A power equipment multi-source monitoring data fusion processing method and system

Abstract

The application relates to the technical field of power equipment monitoring, and discloses a power equipment multi-source monitoring data fusion processing method and system, which comprises the following steps: receiving and sending data streams and performing time axis alignment, separating main features, secondary features and basic environment data streams; monitoring the main feature data stream to generate event anchor points, and combining spatial conduction parameters to generate a reference physical time delay; correcting the reference physical time delay by using the basic environment data stream to generate a medium state time delay drift window; mapping the medium state time delay drift window into a target logical block interval set, encapsulating the logical moving pointer and binding it to the event anchor point landing disk; searching and extracting the underlying data based on the interval set, and performing time axis cutting on the decompressed data by using the medium state time delay drift window. The application corrects the physical wave speed offset in combination with the running working condition, and decouples the time addressing from the underlying physical storage, thereby reducing the index maintenance cost caused by data reorganization and improving the precision of multi-source data fusion alignment.

Description

Technical Field

[0001] This invention relates to the field of power equipment monitoring technology, specifically to a method and system for fusion processing of multi-source monitoring data of power equipment. Background Technology

[0002] Online monitoring of modern power equipment typically relies on multi-source heterogeneous sensor networks to collect state and signal data from different dimensions, such as partial discharge, temperature, and vibration, for comprehensive diagnosis of equipment health. In industrial environments, the sampling frequencies and packet transmission mechanisms of various sensors differ significantly, and data transmission is susceptible to communication network jitter, leading to clock synchronization errors and timing discrepancies in the data packets received.

[0003] When performing multi-source data fusion analysis, conventional processing methods typically set a fixed time window to truncate and align monitoring data from different dimensions. However, the physical properties of the insulating medium inside power equipment change with variations in operating conditions. For example, temperature changes in insulating oil can alter the medium density, leading to a shift in the actual propagation velocity of secondary physical signals such as ultrasonic waves within the medium. A fixed time alignment window lacks consideration of the equipment's real-time operating status and cannot adapt to this physical-level wave velocity delay drift, easily resulting in truncation or missed detection of abnormal event-related response data.

[0004] Meanwhile, industrial monitoring systems typically write the collected data into time-series databases for persistent storage. These databases generally employ variable-length compression algorithms to process continuous data points. Because the compression rate varies across different time periods, the physical sector addresses of compressed data blocks in the disk file system dynamically change due to database backend merging and reorganization operations. If existing data retrieval mechanisms directly and forcibly bind absolute physical addresses to upper-layer exception event indexes, the system must synchronously track and modify all associated index records when underlying data blocks migrate. This increases the disk addressing load and index maintenance costs under high-concurrency write environments, making it difficult to support the system's real-time access and causal retrieval needs for heterogeneous sensor data. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for fusion processing of multi-source monitoring data of power equipment. It solves the problems of clock synchronization errors and communication delays in heterogeneous sensors in industrial fields, and the dynamic changes in the insulation properties of power equipment with operating status, which cause the propagation speed of secondary physical signals to shift. Using a fixed time window for alignment can easily lead to data truncation or missed detection. In addition, the variable-length compression algorithm of time-series databases causes dynamic shifts in the physical storage address of data. Directly using absolute physical addresses to build indexes will increase the index maintenance cost when the underlying data is reorganized, making it difficult to meet the real-time access requirements of high-frequency multi-source monitoring data.

[0006] To address the above problems, the present invention provides the following technical solution: The first aspect of this invention provides a method for fusing and processing multi-source monitoring data of power equipment, comprising: Receive concurrent data streams and perform timeline alignment operations to separate the main feature data stream, secondary feature data stream, and basic environment data stream; Monitor the main feature data stream, generate event anchors when abnormal events are identified, and generate a reference physical delay by combining the spatial conduction parameters of the power equipment; Based on the underlying environmental data stream matched with the event anchor point, the baseline physical delay is corrected for wave velocity, and a secondary feature data stream is generated to respond to the medium-state delay drift window of the anomalous event. The medium-state delay drift window is mapped to a set of target logical block intervals, encapsulated to generate a logical floating pointer and bound to the event anchor point, and asynchronous persistence to disk is performed. The system receives the fusion retrieval command, obtains the logical floating pointer based on the event anchor point, extracts the underlying raw data based on the target logical block interval set, and performs time axis clipping on the decompressed underlying raw data using the medium-state delay drift window to obtain the fusion processing result.

[0007] The above method uses the basic environmental data stream to correct the baseline physical delay, obtains the medium state delay drift window under environmental state changes, and then maps the continuous time interval to a set of target logic block intervals to encapsulate the logic roaming pointer, decoupling the time addressing from the physical storage address of the underlying data.

[0008] Furthermore, perform timeline alignment operations, including: Extract the absolute timestamp of each data packet in the concurrent data stream; calculate and update the event watermark in real time based on the preset maximum tolerable delay time window and the current maximum received event timestamp value; When the absolute timestamp of the data block to be processed is less than or equal to the current event water level, the corresponding multi-source monitoring data message is confirmed to be received completely, and the separation operation to the main feature data stream, secondary feature data stream and basic environmental data stream is performed.

[0009] When processing concurrent data streams, the event watermark is calculated using timestamps and a set delay time window to determine the integrity of data packet reception.

[0010] Furthermore, the main feature data stream is monitored, and event anchors are generated when abnormal events are identified, including: Perform energy integration calculation of instantaneous amplitude on the main feature data stream using a sliding time window; When the calculated energy integral value is greater than or equal to the preset baseline threshold, the start timestamp of the corresponding data slice is marked as the trigger time. Event anchors containing sensor device physical identifiers, trigger times, and energy extremes are generated in the inverted index tree. The generation of event anchors is triggered by the comparison results of energy integral values ​​and benchmark thresholds to filter high-frequency transient noise signals.

[0011] Furthermore, a reference physical delay is generated by combining the spatial conduction parameters of the power equipment, including: Call the system's preset internal conduction distance parameters of power equipment and reference physical wave velocity parameters; The reference physical delay is calculated by using the ratio of the transmission distance parameter to the reference physical wave velocity parameter. This reference physical delay reflects the time difference of the physical signal propagation in the transmission medium of the device under standard environmental conditions.

[0012] Furthermore, a wave velocity correction is applied to the baseline physical delay to generate a secondary characteristic data stream response to the medium-state delay drift window of this anomalous event, including: Using the trigger time as the address key, adjacent real sampling points are extracted and linearly interpolated to calculate the instantaneous medium temperature value; By introducing the temperature coefficient of the medium's physical properties, the reference physical wave velocity parameters are corrected based on the difference between the instantaneous medium temperature and the preset reference temperature, and the actual propagation wave velocity is obtained. The actual physical time delay is calculated using the ratio of the transmission distance parameter to the actual transmission wave velocity; Based on the actual physical time delay, a time tolerance term is introduced, and an absolute time closed interval is constructed in combination with the trigger time to generate a medium state time delay drift window. By introducing temperature parameters and physical property coefficients, the wave velocity is dynamically corrected to obtain a medium state time delay drift window that matches the current operating conditions.

[0013] Furthermore, the medium-state delay drift window is mapped to a set of target logic block intervals, and a logic roaming pointer is encapsulated and generated, including: Logically slice the secondary feature data stream according to a fixed time span; Substitute the upper and lower boundary timestamps of the medium-state delay drift window into the mapping formula and perform a floor operation to complete the continuous integer numbering, thereby generating a discrete set of target logic block intervals. A logical roaming pointer is constructed in memory, containing the target data stream identifier, the target logical block interval set, and the unmapped medium-state delay drift window. This process converts the absolute time boundary into discrete logical block numbers and retains the original time window parameters in the logical roaming pointer.

[0014] Furthermore, asynchronous persistence to disk is performed, including: The logical floating pointer is used as a data value and key-value bound to the event anchor point, and submitted to the asynchronous persistent queue of the inverted index file for disk writing; Simultaneously, the slicing, compression, and disk writing operations of the secondary feature data stream are performed in parallel, using a separate asynchronous writing mechanism to write time-series data to disk while maintaining index data.

[0015] Furthermore, based on the target logical block interval set, the underlying raw data is extracted, including: Parse the target logical block interval set from the logical roaming pointer and extract the target data stream identifier; Extract the integer number from the target logical block range set as an addressing clue, and access the metadata management table to obtain the physical sector address of the disk. The target compressed logical block is extracted using a direct input / output mechanism based on the physical sector address of the disk, and decompressed into a continuous time-series data point array containing the original physical timestamp and monitoring values. During retrieval, the logical number is restored to the physical sector address based on the metadata management table for data reading.

[0016] Furthermore, the timeline is pruned using a medium-state delay drift window on the decompressed underlying raw data, including: Using the absolute time closed interval determined by the medium-state delay drift window as the boundary, a fixed-point stripping is performed on the continuous time-series data point array to remove data points outside the absolute time closed interval; If the set of valid data points after cropping is empty, mark the result and the main feature data as isolated trigger events. If not empty, the result is packaged into a structured fusion processing result and returned. The medium-state delay drift window contained in the logical floating pointer is used to perform a secondary time boundary truncation on the decompressed and restored coarse-grained data block.

[0017] A second aspect of the present invention provides a multi-source monitoring data fusion processing system for power equipment, comprising: The data acquisition module is used to collect status data and signal data of the internal physical medium of the target power equipment, and generate and send concurrent data streams; The access buffer module communicates with the data acquisition module to receive concurrent data streams and perform time axis alignment operations, separating the main feature data stream, secondary feature data stream, and basic environment data stream. The storage engine module, connected to the access buffer module, is used to monitor the main feature data stream generation event anchor point and baseline physical latency, correct the generated medium-state latency drift window based on the basic environment data stream, encapsulate and generate a logical floating pointer, and perform asynchronous persistence to disk. In response to the fusion retrieval command, the underlying raw data is extracted based on the target logical block interval set and time axis clipping is performed using the medium-state delay drift window; The application processing module communicates with the storage engine module to submit fusion retrieval commands to the storage engine module and receive fusion processing results.

[0018] This invention provides a method and system for fusing and processing multi-source monitoring data from power equipment. It has the following beneficial effects: 1. This invention extracts the real-time medium temperature value from the basic environmental data stream, introduces the medium physical property temperature coefficient to dynamically calculate the reference physical wave velocity of the device, and generates a medium state time delay drift window for the secondary feature data stream. This processing method combines the device operating conditions to compensate for the transmission time of the physical signal, avoiding data omission and truncation caused by wave velocity drift when performing multi-dimensional data alignment using a fixed time window.

[0019] 2. This invention maps the time closed interval established by the media-state delay drift window to a discrete set of target logical block intervals, which are then encapsulated into logical floating pointers and bound to event anchor points for disk storage. This two-stage addressing mechanism decouples time characteristics from underlying physical storage addresses. When the underlying data physical sectors change due to variable-length compression or background merging and reorganization of the time-series database, there is no need to synchronously rewrite the fixed logical floating pointers, reducing the index maintenance cost in a concurrent environment.

[0020] 3. In the data retrieval stage, this invention uses the physical sector address corresponding to the target logic block interval set to extract the underlying coarse-grained data block. Then, it calls the precise medium-state delay drift window retained in the logic roaming pointer to perform time axis pruning on the decompressed continuous time-series data point array. This combines the access efficiency of discrete addressing and removes irrelevant data points at both ends of the logic block through fixed-point stripping, outputting the fusion feature result with precise time boundary alignment to the application layer. Attached Figure Description

[0021] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a flowchart of the method of the present invention; Figure 3 This is a timing diagram of streaming data access and alignment according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the event triggering and baseline physical mapping principle of an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the principle of adaptive calibration of the medium state according to an embodiment of the present invention. Figure 6 This is a schematic diagram illustrating the two-stage decoupling index construction principle of an embodiment of the present invention. Figure 7 This is a schematic diagram illustrating the physical causal fusion retrieval principle of an embodiment of the present invention.

[0022] Among them, 10 is the data acquisition module; 20 is the access buffer module; 30 is the storage engine module; and 40 is the application processing module. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example: Please see the appendix Figure 1 This invention provides a multi-source monitoring data fusion processing system for power equipment, comprising: The data acquisition module 10 is installed at the target power equipment. The data acquisition module 10 includes heterogeneous sensors with multiple sampling frequencies. These heterogeneous sensors are deployed in fixed positions according to the three-dimensional spatial topology of the power equipment, and are used to collect state data and signal data of the internal physical medium of the power equipment.

[0025] The access buffer module 20 is deployed in the memory space of the database server. The access buffer module 20 establishes a communication connection with the data acquisition module 10 through the industrial network bus, and is used to receive and buffer concurrent data streams generated by the underlying sensors.

[0026] Storage engine module 30 is deployed on the underlying time-series database server. Storage engine module 30 is connected to access buffer module 20 and is responsible for compressing and persisting streaming data slices to the disk file system, while building the underlying inverted index file based on the physical causal mechanism.

[0027] Application processing module 40 is deployed on the industrial data processing application layer server. Application processing module 40 communicates with storage engine module 30 through a database query interface to initiate multidimensional data retrieval requests and receive aligned data result sets for downstream diagnostic analysis.

[0028] Please see the appendix Figure 2 This invention provides a method for fusing and processing multi-source monitoring data of power equipment, comprising the following steps: S11, the access buffer module 20 receives the concurrent data stream sent by the data acquisition module 10, extracts the physical timestamp from each data packet, performs time axis alignment operation on the concurrent data stream in the memory space according to the preset water level determination rule, and separates the main feature data stream, secondary feature data stream and basic environment data stream. S12, the storage engine module 30 continuously monitors the aligned main feature data stream. When the storage engine module 30 calculates that the energy integral of a certain main feature data slice exceeds the set benchmark threshold, it generates an event anchor point associated with the timestamp in the memory inverted index tree and calls the system's preset three-dimensional transmission distance and benchmark physical wave velocity to generate a benchmark physical delay. S13, the storage engine module 30 extracts the specific values ​​of the basic environmental data stream that match the occurrence time of the event anchor point from the access buffer module 20, uses the values ​​of the basic environmental data stream to substitute into the physical property coefficient formula to correct the physical wave velocity, and generates the secondary feature data stream to respond to the accurate medium state delay drift window of the physical event. S14, the storage engine module 30 discretizes and maps the media-state delay drift window obtained from the previous calculation according to the fixed-time block division rules set by the underlying database to generate a target logical block interval set. It encapsulates the target data stream identifier, the target logical block interval set, and the media-state delay drift window into a logical floating pointer. It binds the logical floating pointer as a key-value pair to the event anchor point and performs asynchronous persistence to disk. At the same time, the system performs regular slice compression and disk write operations on the secondary feature data stream. S15, the application processing module 40 sends an abnormal event fusion retrieval instruction to the storage engine module 30. The storage engine module 30 hits the event anchor point in the underlying index file according to the retrieval instruction and deserializes the logical floating pointer. It extracts the target logical block interval set as the addressing clue, accesses the metadata management table of the storage system to obtain the physical sector address of the disk, extracts the target compressed logical block to the memory buffer and decompresses it using the direct input / output mechanism, performs time axis trimming on the decompressed data according to the time boundary established by the media state delay drift window carried by the logical floating pointer, and packages the filtered primary feature data and secondary feature data into a result set and returns it to the application processing module 40.

[0029] See attached document Figure 3 The specific implementation of receiving concurrent data streams and performing alignment operations in step S11 of the present invention includes the following steps: S111, in the initial access phase of multi-source data, due to the variety of sensors and different communication transmission mechanisms in industrial sites, the system needs to establish a unified data classification index at the memory level.

[0030] Specifically, the access buffer module 20 receives monitoring data messages sent in real time by the data acquisition module 10 via the industrial bus. In order to achieve accurate time-series traceability of the underlying data, the access buffer module 20 parses the header data structure of each monitoring data message and extracts the physical identifier of the sensor device and the high-precision absolute timestamp of the data generation time.

[0031] Subsequently, the access buffer module 20 performs key-value matching between the extracted sensor device physical identifier and the preset metadata mapping table in the system memory. According to the preset configuration parameters of the mapping table, the access buffer module 20 routes the aliased concurrent data packets on the bus input side to different data stream queues in the memory space.

[0032] In this embodiment, the system classifies the data stream queues into primary feature data streams, secondary feature data streams, and basic environment data streams based on the sampling rate and physical attributes.

[0033] S112, the access buffer module 20 allocates sliding buffer areas with fixed capacity boundaries for the various data stream queues after separation in the memory space, thereby constructing a water level alignment mechanism.

[0034] In actual substation operation environments, network jitter often occurs during the operation of industrial communication networks. Furthermore, the packet transmission cycle of the sensors to which the basic environmental data stream belongs is often longer than that of the sensors to which the main feature data stream belongs, resulting in out-of-order and reversed physical processing times for data packets of different physical dimensions arriving at the access buffer module 20.

[0035] To avoid the anomalies caused by the aforementioned delays in subsequent status parameters, the access buffer module 20 mainly relies on the high-precision absolute timestamps inherent in the data packets to unidirectionally advance the system's operating logic time, rather than relying on the physical reception time of the server's local location.

[0036] S113, the access buffer module 20 calculates and updates the system's event water level in real time according to the set hardware clock tick frequency.

[0037] The specific formula for calculating the water level in this event is as follows: ; In this calculation formula, This indicates the timestamp value of the event watermark currently established by the access buffer module 20; This represents the maximum event timestamp value parsed by the access buffer module 20 in all current receive buffer queues; This represents a constant parameter representing the maximum tolerable delay time window that the system pre-sets to cope with the latency of underlying network packets.

[0038] As a preferred method, this The specific value can be dynamically adjusted based on the statistical distribution results of historical packet loss and latency tests of the target industrial network to cover the vast majority of common network jitter. In a typical substation 100 Mbps Ethernet environment, its value range is usually set between 50 and 500 milliseconds to achieve the best balance between multi-source data integrity and real-time system diagnostics.

[0039] S114, the access buffer module 20 traverses the data blocks to be processed in each sliding buffer area through the background polling process, and performs an arithmetic comparison between the event timestamp attached to each data block to be processed and the event water level value currently updated by the system.

[0040] When the event timestamp value corresponding to a certain data block is less than or equal to the current event water level value, the access buffer module 20 confirms at the logical level that the time segment and the multi-source monitoring data messages generated before it have been completely received.

[0041] Subsequently, the access buffer module 20 marks the data block status that meets the time section condition as aligned and releases the memory access read / write lock, allowing the downstream storage engine module 30 to perform event anchor monitoring and status parameter extraction operations on the aligned main feature data stream and basic environment data stream.

[0042] In addition, to prevent individual sensor devices from going offline or experiencing prolonged network outages... Stagnation can lead to extreme situations where the memory cache becomes full and deadlock occurs. The access buffer module 20 is also equipped with a timeout detection mechanism based on the local physical clock.

[0043] If detected If no updates occur within the set keep-alive threshold, the system will forcibly advance the event waterline. Data blocks that fail to arrive within the timeout period are treated as lost packets, thus ensuring the continuous flow of the entire system's analysis chain.

[0044] For the specific network transmission communication protocol of the underlying data packets and the allocation method of hardware-level high-precision timestamps, those skilled in the art can use conventional industrial Ethernet transmission specifications and precise time synchronization protocols to implement them. The network packet parsing process and clock synchronization hardware control logic are well-known technologies in this field and will not be described in detail here.

[0045] See attached document Figure 4 The specific implementation of monitoring the main feature data stream and generating the reference physical delay in step S12 of the present invention includes the following steps: S121 After completing the access of multi-source data and timeline alignment, the system needs to accurately capture the starting point of abnormal events from massive amounts of normal operation data.

[0046] The storage engine module 30 continuously acquires the aligned master feature data stream. In order to identify potential physical mutations inside the power equipment and filter transient high-frequency noise interference, the storage engine module 30 performs energy integral calculation of the instantaneous amplitude of the master feature data stream at a fixed time step.

[0047] From the perspective of signal processing principles, physical anomalies such as partial discharge are usually accompanied by continuous energy release, while sporadic electromagnetic interference noise often manifests as extremely short pulses. Simple amplitude detection is easily triggered by such random pulses, while energy integration mechanisms can effectively smooth transient noise and highlight events with a certain amount of energy accumulation.

[0048] To avoid truncating abnormal signals due to a fixed time window, thus preventing missed detection dead zones, this embodiment uses a sliding time window for integration calculation. Let the instantaneous amplitude of the main feature data stream be... The time step of the sliding integral is The energy integral value within the current observation window The calculation formula is: ; In this calculation formula, Represents the integration variable time and the time step. The value of is usually determined by the sampling frequency of the main feature data stream and the typical width of the monitored physical anomaly pulse.

[0049] As a preferred embodiment, this time step is used when monitoring ultra-high frequency partial discharge signals. The magnitude of this value is usually matched with the duration of the abnormal discharge pulse.

[0050] Storage engine module 30 will calculate the energy integral value in real time. Compared with the preset benchmark threshold A comparison is performed. This benchmark threshold... It can be determined comprehensively based on the historical statistical average value of background noise energy collected when the target power equipment is operating under normal fault-free conditions, combined with the safety margin coefficient set in engineering practice.

[0051] S122, during continuous sliding monitoring, the energy integral value of a certain main feature data slice is determined. Greater than or equal to the baseline threshold In this case, the system identifies it as a potential physical event of equipment malfunction and marks the start timestamp corresponding to that data slice as the trigger time. .

[0052] Based on the trigger identification result, the storage engine module 30 generates a value in the underlying inverted memory index tree that corresponds to the trigger time. Strongly correlated event anchors. In this embodiment, the event anchor is a lightweight index node structure at the logical level, which internally records the physical identifier of the main characteristic sensor device that triggered the event and the trigger time. And the corresponding energy limit extreme value, which serves as the primary key for subsequent joint cross-retrieval of multi-source heterogeneous data.

[0053] S123, after establishing the event anchor point, the system needs to construct the causal transmission relationship between data streams of different physical dimensions, that is, to transform the physical spatial geometric relationship of the device into a time dimension mapping value.

[0054] Specifically, the storage engine module 30 calls the system's preset internal power transmission distance parameters. Compared with the reference physical wave velocity parameters Transmission distance parameter The physical distance in three-dimensional space between the monitoring point that generates the primary characteristic signal and the sensor that senses the secondary characteristic signal is typically written as static metadata into the configuration table of the database during the substation sensor engineering deployment and topology survey phase. Reference physical wave velocity parameter. It represents the propagation speed of a physical signal in the insulating medium of electrical equipment under standard environmental conditions. For example, it represents the theoretical constant wave speed of an ultrasonic signal propagating in the insulating oil inside a transformer cavity at a set reference insulating oil temperature of 20 degrees Celsius.

[0055] S124, after extracting the above spatial parameters and medium physical property parameters, the storage engine module 30 calculates the baseline physical delay under ideal conditions by the ratio of conduction distance to propagation speed.

[0056] The benchmark physical delay The specific calculation formula is as follows: ; Based on the calculation results of the above formula, the system establishes a baseline time interval across heterogeneous data streams. This baseline physical delay, at the underlying logic level, clarifies the basic physical propagation time required for secondary characteristic signals to respond to the device's abnormal event, thus providing a deterministic calculation baseline for subsequent dynamic adaptive correction of time drift by introducing environmental state variables.

[0057] See attached document Figure 5 The specific implementation of step S13 of the present invention, which involves extracting basic environmental parameters and generating a medium-state delay drift window, includes the following steps: S131, locking the trigger time of the abnormal event. After that, the system needs to obtain the equipment operating status that matches that moment.

[0058] Thanks to the previously established water level alignment mechanism, it occurs at the trigger time. The underlying environmental data streams have already resided in the memory sliding cache of the access buffer module 20. During this stage, the storage engine module 30 bypasses the conventional disk database query chain and directly uses the trigger time... For addressing key values, the instantaneous medium temperature value is extracted from the access buffer module 20 with constant time complexity.

[0059] In actual industrial operating environments, the sampling frequency of the basic environmental data stream is usually lower than that of the main feature data stream. To avoid data loss dead zones caused by incomplete overlap of sampling time points, the storage engine module 30 extracts the timestamp immediately adjacent to the trigger time from the sliding cache. The corresponding trigger time is calculated using a linear interpolation algorithm from the two actual sampling points before and after. instantaneous medium temperature value .

[0060] In this embodiment, if a sensor malfunction prevents the detection of two adjacent real sampling points within the sliding buffer, the system automatically calls the preset reference temperature from the configuration table. As an alternative, to ensure the continuity of the entire computing chain.

[0061] S132 addresses the objective situation that the physical properties of insulating media in industrial settings are susceptible to temperature fluctuations, and the system aims to correct the drift in physical signal transmission velocity caused by changes in environmental conditions.

[0062] From a physical perspective, the density and bulk modulus of media such as transformer insulating oil or sulfur hexafluoride gas change with the rise and fall of equipment operating temperature. This change in fluid properties causes the actual propagation speed of secondary characteristic signals to deviate from the static reference parameters.

[0063] To compensate for such physical mechanism errors, the storage engine module 30 introduces a preset temperature coefficient for the dielectric properties. Based on the instantaneous medium temperature value Compared with the reference temperature The difference, relative to the reference physical wave velocity Perform dynamic wave velocity correction. Actual propagated wave velocity. The calculation formula is: ; In this calculation formula, the temperature coefficient of the medium's physical properties Characterizes the relative drift rate of wave velocity under unit temperature change in a specific conductive medium.

[0064] As a preferred method, this physical property coefficient can be determined by offline temperature rise physical experiments on the same batch of insulation media in the substation, and then fixed as a constant in the system configuration table. Its value range is typically -1.5 × 10⁻⁶. -3~-3.5×10 -3 The specific value depends on the grade of the insulating oil.

[0065] S133, based on the dynamically corrected actual propagation wave velocity The storage engine module 30 utilizes three-dimensional transmission distance parameters. With actual propagation wave velocity The ratio is used to recalculate the actual physical delay of the signal under the current operating conditions of the device. ,Right now In industrial network environments, even with the deployment of precise time synchronization protocols, heterogeneous sensor networks will inevitably experience inherent time synchronization jitter due to the underlying hardware crystal oscillators and analog-to-digital conversion circuits of different sensors.

[0066] To ensure that secondary feature data can be completely recalled without missed detections during the retrieval phase, the system introduces a time tolerance term on top of the actual physical latency. The time tolerance term The specific value is determined by the upper limit of the comprehensive time synchronization error test of the heterogeneous sensor network in the field, and is usually taken as a few milliseconds to tens of milliseconds in conventional substation configurations. This is combined with the trigger time. Actual physical delay With time tolerance term The storage engine module 30 derives the medium-state delay drift window of the secondary feature data stream in response to the physical event. Its mathematical expression is shown below as the formula for an absolute time closed interval: ; Through this calculation, the system establishes the expected time boundary for the arrival of secondary feature data within a certain future period, thereby transforming the physical medium drift affected by operating conditions such as temperature into a deterministic numerical range. This medium-state time delay drift window forms the basis for subsequent time partitioning logical block mapping and precise data pruning.

[0067] See attached document Figure 6 The specific implementation of generating a logical floating pointer and performing asynchronous disk write in step S14 of the present invention includes the following steps: S141. When performing data persistence operations in a time-series database, in order to save disk space, the underlying layer usually uses a variable-length compression algorithm to process continuous data points.

[0068] Because data fluctuations vary across different time periods, compression ratios differ, resulting in dynamic changes in the physical byte offset of data within the underlying file system. If multi-source data association indexes are directly constructed using absolute physical addresses, the computational overhead for index maintenance often increases dramatically should the underlying data undergo background merging or reorganization. To mitigate the address uncertainty caused by underlying data compression at the index level, the system introduces the concept of time-partitioned logical blocks, independent of physical storage configurations.

[0069] Specifically, the storage engine module 30 logically slices the secondary feature data stream according to a fixed time span set by the underlying database. Let this time span be... For any given timestamp The specific formula for calculating the identifier of its logical block is as follows: ; in, This indicates the floor function; Integer number for the generated time-separated logical blocks. Time span. The specific value can be set according to the available memory page size of the database system and the actual data write throughput of the device.

[0070] As a preferred method, in high-frequency monitoring scenarios for power equipment, this time span It can be set to a constant of 1 to 10 seconds, which can achieve a balance between addressing precision and the total number of logic blocks.

[0071] S142 After constructing the mapping rules from the time dimension to the logical number, it is necessary to further discretize the previously obtained continuous time intervals.

[0072] The storage engine module 30 substitutes the upper and lower boundary timestamps of the media-state delay drift window, calculated previously, into the mapping formula. Through rounding, the system can obtain the set of target logical block intervals that the secondary feature data stream is expected to fall into when responding to the abnormal event. This set of target logical block intervals... The derivation formula is as follows: ; In this embodiment, if the upper and lower bound logical numbers calculated by the above formula are not equal, it indicates that the time drift window spans multiple adjacent logical blocks. In this case, the system will automatically fill in all consecutive integer number elements between the upper and lower bound numbers to ensure that the data coverage has no logical dead zones. Through this conversion mechanism, the system decouples the continuous absolute time interval from the strong binding relationship with the physical address and transforms it into a set of discrete integer numbers, so that the upper-level index points do not depend on the actual compression ratio of the lower-level data block.

[0073] S143, after obtaining the target logical block range set, the storage engine module 30 constructs a data structure containing physical causal relationships, namely a logical floating pointer, in memory. This logical floating pointer data structure encapsulates three core fields.

[0074] The target data stream identifier is used to precisely point to a specific secondary sensor data channel during subsequent retrieval. Target logical block interval set. This pointer serves as a coarse-grained addressing clue for the system to initiate data reads from the file system. Furthermore, it internally preserves the complete, unmapped, precise media-state delay drift window. It is used to provide fine-grained time clipping boundaries after data is loaded into memory and decompressed.

[0075] This composite structure design retains both broad addressing capabilities to cope with changes in the underlying blocks and maintains the microsecond-level time precision required for data alignment.

[0076] S144, during the write phase at the database level, the system employs a decoupled concurrent timing mechanism to handle the logical floating pointer and the original monitoring data.

[0077] The storage engine module 30 uses the encapsulated logical floating pointer as the data value in the key-value pair data system, binds it to the primary key of the previously generated event anchor point, and submits it to the asynchronous persistent queue of the inverted index file for disk write operation. At the same time, the internal data flow components of the system still process the routine slicing, compression, and disk write operations of the secondary feature data stream in parallel.

[0078] Since the logical roaming pointer only records the logical number of the time partition, its index write-to-disk action will not interfere with or block the variable-length compression and writing process of regular data blocks, thereby reducing the impact of index updates on the system throughput in a high-concurrency write environment of the database.

[0079] For the variable-length time-series data compression algorithm and the specific tree-like data structure layout of the inverted index used by the underlying database, those skilled in the art can choose the underlying framework of the mainstream time-series storage engine according to the actual engineering needs. The specific coding implementation of its file stream compression logic and asynchronous disk flushing scheduling is a well-known technology in this field and will not be elaborated here.

[0080] See attached document Figure 7 The specific implementation of the data retrieval and fusion cropping in step S15 of the present invention includes the following steps: S151, when the target power equipment triggers a downstream diagnostic task and needs to make a comprehensive judgment on the equipment abnormality, the application processing module 40 submits an abnormal event fusion retrieval instruction to the underlying storage engine module 30 through the database query interface.

[0081] The retrieval instruction typically carries a specified time span parameter or a specific event identifier. The storage engine module 30 receives and parses the retrieval instruction, and performs a traversal and matching in the in-memory inverted index tree according to the query conditions, thereby hitting the corresponding event anchor point in the underlying index file.

[0082] S152, after hitting the event anchor point, in order to extract the original signal related to the event, the system needs to convert the static index information into the physical access path of the underlying data.

[0083] The storage engine module 30 extracts and deserializes the logical roaming pointer bound to the anchor point from the index structure, thereby resolving the target data stream identifier and the target logical block range set. During the operation of the time-series database, the underlying data files continuously undergo merging and reorganization operations, causing dynamic changes in the physical storage location of the data.

[0084] To obtain the current actual data storage location, the storage engine module 30 extracts the integer number from the target logical block range set as an addressing clue to access the metadata management table maintained internally by the storage system. This query method based on logical numbers achieves a delayed binding between the logical view and the physical address, enabling the system to quickly obtain the actual physical sector address of the target compressed logical block on the disk from the metadata management table even after the data block has been merged and migrated in the background.

[0085] In this embodiment, if the corresponding logical block number cannot be found in the metadata management table due to data aging or obsolescence, the system determines that the historical data has expired and returns a data truncation flag to the upper layer to avoid the underlying process getting stuck in a deadlock.

[0086] S153, after establishing the physical sector address, the storage engine module 30 extracts the underlying raw data.

[0087] Since the logic blocks related to abnormal events are independent segments distributed in a hashed manner in the massive time-series data stream, in order to avoid loading a large number of unrelated adjacent time-series data blocks into memory and causing system bus congestion, the storage engine module 30 abandons the conventional continuous range scan addressing mechanism.

[0088] As a preferred approach, the engine employs a direct input / output mechanism, bypassing the operating system's page cache and using discrete addressing based on physical sector addresses to directly extract the target compressed logic block from the disk into the memory buffer. This direct read / write mechanism effectively prevents low-frequency abnormal retrieval operations from polluting the system's existing cache hit rate.

[0089] Subsequently, the storage engine module 30 calls the decompression algorithm that matches the underlying write operation to restore the compressed logical block into a continuous time-series data point array containing the original physical timestamps and monitoring values.

[0090] S154. Although the decompressed time-series data point array covers the physical response period of the abnormal event, since the logic blocks are divided according to a fixed coarse-grained time span, the two ends of the array usually contain a large number of redundant data points.

[0091] To provide the application layer with accurate causal correlation data, the storage engine module 30 extracts the undiscrete mapped medium-state delay drift window from the logical wandering pointer. The system uses the absolute time closed interval established by this window as the boundary to perform precise timeline pruning on the decompressed data in memory. Let the original set of decompressed time-series data points be... The effective data point set after cropping is Then, the mathematical expression of its fixed-point stripping logic is: ; In this formula, This represents the original absolute timestamp of each extracted data point; This represents the secondary feature monitoring value at the corresponding timestamp. After this precise pruning process, only the response data that conforms to the physical laws of medium conduction is retained in memory. If the pruned set of valid data points... If the value is empty, it indicates that no valid secondary feature response was detected within the time delay window, and the system will mark the result along with the primary feature data as an isolated trigger event.

[0092] Finally, the storage engine module 30 packages the filtered primary feature data and the pruned secondary feature data into a structured result set and returns it to the application processing module 40 for the system to perform more refined downstream diagnostic analysis.

[0093] For the specific structure maintenance of the metadata management table in the underlying database, and the design of the system-level call interface at the bottom layer of the direct input / output mechanism, those skilled in the art can use the standard file system bottom layer instructions of the operating system and the conventional table structure mapping mechanism to implement it. Its concurrency access control and pointer offset logic are well-known technologies in this field, and will not be elaborated here.

[0094] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for fusing and processing multi-source monitoring data of power equipment, characterized in that, Includes the following steps: Receive concurrent data streams and perform timeline alignment operations to separate the main feature data stream, secondary feature data stream, and basic environment data stream; The main feature data stream is monitored, and when an abnormal event is detected, an event anchor point is generated, and a reference physical delay is generated by combining the spatial conduction parameters of the power equipment. Based on the basic environmental data stream matched with the event anchor point, the reference physical delay is corrected for wave velocity, and a secondary feature data stream is generated to respond to the medium state delay drift window of the abnormal event. The medium-state delay drift window is mapped to a set of target logical block intervals, encapsulated to generate a logical floating pointer and bound to the event anchor point, and asynchronous persistence to disk is performed. The system receives a fusion retrieval command, obtains the logical floating pointer based on the event anchor point, extracts the underlying raw data based on the target logical block interval set, and performs time axis clipping on the decompressed underlying raw data using the medium-state delay drift window to obtain the fusion processing result.

2. The method for fusion processing of multi-source monitoring data of power equipment according to claim 1, characterized in that, The timeline alignment operation includes: Extract the absolute timestamp of each data packet in the concurrent data stream; The event water level is calculated and updated in real time based on the preset maximum tolerable delay time window and the current maximum event timestamp value. When the absolute timestamp of the data block to be processed is less than or equal to the current event water level, it is confirmed that the corresponding multi-source monitoring data message has been received completely, and the separation operation to the main feature data stream, secondary feature data stream and basic environmental data stream is performed.

3. The method for fusion processing of multi-source monitoring data of power equipment according to claim 1, characterized in that, The monitoring of the main feature data stream, and the generation of event anchors when an abnormal event is identified, includes: The energy integral of the instantaneous amplitude is performed on the main feature data stream using a sliding time window; When the calculated energy integral value is greater than or equal to the preset baseline threshold, the start timestamp of the corresponding data slice is marked as the trigger time. Generate the event anchor point in the inverted index tree, which includes the physical identifier of the sensor device, the trigger time, and the energy extreme value.

4. The method for fusion processing of multi-source monitoring data of power equipment according to claim 3, characterized in that, The process of generating a reference physical delay by combining the spatial conduction parameters of the power equipment includes: Call the system's preset internal conduction distance parameters of power equipment and reference physical wave velocity parameters; The reference physical delay is calculated by using the ratio of the transmission distance parameter to the reference physical wave velocity parameter.

5. The method for fusion processing of multi-source monitoring data of power equipment according to claim 4, characterized in that, The step of correcting the reference physical delay by wave velocity and generating a secondary characteristic data stream response medium-state delay drift window for the anomalous event includes: Using the trigger time as the address key, adjacent real sampling points are extracted and linearly interpolated to calculate the instantaneous medium temperature value; By introducing a medium physical property temperature coefficient, the reference physical wave velocity parameter is corrected based on the difference between the instantaneous medium temperature value and the preset reference temperature, and the actual transmission wave velocity is obtained. The actual physical time delay is calculated using the ratio of the transmission distance parameter to the actual transmission wave velocity; Based on the actual physical delay, a time tolerance term is introduced, and an absolute time closed interval is constructed in combination with the trigger time to generate the medium state delay drift window.

6. The method for fusion processing of multi-source monitoring data of power equipment according to claim 1, characterized in that, The step of mapping the medium-state delay drift window to a set of target logic block intervals and encapsulating it to generate a logic roaming pointer includes: Logically slice the secondary feature data stream according to a fixed time span; Substitute the upper and lower boundary timestamps of the medium-state delay drift window into the mapping formula and perform a floor operation to complete the continuous integer numbering, thereby generating a discrete set of the target logic block intervals; Construct in memory the logical roaming pointer containing the target data stream identifier, the target logical block interval set, and the unmapped medium-state delay drift window.

7. The method for fusion processing of multi-source monitoring data of power equipment according to claim 1, characterized in that, The asynchronous persistence to disk execution includes: The logical roaming pointer is used as a data value and key-value bound to the event anchor point, and submitted to the asynchronous persistent queue of the inverted index file for disk writing; Simultaneously, slicing, compression, and disk writing operations on the secondary feature data stream are performed in parallel.

8. The method for fusion processing of multi-source monitoring data of power equipment according to claim 1, characterized in that, The extraction of underlying raw data based on the target logical block interval set includes: The target logical block interval set is parsed from the logical traversing pointer, and the target data stream identifier is extracted; Extract the integer number from the target logical block range set as an addressing clue, and access the metadata management table to obtain the physical sector address of the disk; The target compressed logic block is extracted using a direct input / output mechanism based on the physical sector address of the disk, and then decompressed into a continuous time-series data point array containing the original physical timestamp and monitoring values.

9. A method for fusion processing of multi-source monitoring data of power equipment according to claim 8, characterized in that, The step of performing time-axis clipping on the decompressed underlying raw data using the medium-state delay drift window includes: Using the absolute time closed interval determined by the medium-state delay drift window as the boundary, fixed-point stripping is performed on the continuous time-series data point array to remove data points outside the absolute time closed interval; If the set of valid data points after cropping is empty, mark the result and the main feature data as isolated trigger events. If not empty, the result of the fusion process is packaged into a structured form and returned.

10. A power equipment multi-source monitoring data fusion processing system, used to implement the power equipment multi-source monitoring data fusion processing method as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to collect status data and signal data of the internal physical medium of the target power equipment, and generate and send concurrent data streams; The access buffer module is communicatively connected to the data acquisition module and is used to receive the concurrent data stream and perform time axis alignment operation to separate the main feature data stream, secondary feature data stream and basic environment data stream. The storage engine module, connected to the access buffer module, is used to monitor the main feature data stream generation event anchor point and the baseline physical latency, correct the generated medium-state latency drift window according to the basic environment data stream, encapsulate and generate a logical floating pointer and perform asynchronous persistence to disk; In response to the fusion retrieval command, the underlying raw data is extracted based on the target logical block interval set and time axis clipping is performed using the medium-state delay drift window; The application processing module is communicatively connected to the storage engine module, and is used to submit the fusion retrieval instruction to the storage engine module and receive the fusion processing result.

Images