversation category classification data, and each group of conversation category classification data among the J groups of conversation category classification data includes M conversation category classification data, and the M conversation category classification data covered by the s-th group of conversation category classification data reflect the possibility that the s-th active conversation category corresponding to the s-th target feature representation is each candidate conversation category among the M set active conversation categories, and the s-th group of conversation category classification data corresponds to the s-th target feature representation among the J target feature representations, where M is a positive integer greater than 1, J is a positive integer not less than 2 and less than X, and s is a positive integer not less than 1 and not greater than J; Determine the sth active session category among J active session categories based on the sth group of session category classification data in the sequence of session category classification data by using the following operations: The sth active session category is determined to be equal to the target active session category among the M set active session categories, and among the M session category classification data covered by the sth group of session category classification data, the sth active session category is most likely to be the target active session category.

10. A status monitoring system, characterized in that: The state monitoring system includes a processor and a readable storage medium, wherein the readable storage medium stores a program, and when the program is executed by the processor, the switch cabinet state monitoring method based on artificial intelligence according to any one of claims 1 to 9 is implemented.

Images