A computer room power equipment monitoring system based on edge computing and fault prediction

Abstract

The application discloses a kind of based on edge computing and failure prediction's computer room power equipment monitoring system, including the following steps: data acquisition module, for collecting data;Edge processing module, for obtaining standardized power equipment operation data;State extraction module, for extracting continuous operation parameters and discrete operating state;Feature construction module, for constructing continuous state feature data and discrete state jump data;Hybrid trajectory construction module, for constructing equipment hybrid state data to obtain equipment hybrid state trajectory;Association construction module, for constructing equipment association;Failure prediction module, for obtaining equipment state evolution result and linkage failure risk result according to edge hybrid failure prediction algorithm;Result generation module, for generating failure prediction result;Alarm output module, for sending and generating local alarm record.The application utilizes edge hybrid prediction, realizes computer room power equipment early warning, with the advantages of timely, accurate, stable.

Description

Technical Field

[0001] This invention relates to the field of data center power equipment operation monitoring technology, and in particular to a data center power equipment monitoring system based on edge computing and fault prediction. Background Technology

[0002] Monitoring of power equipment in a computer room typically involves connecting to UPS power supplies, battery banks, distribution cabinets, PDUs, precision air conditioners, generators, and temperature and humidity sensors through a monitoring platform. This platform collects data on voltage, current, power, load rate, temperature, humidity, battery capacity, alarm records, and maintenance records. The platform then displays the operating status and issues abnormal alarms through protocol conversion, data uploading, and threshold comparison.

[0003] The above methods mainly rely on whether a single operating parameter exceeds the threshold or whether a single alarm event occurs to make judgments. It is difficult to simultaneously utilize the changing trends of continuous operating parameters and the transition process of discrete operating states. It is also difficult to combine the power supply connection path, load access path and the installation location of environmental acquisition equipment to analyze the linkage effect between devices. As a result, the data center site side cannot identify the evolution trend of power equipment failure in a timely manner, and the accuracy of failure prediction results and the ability to judge linkage risks are insufficient.

[0004] Therefore, how to provide a monitoring system for data center power equipment based on edge computing and fault prediction is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a monitoring system for power equipment in a data center based on edge computing and fault prediction. This invention utilizes edge promiscuous prediction to achieve early warning of power equipment in the data center, and has the advantages of being timely, accurate, and stable.

[0006] A data center power equipment monitoring system based on edge computing and fault prediction according to an embodiment of the present invention includes the following steps: The data acquisition module is used to collect the operating data and equipment configuration data of the power equipment of the monitoring objects in the computer room; The edge processing module is used to perform edge standardization processing on the power equipment operation data to obtain standardized power equipment operation data. The status extraction module is used to extract continuous operating parameters and discrete operating states based on standardized power equipment operating data. The feature construction module is used to construct continuous state feature data based on continuous operating parameters and discrete state transition data based on discrete operating states. The hybrid trajectory construction module is used to construct hybrid state data of the device and arrange it in time sequence to obtain the hybrid state trajectory of the device. The association construction module is used to construct equipment associations based on the equipment identifiers and equipment configuration data carried in the standardized power equipment operation data. The fault prediction module is used to input the mixed state trajectory of the equipment into the edge mixed fault prediction algorithm, jointly identify the continuous state feature data and discrete state transition data to obtain the equipment state evolution result, and perform linkage risk analysis based on the equipment correlation relationship to obtain the linkage fault risk result. The result generation module is used to generate fault prediction results based on the equipment status evolution results and the linkage fault risk results; The alarm output module is used to send fault prediction results and generate local alarm records.

[0007] Optionally, the data acquisition module includes: The power monitoring objects in the computer room include power supply equipment, energy storage equipment, power distribution equipment, cooling equipment, backup power generation equipment, and environmental acquisition equipment. The power supply equipment includes UPS power supplies and PDUs, the energy storage equipment includes battery packs, the power distribution equipment includes distribution cabinets, the cooling equipment includes precision air conditioners, the backup power generation equipment includes generators, and the environmental acquisition equipment includes temperature and humidity sensors. The power equipment operation data includes continuously collected data, status switching data, alarm record data, and maintenance record data. The equipment configuration data includes equipment access location, power supply connection path, load access path, and installation location of the environmental acquisition equipment.

[0008] Optionally, the edge processing module includes: The edge standardization process includes protocol conversion, timestamp unification, device identifier binding, missing data completion, abnormal data removal, and data format conversion.

[0009] Optionally, the state extraction module includes: Voltage, current, power, load rate, temperature, humidity, and battery capacity values ​​are extracted from continuously collected data in standardized power equipment operation data. Based on the voltage, current, power, load rate, temperature, humidity, and battery capacity data collected, voltage change sequences, current change sequences, power change sequences, load rate change sequences, temperature change sequences, humidity change sequences, and battery capacity change sequences are generated and aggregated into continuous operating parameters. Extract power supply status records, energy storage status records, cooling status records, power distribution status records, and maintenance recovery status records from the status switching data, alarm record data, and maintenance record data in the standardized power equipment operation data; Power supply status is generated based on power supply status records, energy storage status is generated based on energy storage status records, cooling status is generated based on cooling status records, power distribution status is generated based on power distribution status records, and maintenance and recovery status is generated based on maintenance and recovery status records. Power supply status, energy storage status, cooling status, power distribution status, and maintenance and recovery status are then aggregated into discrete operating statuses.

[0010] Optionally, the feature construction module includes: The voltage change sequence, current change sequence, power change sequence, load rate change sequence, temperature change sequence, humidity change sequence, and battery capacity change sequence in the continuous operating parameters are compared with adjacent sequence values ​​to obtain the direction and magnitude of change. The continuous acquisition interval that maintains the same direction of change is determined based on the direction of change. The duration is obtained based on the start and end times of the continuous acquisition interval. The change frequency is obtained based on the number of times the direction of change is switched between adjacent continuous acquisition intervals. The direction, magnitude, duration, and frequency of change are bound to the voltage change sequence, current change sequence, power change sequence, load rate change sequence, temperature change sequence, humidity change sequence, and battery capacity change sequence to obtain continuous state characteristic data. The jump sequence is obtained according to the order in which the power supply state, energy storage state, cooling state, power distribution state, and maintenance and recovery state occur in the discrete operation state, and the jump number is obtained according to the number of times the power supply state, energy storage state, cooling state, power distribution state, and maintenance and recovery state are switched. The duration of a state is obtained from the start and end times of the same state in discrete running states, and the interval between adjacent states is obtained from the end and start times of two adjacent discrete running states. The jump order, jump count, state duration, and interval between adjacent states are bound to the corresponding power supply state, energy storage state, cooling state, power distribution state, and maintenance and recovery state to obtain discrete state jump data.

[0011] Optionally, the hybrid trajectory construction module includes: The continuous state characteristic data and discrete state transition data are aggregated according to the device identifier to obtain the continuous state characteristic data and discrete state transition data corresponding to the same device identifier. Based on the timestamp, the continuous state feature data and discrete state transition data corresponding to the same device identifier are time-aligned to obtain the continuous state feature data and discrete state transition data corresponding to the same device identifier and the same timestamp; By binding continuous state feature data, discrete state transition data, device identifier and timestamp corresponding to the same device identifier and the same timestamp, mixed state data of the device is obtained. Arrange the mixed state data of devices corresponding to the same device identifier in a time sequence according to the timestamp to obtain the mixed state data sequence of devices. According to the arrangement order of the mixed state data in the mixed state data sequence, the adjacent mixed state connection relationship is established between two adjacent mixed state data, and the mixed state data sequence and the adjacent mixed state connection relationship are combined to obtain the mixed state trajectory of the equipment.

[0012] Optionally, the association construction module includes: Extract equipment identifiers from standardized power equipment operation data, and extract equipment access locations, power supply connection paths, load access paths, and installation locations of environmental acquisition equipment from equipment configuration data; Based on the device identifier and the device access location, the device identifier is matched with the device access location to obtain the device location correspondence; Based on the device location correspondence and power supply connection path, the device identifiers with power supply connections in the power supply connection path are matched to obtain the power supply association relationship; Based on the device location correspondence and load access path, the device identifiers with load access in the load access path are matched to obtain the load association relationship; Based on the device location correspondence and the installation location of the environmental acquisition device, the device identifier of the environmental acquisition device is matched with the device identifier of the device access location corresponding to the installation location of the environmental acquisition device to obtain the environmental association relationship; The power supply association, load association, and environmental association are bound according to the device identifier to obtain the device association.

[0013] Optionally, the fault prediction module includes: Extract the hybrid state trajectory points corresponding to the same device identifier according to the time sequence in the hybrid state trajectory of the device to obtain the hybrid state point sequence; The continuous state feature data contained in adjacent mixed state trajectory points are read sequentially from the mixed state point sequence. The change direction, change amplitude, duration and change frequency in the continuous state feature data are concatenated in time order to obtain the continuous state change process. The discrete state transition data contained in adjacent mixed state trajectory points are read sequentially from the mixed state point sequence. The transition order, number of transitions, state duration and adjacent state interval time in the discrete state transition data are concatenated in time order to obtain the discrete state transition process. A continuous direction sequence is generated based on the direction of change during the continuous state change process; a continuous amplitude sequence is generated based on the amplitude of change during the continuous state change process; a continuous duration sequence is generated based on the duration of change during the continuous state change process; and a continuous frequency sequence is generated based on the frequency of change during the continuous state change process. The continuous direction sequence, continuous amplitude sequence, continuous duration sequence, and continuous frequency sequence are then aggregated into continuous evolution recognition data. A discrete jump sequence is generated based on the jump order in the discrete state jump process, a discrete number sequence is generated based on the number of jumps in the discrete state jump process, a discrete duration sequence is generated based on the state duration in the discrete state jump process, and a discrete interval sequence is generated based on the time interval between adjacent states in the discrete state jump process. The discrete jump sequence, discrete number sequence, discrete duration sequence and discrete interval sequence are then collected into discrete evolution recognition data. The continuous evolution identification data is input into the edge hybrid fault prediction algorithm, and the continuous direction sequence, continuous amplitude sequence, continuous duration sequence and continuous frequency sequence are combined for identification to obtain the continuous evolution label corresponding to each hybrid state trajectory point; The discrete evolution identification data is input into the edge hybrid fault prediction algorithm to combine and identify discrete jump sequences, discrete number sequences, discrete duration sequences and discrete interval sequences to obtain the discrete evolution labels corresponding to each hybrid state trajectory point. Following the order of high-risk operating state, continuous degradation state, slight degradation state, fluctuating operating state, and normal operating state, the continuous evolution markers and discrete evolution markers corresponding to the same mixed state trajectory points are matched to obtain the normal operating state, fluctuating operating state, slight degradation state, continuous degradation state, and high-risk operating state, which are then aggregated into the equipment state evolution results. Based on the device association relationships, the device status evolution results corresponding to the device identifiers with power supply association relationships, load association relationships, and environmental association relationships are matched to obtain the associated status evolution data; Based on the device identifier, time sequence, and device association relationship corresponding to the device state evolution results in the associated state evolution data, the risk result of linkage failure is determined.

[0014] Optionally, the result generation module includes: Based on the equipment status evolution results, the fault risk level is determined as follows: normal operation status, fluctuating operation status, slight degradation status, continuous degradation status, and high-risk operation status. The fault type is determined based on the state category corresponding to the equipment state evolution result and the equipment association relationship corresponding to the linkage fault risk result; The scope of impact is determined based on the equipment relationships and the corresponding computer room power monitoring objects based on the fault type; Generate maintenance reminders based on the fault risk level, fault type, and scope of impact; The fault risk level, fault type, scope of impact, and maintenance reminders are compiled into fault prediction results.

[0015] Optionally, the alarm output module includes: The fault prediction results are read from the edge computing nodes and converted into fault prediction data for transmission. The fault prediction data is sent to the data center power equipment monitoring platform and on-site alarm terminal that are connected to the edge computing node through the edge computing node; Local alarm records are generated based on the fault prediction results.

[0016] The beneficial effects of this invention are: This invention deploys edge computing nodes on-site in the data center to perform edge standardization processing on the operating data of the power equipment being monitored. This enables continuously collected data, status switching data, alarm record data, and maintenance record data generated by UPS power supplies, PDUs, battery banks, distribution cabinets, precision air conditioners, generators, and temperature and humidity sensors to form standardized power equipment operating data with equipment identifiers and unified timestamps. This reduces data inconsistencies caused by multiple devices, protocols, and acquisition frequencies, providing a unified data foundation for subsequent status extraction, trajectory construction, and fault prediction. It also reduces reliance on real-time computing from remote platforms and improves the on-site response capability of data center power equipment monitoring.

[0017] This invention combines continuous operating parameters and discrete operating states for processing. It constructs continuous state feature data through voltage change sequences, current change sequences, power change sequences, load rate change sequences, temperature change sequences, humidity change sequences, and battery capacity change sequences. It also constructs discrete state transition data through power supply state, energy storage state, cooling state, power distribution state, and maintenance recovery state. Finally, it combines equipment identifiers and timestamps to form a mixed state trajectory of the equipment. Compared with methods that rely solely on a single parameter threshold or a single alarm event, this invention can simultaneously reflect the numerical change trend of power equipment and the process of operating state transitions. It is more suitable for identifying the pre-fault evolution process of gradually transitioning from a normal operating state to a fluctuating operating state, a slightly degraded state, a continuously degraded state, and a high-risk operating state.

[0018] This invention also constructs device correlations based on equipment configuration data, incorporating power supply correlations, load correlations, and environmental correlations into the fault prediction process, thus expanding fault prediction beyond a single device. Through combined analysis of device state evolution results and device correlations, it can identify the linked fault risks between power supply equipment, energy storage equipment, power distribution equipment, cooling equipment, and environmental data acquisition equipment, thereby generating fault prediction results that include fault risk level, fault type, impact range, and maintenance reminders. This improves the matching degree of fault prediction results to the actual data center operation chain, assisting maintenance personnel in proactively locating risky devices and affected areas, and reducing the impact of sudden power supply anomalies, battery degradation, power distribution anomalies, and cooling environment anomalies on the stable operation of the data center. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a data center power equipment monitoring system based on edge computing and fault prediction proposed in this invention. Figure 2 This is a schematic diagram illustrating the construction of equipment association relationships in a data center power equipment monitoring system based on edge computing and fault prediction, as proposed in this invention. Figure 3 This is a schematic diagram of the processing flow of an edge hybrid fault prediction algorithm in a data center power equipment monitoring system based on edge computing and fault prediction proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figures 1-3 A monitoring system for power equipment in a data center based on edge computing and fault prediction includes the following steps: The data acquisition module is used to collect the operating data and equipment configuration data of the power equipment of the monitoring objects in the computer room; The edge processing module is used to perform edge standardization processing on the power equipment operation data through edge computing nodes deployed on the data center side, so as to obtain standardized power equipment operation data with equipment identification and unified timestamp; The status extraction module is used to extract continuous operating parameters from the continuously acquired data in the standardized power equipment operation data, and to extract discrete operating states from the status switching data, alarm record data, and maintenance record data in the standardized power equipment operation data. The continuous operating parameters include voltage change sequence, current change sequence, power change sequence, load rate change sequence, temperature change sequence, humidity change sequence, and battery capacity change sequence. The discrete operating states include power supply state, energy storage state, cooling state, power distribution state, and maintenance recovery state. The power supply state includes UPS normal power supply state and UPS bypass power supply state. The energy storage state includes battery discharge state. The cooling state includes air conditioning cooling state and air conditioning shutdown state. The power distribution state includes power distribution branch alarm state. The feature construction module is used to construct continuous state feature data based on continuous operating parameters and discrete state transition data based on discrete operating states. The continuous state feature data includes the direction of change, the magnitude of change, the duration, and the frequency of change. The discrete state transition data includes the transition order, the number of transitions, the state duration, and the interval between adjacent states. The hybrid trajectory construction module is used to construct hybrid state data of equipment based on continuous state feature data, discrete state transition data, equipment identifier and timestamp, and to arrange the hybrid state data of equipment corresponding to the same equipment identifier in time sequence according to the timestamp to obtain the hybrid state trajectory of equipment. The association construction module is used to construct equipment associations based on the equipment identifiers and equipment configuration data carried in the standardized power equipment operation data. The equipment associations include power supply associations, load associations, and environmental associations. The fault prediction module is used to input the hybrid state trajectory of the equipment into the edge hybrid fault prediction algorithm, jointly identify the continuous state feature data and discrete state transition data in the hybrid state trajectory of the equipment, obtain the equipment state evolution result, and perform linkage risk analysis based on the equipment state evolution result and equipment correlation relationship to obtain the linkage fault risk result. The equipment state evolution result includes normal operation state, fluctuating operation state, slight degradation state, continuous degradation state and high-risk operation state; The result generation module is used to generate fault prediction results based on the equipment status evolution results and linkage fault risk results. The fault prediction results include fault risk level, fault type, scope of impact and maintenance reminder content. The alarm output module is used to send the fault prediction results to the data center power equipment monitoring platform and the on-site alarm terminal that are connected to the edge computing node through the edge computing node, and generate local alarm records based on the fault prediction results.

[0022] In this embodiment, the data acquisition module includes: The power monitoring objects in the computer room include power supply equipment, energy storage equipment, power distribution equipment, cooling equipment, backup power generation equipment, and environmental acquisition equipment. Power supply equipment includes UPS power supplies and PDUs, energy storage equipment includes battery packs, power distribution equipment includes distribution cabinets, cooling equipment includes precision air conditioners, backup power generation equipment includes generators, and environmental acquisition equipment includes temperature and humidity sensors. The power equipment operation data includes continuously acquired data, status switching data, alarm record data, and maintenance record data. The equipment configuration data includes equipment access location, power supply connection path, load access path, and installation location of environmental acquisition equipment.

[0023] In this embodiment, the edge processing module includes: Edge standardization processing includes protocol conversion, timestamp unification, device identifier binding, missing data completion, abnormal data removal, and data format conversion.

[0024] In this embodiment, the state extraction module includes: Voltage, current, power, load rate, temperature, humidity, and battery capacity values ​​are extracted from continuously collected data in standardized power equipment operation data. Based on the voltage, current, power, load rate, temperature, humidity, and battery capacity data collected, voltage change sequences, current change sequences, power change sequences, load rate change sequences, temperature change sequences, humidity change sequences, and battery capacity change sequences are generated and aggregated into continuous operating parameters. Extract power supply status records, energy storage status records, cooling status records, power distribution status records, and maintenance recovery status records from the status switching data, alarm record data, and maintenance record data in the standardized power equipment operation data; Power supply status is generated based on power supply status records, energy storage status is generated based on energy storage status records, cooling status is generated based on cooling status records, power distribution status is generated based on power distribution status records, and maintenance and recovery status is generated based on maintenance and recovery status records. Power supply status, energy storage status, cooling status, power distribution status, and maintenance and recovery status are then aggregated into discrete operating statuses.

[0025] In this embodiment, the feature construction module includes: The voltage change sequence, current change sequence, power change sequence, load rate change sequence, temperature change sequence, humidity change sequence, and battery capacity change sequence in the continuous operating parameters are compared with adjacent sequence values ​​to obtain the direction and magnitude of change. The continuous acquisition interval that maintains the same direction of change is determined based on the direction of change. The duration is obtained based on the start and end times of the continuous acquisition interval. The change frequency is obtained based on the number of times the direction of change is switched between adjacent continuous acquisition intervals. The direction, magnitude, duration, and frequency of change are bound to the voltage change sequence, current change sequence, power change sequence, load rate change sequence, temperature change sequence, humidity change sequence, and battery capacity change sequence to obtain continuous state characteristic data. The jump sequence is obtained according to the order in which the power supply state, energy storage state, cooling state, power distribution state, and maintenance and recovery state occur in the discrete operation state, and the jump number is obtained according to the number of times the power supply state, energy storage state, cooling state, power distribution state, and maintenance and recovery state are switched. The duration of a state is obtained from the start and end times of the same state in discrete running states, and the interval between adjacent states is obtained from the end and start times of two adjacent discrete running states. The jump order, jump count, state duration, and interval between adjacent states are bound to the corresponding power supply state, energy storage state, cooling state, power distribution state, and maintenance and recovery state to obtain discrete state jump data.

[0026] In this embodiment, the hybrid trajectory construction module includes: The continuous state characteristic data and discrete state transition data are aggregated according to the device identifier to obtain the continuous state characteristic data and discrete state transition data corresponding to the same device identifier. Based on the timestamp, the continuous state feature data and discrete state transition data corresponding to the same device identifier are time-aligned to obtain the continuous state feature data and discrete state transition data corresponding to the same device identifier and the same timestamp; By binding continuous state feature data, discrete state transition data, device identifier and timestamp corresponding to the same device identifier and the same timestamp, mixed state data of the device is obtained. Arrange the mixed state data of devices corresponding to the same device identifier in a time sequence according to the timestamp to obtain the mixed state data sequence of devices. According to the arrangement order of the mixed state data in the mixed state data sequence, establish the adjacent mixed state connection relationship between two adjacent mixed state data, and combine the mixed state data sequence and the adjacent mixed state connection relationship to obtain the mixed state trajectory of the equipment. The connection relationship between adjacent hybrid states includes the timestamp of the hybrid state data of the preceding device, the timestamp of the hybrid state data of the following device, the connection direction, the connection interval, the difference in continuous state characteristic data, and the difference in discrete state transition data. Among them, the connection direction is determined by the order of two adjacent timestamps corresponding to the same device identifier, and the connection interval is determined by the time difference between the following timestamp and the preceding timestamp. The difference in continuous state characteristic data is obtained by comparing the change direction, change amplitude, duration, and change frequency in the hybrid state data of two adjacent devices item by item. The difference in discrete state transition data is obtained by comparing the transition order, number of transitions, state duration, and adjacent state interval time in the hybrid state data of two adjacent devices item by item. The connection direction, connection interval, difference in continuous state characteristic data, and difference in discrete state transition data are written into the connection relationship between adjacent hybrid states.

[0027] In this embodiment, the association construction module includes: Extract equipment identifiers from standardized power equipment operation data, and extract equipment access locations, power supply connection paths, load access paths, and installation locations of environmental acquisition equipment from equipment configuration data; Based on the device identifier and the device access location, the device identifier is matched with the device access location to obtain the device location correspondence; Based on the device location correspondence and power supply connection path, the device identifiers with power supply connections in the power supply connection path are matched to obtain the power supply association relationship; Based on the device location correspondence and load access path, the device identifiers with load access in the load access path are matched to obtain the load association relationship; Based on the device location correspondence and the installation location of the environmental data acquisition device, the device identifier of the environmental data acquisition device is matched with the device identifier of the device access location corresponding to the installation location of the environmental data acquisition device to obtain the environmental association relationship; The installation location and access location of the environmental data acquisition equipment include the location of the computer room area, the location of the cabinet, and the location of the passage. When the installation location of the environmental data acquisition equipment has the same computer room area location, the same cabinet location, or the same passage location as the access location corresponding to any equipment identifier, the equipment identifier of the environmental data acquisition equipment is matched with the equipment identifier. The matching result is used as the environmental impact object pair in the environmental association relationship. The environmental impact object pair includes the equipment identifier of the environmental data acquisition equipment, the corresponding equipment identifier, and the installation location and access location of the two. The environmental association relationship is formed by all environmental impact object pairs. The power supply association, load association, and environmental association are bound according to the device identifier to obtain the device association.

[0028] In this embodiment, the fault prediction module includes: Extract the hybrid state trajectory points corresponding to the same device identifier according to the time sequence in the hybrid state trajectory of the device to obtain the hybrid state point sequence; The continuous state feature data contained in adjacent mixed state trajectory points are read sequentially from the mixed state point sequence. The change direction, change amplitude, duration and change frequency in the continuous state feature data are concatenated in time order to obtain the continuous state change process. The discrete state transition data contained in adjacent mixed state trajectory points are read sequentially from the mixed state point sequence. The transition order, number of transitions, state duration and adjacent state interval time in the discrete state transition data are concatenated in time order to obtain the discrete state transition process. A continuous direction sequence is generated based on the direction of change during the continuous state change process; a continuous amplitude sequence is generated based on the amplitude of change during the continuous state change process; a continuous duration sequence is generated based on the duration of change during the continuous state change process; and a continuous frequency sequence is generated based on the frequency of change during the continuous state change process. The continuous direction sequence, continuous amplitude sequence, continuous duration sequence, and continuous frequency sequence are then aggregated into continuous evolution recognition data. A discrete jump sequence is generated based on the jump order in the discrete state jump process, a discrete number sequence is generated based on the number of jumps in the discrete state jump process, a discrete duration sequence is generated based on the state duration in the discrete state jump process, and a discrete interval sequence is generated based on the time interval between adjacent states in the discrete state jump process. The discrete jump sequence, discrete number sequence, discrete duration sequence and discrete interval sequence are then collected into discrete evolution recognition data. The continuous evolution identification data is input into the edge hybrid fault prediction algorithm. The continuous direction sequence, continuous amplitude sequence, continuous persistence sequence and continuous frequency sequence are combined and identified to obtain the continuous evolution label corresponding to each hybrid state trajectory point. The continuous evolution label includes continuous stationary label, continuous fluctuation label, continuous degradation label, continuous persistent degradation label and continuous high risk label. The generation of continuous volatility markers and continuous high-risk markers specifically includes: When combining and identifying continuous direction sequences, continuous amplitude sequences, continuous persistence sequences, and continuous frequency sequences, the median of amplitude and the interquartile range of amplitude are first calculated from the continuous amplitude sequence to form the amplitude identification limit. The amplitude identification limit is the sum of the median of amplitude and twice the interquartile range of amplitude. The persistent median and persistent interquartile range are calculated from the continuous persistent series to form the persistent identification limit, which is the sum of the persistent median and twice the persistent interquartile range; When the continuous amplitude sequence corresponding to the same mixed state trajectory point reaches the amplitude recognition limit and the continuous persistence sequence reaches the persistence recognition limit, a continuous persistence degradation mark is generated. When the continuous amplitude sequence reaches the amplitude recognition limit but the continuous persistence sequence does not reach the persistence recognition limit, a continuous degradation mark is generated. When the continuous direction sequence changes direction and the continuous frequency sequence reaches the frequency median, a continuous fluctuation mark is generated. When the continuous amplitude sequence, continuous persistence sequence, and continuous frequency sequence have not reached the corresponding recognition limit, and the continuous direction sequence maintains the same direction or has no directional change, a continuous stationary marker is generated; when the continuous amplitude sequence, continuous persistence sequence, and continuous frequency sequence have all reached the corresponding recognition limit, a continuous high-risk marker is generated. The edge hybrid fault prediction algorithm integrates continuous state feature data and discrete state transition data into a hybrid state trajectory of the equipment. It identifies the pre-fault evolution process of the equipment through continuous evolution markers, discrete evolution markers and state judgment priority. Then, it outputs the linkage fault risk results by combining the equipment correlation relationship, thereby improving the advance, accuracy and linkage risk identification capability of the fault prediction of the power equipment in the data center. The discrete evolution identification data is input into the edge hybrid fault prediction algorithm to combine and identify discrete jump sequences, discrete number sequences, discrete persistence sequences and discrete interval sequences to obtain discrete evolution labels corresponding to each hybrid state trajectory point. Discrete evolution labels include discrete stationary labels, discrete fluctuation labels, discrete degradation labels, discrete persistent degradation labels and discrete high-risk labels. The frequency identification limit of a discrete frequency sequence is determined by the sum of the frequency median of the discrete frequency sequence and the interquartile range of twice the frequency; the persistence identification limit of a discrete persistence sequence is determined by the sum of the persistence median of the discrete persistence sequence and the interquartile range of twice the persistence. When the discrete number sequence reaches the number recognition limit and the discrete persistence sequence reaches the persistence recognition limit, a discrete persistence degradation flag is generated. When the discrete number sequence reaches the number recognition limit but the discrete persistence sequence does not reach the persistence recognition limit, a discrete degradation flag is generated. When the power supply state, energy storage state, cooling state, power distribution state or maintenance and recovery state in the discrete jump sequence switch back and forth, a discrete fluctuation flag is generated. When neither the discrete frequency sequence nor the discrete persistence sequence reaches the corresponding recognition limit, and the discrete jump sequence does not switch back and forth, a discrete stationary marker is generated. When the discrete frequency sequence reaches the frequency recognition limit, the discrete persistence sequence reaches the persistence recognition limit, and the discrete interval sequence continuously shortens, a discrete high-risk marker is generated. Following the sequence of high-risk operation state, continuous degradation state, slight degradation state, fluctuating operation state, and normal operation state, the continuous evolution markers and discrete evolution markers corresponding to the same mixed state trajectory points are mapped. When the continuous evolution marker is a continuous high-risk marker or the discrete evolution marker is a discrete high-risk marker, a high-risk operation state is determined. When it does not belong to a high-risk operation state and the continuous evolution marker is a continuous continuous degradation marker or the discrete evolution marker is a discrete continuous degradation marker, a continuous degradation state is determined. When it does not belong to either a high-risk operation state or a continuous degradation state and the continuous evolution marker is a discrete continuous degradation marker, a continuous degradation state is determined. When the continuous evolution marker is a continuous degradation marker or the discrete evolution marker is a discrete degradation marker, a slight degradation state is determined. When it does not belong to a high-risk operating state, a continuous degradation state, or a slight degradation state, and the continuous evolution marker is a continuous fluctuation marker or the discrete evolution marker is a discrete fluctuation marker, a fluctuating operating state is determined. When the continuous evolution marker is a continuous stable marker and the discrete evolution marker is a discrete stable marker, a normal operating state is determined. The normal operating state, fluctuating operating state, slight degradation state, continuous degradation state, and high-risk operating state are aggregated into the equipment state evolution results. When mapping continuous evolution markers and discrete evolution markers corresponding to the same mixed state trajectory point, unique state determination is performed in the order of high-risk operating state, continuous degradation state, slight degradation state, fluctuating operating state, and normal operating state. When the same mixed state trajectory point has been determined to be in a high-risk operating state, continuous degradation state, slight degradation state, fluctuating operating state, and normal operating state are no longer judged. When the same mixed state trajectory point has been determined to be in a continuous degradation state, slight degradation state, fluctuating operating state, and normal operating state are no longer judged. Based on the device association relationships, the device status evolution results corresponding to the device identifiers with power supply association relationships, load association relationships, and environmental association relationships are matched in chronological order to obtain the associated status evolution data; Based on the device identifier, time sequence, and device association relationship corresponding to the device state evolution results in the associated state evolution data, the risk result of linkage failure is determined.

[0029] In this embodiment, the result generation module includes: Based on the equipment status evolution results, the fault risk level is determined as follows: normal operation status, fluctuating operation status, slight degradation status, continuous degradation status, and high-risk operation status. The fault type is determined based on the state category corresponding to the equipment state evolution result and the equipment association relationship corresponding to the linkage fault risk result; The determination of fault types specifically includes: when determining fault types based on equipment status evolution results and linkage fault risk results, extracting equipment identifiers in slightly degraded, continuously degraded, and high-risk operating states, and determining the association between the data center power monitoring object and the equipment corresponding to the equipment identifier. When the data center power monitoring object is a power supply device, it is determined to be a power supply fault type; when the data center power monitoring object is an energy storage device, it is determined to be an energy storage fault type; when the data center power monitoring object is a power distribution device, it is determined to be a power distribution fault type; and when the data center power monitoring object is a cooling device or an environmental acquisition device, it is determined to be a cooling environment fault type. The scope of impact is determined based on the equipment relationships and the data center power monitoring objects corresponding to the fault types involved in the linkage fault risk results. The determination of the scope of impact specifically includes: when determining the scope of impact based on the results of the linkage failure risk and the failure type, extracting the equipment association and equipment identifier corresponding to the failure type from the results of the linkage failure risk; when the failure type is a power supply failure, energy storage failure, or power distribution failure, determining the data center power monitoring object corresponding to the power supply association and load association as the scope of impact; when the failure type is a cooling environment failure, determining the data center power monitoring object corresponding to the environment association as the scope of impact. Generate maintenance reminders based on the fault risk level, fault type, and scope of impact; The fault risk level, fault type, scope of impact, and maintenance reminders are compiled into fault prediction results.

[0030] In this embodiment, the alarm output module includes: The fault prediction results are read by the edge computing node and converted into fault prediction transmission data. The fault prediction transmission data includes the fault risk level, fault type, scope of impact and maintenance reminder content. The fault prediction data is sent to the data center power equipment monitoring platform and on-site alarm terminal that are connected to the edge computing node through the edge computing node; Local alarm records are generated based on the fault prediction results.

[0031] Example 1: To verify the feasibility of this invention in practice, it was applied to a power monitoring scenario in a data center of a large office park. This data center is equipped with UPS power supplies, PDUs, battery banks, distribution cabinets, precision air conditioners, backup generators, and temperature and humidity sensors. The original monitoring method mainly relied on a monitoring platform to collect current equipment values ​​and to determine thresholds based on voltage, current, temperature, humidity, battery capacity, and alarm status. While this method can detect anomalies, it struggles to promptly identify the process of equipment gradually transitioning from normal operation to fluctuating operation, slight degradation, continuous degradation, or high-risk operation. It also has difficulty determining whether there are any interconnected risks between the UPS power supply, battery banks, distribution cabinets, PDUs, precision air conditioners, and environmental monitoring equipment.

[0032] In this scenario, UPS power supplies, PDUs, battery banks, distribution cabinets, precision air conditioners, backup generators, and temperature and humidity sensors are used as power monitoring targets in the data center. Operating data and configuration data of the power equipment are collected. Operating data includes continuously collected data, status switching data, alarm records, and maintenance records. Configuration data includes equipment access locations, power supply connection paths, load access paths, and the installation locations of environmental monitoring devices. Edge computing nodes are deployed on-site in the data center to perform protocol conversion, timestamp unification, equipment identifier binding, missing data completion, abnormal value removal, and data format conversion on the operating data. This results in standardized operating data for the power equipment, carrying equipment identifiers and unified timestamps, enabling data from different devices, protocols, and data sources to enter a unified processing flow.

[0033] Edge computing nodes extract continuous operating parameters from continuously acquired data in standardized power equipment operation data, including voltage change sequences, current change sequences, power change sequences, load rate change sequences, temperature change sequences, humidity change sequences, and battery capacity change sequences. Simultaneously, they extract discrete operating states from state switching data, alarm record data, and maintenance record data, including power supply state, energy storage state, cooling state, power distribution state, and maintenance recovery state. Subsequently, continuous state feature data is constructed based on the continuous operating parameters, discrete state transition data is constructed based on the discrete operating states, and combined with equipment identifiers and timestamps to form mixed state data. Then, the mixed state data corresponding to the same equipment identifier are sequentially arranged according to timestamps to obtain the mixed state trajectory. Through this mixed state trajectory, the system can simultaneously express equipment numerical changes and state switching processes, rather than judging anomalies based on a single parameter at a single moment.

[0034] Simultaneously, edge computing nodes construct device relationships based on device identifiers and configuration data carried in standardized power equipment operation data; power supply relationships are formed based on device access locations and power supply connection paths; load relationships are formed based on device access locations and load access paths; and environmental relationships are formed based on the installation locations of environmental acquisition devices and their access locations. Through these device relationships, the system can determine whether a status change of a power monitoring object in one data center will affect power monitoring objects in other data centers, avoiding the treatment of multiple related alarms as isolated events.

[0035] During fault prediction, edge computing nodes input the hybrid state trajectory of equipment into the edge hybrid fault prediction algorithm. The algorithm jointly identifies continuous state feature data and discrete state transition data from the hybrid state trajectory to obtain the equipment state evolution result. This result includes normal operation, fluctuating operation, slight degradation, continuous degradation, and high-risk operation. If the load rate, temperature, and current of a UPS power supply continuously change, and related PDUs, battery banks, or distribution cabinets show state switching or alarm records, the system will combine continuous changes and discrete transitions for judgment. Furthermore, it will perform a linkage risk analysis based on power supply correlation, load correlation, and environmental correlation to obtain the linkage fault risk result.

[0036] Edge computing nodes generate fault prediction results based on device status evolution and associated fault risk outcomes. These predictions include the fault risk level, fault type, impact scope, and maintenance reminders. Subsequently, the edge computing nodes send these predictions to the data center power equipment monitoring platform and on-site alarm terminals, generating local alarm records accordingly. Maintenance personnel can then view the risk source and associated devices based on the fault type and impact scope, and proactively perform checks, load adjustments, heat dissipation measures, or equipment maintenance based on the maintenance reminders.

[0037] Through the above implementation process, the present invention can complete the processing of power equipment operation data, the construction of mixed equipment state trajectories, the construction of equipment correlation relationships, and the analysis of linkage fault risks at the data center site. Compared with the method of relying solely on threshold alarms, the present invention can combine continuous changes in operating parameters and discrete operating state jumps to identify the pre-fault evolution trend, and combine equipment correlation relationships to determine the scope of risk impact, thereby improving the timeliness of data center power equipment monitoring, the accuracy of fault prediction, and the efficiency of on-site operation and maintenance.

[0038] Table 1 Comparison of Fault Prediction and Linkage Early Warning Effects of Computer Room Power Equipment

[0039] As shown in Table 1, the fault pre-identification accuracy of the method of this invention is 89.6%, which is 12.8 percentage points higher than the 76.8% of the traditional threshold alarm method and 5.5 percentage points higher than the 84.1% of the ordinary trend prediction method. This improvement is mainly due to the fact that this invention does not rely solely on a single parameter such as voltage, current, temperature, load rate, or battery capacity for judgment. Instead, it constructs continuous operating parameters as continuous state feature data, and simultaneously constructs power supply status, energy storage status, cooling status, power distribution status, and maintenance recovery status as discrete state transition data. Then, it identifies the equipment state evolution results through the mixed state trajectory of the equipment. Therefore, when the power equipment has not yet reached the threshold alarm condition but has already shown a continuous degradation state or a high-risk operating state, this invention can identify the pre-fault evolution trend in advance.

[0040] Regarding the average early warning time, the method of this invention achieves 52.7 minutes, compared to 18.5 minutes for the traditional threshold alarm method and 31.2 minutes for the ordinary trend prediction method. This invention achieves an early warning time 34.2 minutes faster than the traditional threshold alarm method and 21.5 minutes faster than the ordinary trend prediction method. This result demonstrates that this invention, through a hybrid fault prediction algorithm, jointly identifies continuous state characteristic data and discrete state transition data. It can establish a time correlation between continuous changes in UPS load, abnormal battery discharge status, the occurrence of alarm status in distribution branches, and the switching of precision air conditioning operating status, thereby generating fault prediction results before obvious faults occur.

[0041] Regarding the number of false alarms and false negatives, traditional threshold alarm methods resulted in 31 and 17 false alarms respectively, while ordinary trend prediction methods resulted in 24 and 12 false alarms respectively. The method of this invention reduces these numbers to 11 and 5 respectively. The reduction in false alarms is due to the fact that this invention does not directly classify single temperature fluctuations, load fluctuations, or short-term state transitions as fault risks. Instead, it comprehensively identifies fault risks by combining the direction of change, amplitude of change, duration, frequency of change, transition sequence, number of transitions, state duration, and interval between adjacent states. The reduction in false negatives is due to the invention's ability to identify situations where continuous operating parameters have not yet exceeded limits but state transitions are abnormally concentrated, making the pre-fault evolution process less likely to be ignored.

[0042] In terms of the accuracy of linked risk identification, the method of this invention achieves 88.9%, significantly better than the 62.4% of traditional threshold alarm methods and the 70.5% of ordinary trend prediction methods. Traditional threshold alarm methods typically only output alarms for individual devices, making it difficult to determine the influence relationships between UPS power supplies, PDUs, battery banks, distribution cabinets, precision air conditioners, and environmental data acquisition equipment. This invention constructs device correlation relationships based on device configuration data, incorporating power supply correlation relationships, load correlation relationships, and environmental correlation relationships into the fault prediction process. Therefore, it can determine whether a change in the state of a power supply device will affect power distribution equipment or energy storage equipment along the power supply connection path, and it can also determine whether a change in the state of a cooling device will affect the power equipment in the corresponding cabinet through environmental correlation relationships.

[0043] Regarding local alarm response time, the method of this invention achieves 5.4 seconds, compared to 12.6 seconds for traditional threshold alarm methods and 8.9 seconds for ordinary trend prediction methods. This metric reflects the role of edge computing nodes. This invention completes standardized power equipment operation data processing, equipment mixed state trajectory construction, equipment state evolution identification, and linkage fault risk analysis at the data center site. It eliminates the need to upload all raw data for unified judgment by a remote platform, thus shortening the time for generating fault prediction results and local alarm records, and improving on-site alarm response capabilities.

[0044] As can be seen from the data in Table 1, the method of the present invention outperforms traditional threshold alarm methods and ordinary trend prediction methods in six aspects: accuracy of fault pre-identification, average early warning time, number of false alarms, number of missed alarms, accuracy of linkage risk identification, and local alarm response time. The core reason for its performance improvement is that the present invention simultaneously utilizes continuous operating parameters and discrete operating states to construct mixed state trajectories of equipment, and combines equipment correlation relationships to conduct linkage risk analysis. This allows the fault prediction results to not only reflect the state changes of a single data center power monitoring object, but also reflect the risk transmission relationship between multiple data center power monitoring objects, thereby improving the timeliness, accuracy, and operation and maintenance guidance value of data center power equipment monitoring.

[0045] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An edge-computing and failure-prediction-based machine room power equipment monitoring system, characterized in that, Includes the following steps: The data acquisition module is used to collect the operating data and equipment configuration data of the power equipment of the monitoring objects in the computer room; The edge processing module is used to perform edge standardization processing on the power equipment operation data to obtain standardized power equipment operation data. The status extraction module is used to extract continuous operating parameters and discrete operating states based on standardized power equipment operating data. The feature construction module is used to construct continuous state feature data based on continuous operating parameters and discrete state transition data based on discrete operating states. The hybrid trajectory construction module is used to construct hybrid state data of the device and arrange it in time sequence to obtain the hybrid state trajectory of the device. The association construction module is used to construct equipment associations based on the equipment identifiers and equipment configuration data carried in the standardized power equipment operation data. The fault prediction module is used to input the mixed state trajectory of the equipment into the edge mixed fault prediction algorithm, jointly identify the continuous state feature data and discrete state jump data to obtain the equipment state evolution result, and perform linkage risk analysis based on the equipment correlation relationship to obtain the linkage fault risk result. The result generation module is used to generate fault prediction results based on the equipment status evolution results and the linkage fault risk results; The alarm output module is used to send fault prediction results and generate local alarm records. 2.The machine room power equipment monitoring system based on edge computing and failure prediction of claim 1, wherein, The data acquisition module includes: The power monitoring objects in the computer room include power supply equipment, energy storage equipment, power distribution equipment, cooling equipment, backup power generation equipment, and environmental acquisition equipment. The power supply equipment includes UPS power supplies and PDUs, the energy storage equipment includes battery packs, the power distribution equipment includes distribution cabinets, the cooling equipment includes precision air conditioners, the backup power generation equipment includes generators, and the environmental acquisition equipment includes temperature and humidity sensors. The power equipment operation data includes continuously collected data, status switching data, alarm record data, and maintenance record data. The equipment configuration data includes equipment access location, power supply connection path, load access path, and installation location of the environmental acquisition equipment. 3.The machine room power equipment monitoring system based on edge computing and failure prediction of claim 1, wherein, The edge processing module includes: The edge standardization process includes protocol conversion, timestamp unification, device identifier binding, missing data completion, abnormal data removal, and data format conversion.

4. The machine room power equipment monitoring system based on edge computing and failure prediction of claim 1, wherein, The state extraction module includes: Voltage, current, power, load rate, temperature, humidity, and battery capacity values ​​are extracted from continuously collected data in standardized power equipment operation data. Based on the voltage, current, power, load rate, temperature, humidity, and battery capacity data collected, voltage change sequences, current change sequences, power change sequences, load rate change sequences, temperature change sequences, humidity change sequences, and battery capacity change sequences are generated and aggregated into continuous operating parameters. Extract power supply status records, energy storage status records, cooling status records, power distribution status records, and maintenance recovery status records from the status switching data, alarm record data, and maintenance record data in the standardized power equipment operation data; Power supply status is generated based on power supply status records, energy storage status is generated based on energy storage status records, cooling status is generated based on cooling status records, power distribution status is generated based on power distribution status records, and maintenance and recovery status is generated based on maintenance and recovery status records. Power supply status, energy storage status, cooling status, power distribution status, and maintenance and recovery status are then aggregated into discrete operating statuses.

5. The machine room power equipment monitoring system based on edge computing and failure prediction according to claim 1, characterized in that, The feature construction module includes: The voltage change sequence, current change sequence, power change sequence, load rate change sequence, temperature change sequence, humidity change sequence, and battery capacity change sequence in the continuous operating parameters are compared with adjacent sequence values ​​to obtain the direction and magnitude of change. The continuous acquisition interval that maintains the same direction of change is determined based on the direction of change. The duration is obtained based on the start and end times of the continuous acquisition interval. The change frequency is obtained based on the number of times the direction of change is switched between adjacent continuous acquisition intervals. The direction, magnitude, duration, and frequency of change are bound to the voltage change sequence, current change sequence, power change sequence, load rate change sequence, temperature change sequence, humidity change sequence, and battery capacity change sequence to obtain continuous state characteristic data. The jump sequence is obtained according to the order in which the power supply state, energy storage state, cooling state, power distribution state, and maintenance and recovery state occur in the discrete operation state, and the jump number is obtained according to the number of times the power supply state, energy storage state, cooling state, power distribution state, and maintenance and recovery state are switched. The duration of a state is obtained from the start and end times of the same state in discrete running states, and the interval between adjacent states is obtained from the end and start times of two adjacent discrete running states. The jump order, jump count, state duration, and interval between adjacent states are bound to the corresponding power supply state, energy storage state, cooling state, power distribution state, and maintenance and recovery state to obtain discrete state jump data. 6.The machine room power equipment monitoring system based on edge computing and failure prediction of claim 1, wherein, The hybrid trajectory construction module includes: The continuous state characteristic data and discrete state transition data are aggregated according to the device identifier to obtain the continuous state characteristic data and discrete state transition data corresponding to the same device identifier. Based on the timestamp, the continuous state feature data and discrete state transition data corresponding to the same device identifier are time-aligned to obtain the continuous state feature data and discrete state transition data corresponding to the same device identifier and the same timestamp; By binding continuous state feature data, discrete state transition data, device identifier and timestamp corresponding to the same device identifier and the same timestamp, mixed state data of the device is obtained. Arrange the mixed state data of devices corresponding to the same device identifier in a time sequence according to the timestamp to obtain the mixed state data sequence of devices. According to the arrangement order of the mixed state data in the mixed state data sequence, the adjacent mixed state connection relationship is established between two adjacent mixed state data, and the mixed state data sequence and the adjacent mixed state connection relationship are combined to obtain the mixed state trajectory of the equipment.

7. The machine room power equipment monitoring system based on edge computing and failure prediction of claim 1, wherein, The association construction module includes: Extract equipment identifiers from standardized power equipment operation data, and extract equipment access locations, power supply connection paths, load access paths, and installation locations of environmental acquisition equipment from equipment configuration data; Based on the device identifier and the device access location, the device identifier is matched with the device access location to obtain the device location correspondence; Based on the device location correspondence and power supply connection path, the device identifiers with power supply connections in the power supply connection path are matched to obtain the power supply association relationship; Based on the device location correspondence and load access path, the device identifiers with load access in the load access path are matched to obtain the load association relationship; Based on the device location correspondence and the installation location of the environmental acquisition device, the device identifier of the environmental acquisition device is matched with the device identifier of the device access location corresponding to the installation location of the environmental acquisition device to obtain the environmental association relationship; The power supply association, load association, and environmental association are bound according to the device identifier to obtain the device association. 8.The machine room power equipment monitoring system based on edge computing and failure prediction of claim 1, wherein, The fault prediction module includes: Extract the hybrid state trajectory points corresponding to the same device identifier according to the time sequence in the hybrid state trajectory of the device to obtain the hybrid state point sequence; The continuous state feature data contained in adjacent mixed state trajectory points are read sequentially from the mixed state point sequence. The change direction, change amplitude, duration and change frequency in the continuous state feature data are concatenated in time order to obtain the continuous state change process. The discrete state transition data contained in adjacent mixed state trajectory points are read sequentially from the mixed state point sequence. The transition order, number of transitions, state duration and adjacent state interval time in the discrete state transition data are concatenated in time order to obtain the discrete state transition process. A continuous direction sequence is generated based on the direction of change during the continuous state change process; a continuous amplitude sequence is generated based on the amplitude of change during the continuous state change process; a continuous duration sequence is generated based on the duration of change during the continuous state change process; and a continuous frequency sequence is generated based on the frequency of change during the continuous state change process. The continuous direction sequence, continuous amplitude sequence, continuous duration sequence, and continuous frequency sequence are then aggregated into continuous evolution recognition data. A discrete jump sequence is generated based on the jump order in the discrete state jump process, a discrete number sequence is generated based on the number of jumps in the discrete state jump process, a discrete duration sequence is generated based on the state duration in the discrete state jump process, and a discrete interval sequence is generated based on the time interval between adjacent states in the discrete state jump process. The discrete jump sequence, discrete number sequence, discrete duration sequence and discrete interval sequence are then collected into discrete evolution recognition data. The continuous evolution identification data is input into the edge hybrid fault prediction algorithm, and the continuous direction sequence, continuous amplitude sequence, continuous duration sequence and continuous frequency sequence are combined for identification to obtain the continuous evolution label corresponding to each hybrid state trajectory point; The discrete evolution identification data is input into the edge hybrid fault prediction algorithm to combine and identify discrete jump sequences, discrete number sequences, discrete duration sequences and discrete interval sequences to obtain the discrete evolution labels corresponding to each hybrid state trajectory point. Following the order of high-risk operating state, continuous degradation state, slight degradation state, fluctuating operating state, and normal operating state, the continuous evolution markers and discrete evolution markers corresponding to the same mixed state trajectory points are matched to obtain the normal operating state, fluctuating operating state, slight degradation state, continuous degradation state, and high-risk operating state, which are then aggregated into the equipment state evolution results. Based on the device association relationships, the device status evolution results corresponding to the device identifiers with power supply association relationships, load association relationships, and environmental association relationships are matched to obtain the associated status evolution data; Based on the device identifier, time sequence, and device association relationship corresponding to the device state evolution results in the associated state evolution data, the risk result of linkage failure is determined. 9.The machine room power equipment monitoring system based on edge computing and failure prediction of claim 1, wherein, The result generation module includes: Based on the equipment status evolution results, the fault risk level is determined as follows: normal operation status, fluctuating operation status, slight degradation status, continuous degradation status, and high-risk operation status. The fault type is determined based on the state category corresponding to the equipment state evolution result and the equipment association relationship corresponding to the linkage fault risk result; The scope of impact is determined based on the equipment relationships and the corresponding computer room power monitoring objects based on the fault type; Generate maintenance reminders based on the fault risk level, fault type, and scope of impact; The fault risk level, fault type, scope of impact, and maintenance reminders are compiled into fault prediction results. 10.The machine room power equipment monitoring system based on edge computing and failure prediction of claim 1, wherein, The alarm output module includes: The fault prediction results are read from the edge computing nodes and converted into fault prediction data for transmission. The fault prediction data is sent to the data center power equipment monitoring platform and on-site alarm terminal that are connected to the edge computing node through the edge computing node; Local alarm records are generated based on the fault prediction results.

Images