A direct current power supply monitoring system

Abstract

The application provides a direct-current power supply monitoring system, comprising: obtaining operation data from a direct-current power supply device to obtain a real-time collected operation data stream; pre-processing the operation data stream through an edge computing node to generate a uniformly encoded operation data stream; converging the uniformly encoded operation data stream to a central platform by using a distributed message queue mechanism to obtain a multi-source data set; performing data cleaning and format unification on the multi-source data set to generate a globally consistent data set; storing the globally consistent data set in a time series database to obtain a historical trend data set that can be quickly queried; analyzing the historical trend data set by using a neural network model to extract a business cycle feature vector; predicting a global power supply capacity according to the business cycle feature vector to obtain a capacity prediction result; identifying a bottleneck area in the capacity prediction result by using an anomaly detection algorithm to generate a global capacity state display.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a DC power supply monitoring system. Background Technology

[0002] Background of the problem: DC power supply monitoring systems are crucial in modern data centers, communication base stations, and industrial automation, as their stable operation directly impacts equipment safety and business continuity. With increasing informatization, the number of power supply devices has surged, significantly increasing management complexity and necessitating efficient monitoring methods to ensure power supply reliability. However, many current monitoring systems employ decentralized management, limiting data processing capabilities and hindering global optimization. Data from independently operating devices is difficult to integrate, preventing managers from timely understanding of overall power capacity status and restricting their ability to predict future demand.

[0003] Against this backdrop, centralized monitoring architecture has become a key research focus, but its implementation faces significant technical challenges. Distributed DC power supply equipment operating data needs to be efficiently aggregated to a central platform, requiring the system to possess high-performance data acquisition and transmission capabilities. However, after data aggregation, how to uniformly process multi-source, heterogeneous load data to form a global power capacity view becomes a deeper challenge. Inconsistencies in data processing can lead to distorted analysis results, thus affecting the accuracy of capacity prediction. Furthermore, the implementation of complex prediction algorithms relies on in-depth analysis of historical trends and business cycles; if the data processing engine cannot efficiently integrate multi-dimensional information, the prediction model will struggle to capture potential capacity bottlenecks.

[0004] Therefore, designing a centralized monitoring system that can efficiently aggregate and process operational data from distributed devices, and achieve accurate capacity prediction through a high-performance data processing engine, has become a key issue in the field of DC power supply monitoring. Summary of the Invention

[0005] This invention provides a DC power supply monitoring system, mainly comprising: Operational data is acquired from DC power supply equipment to obtain a real-time operational data stream. This data stream is preprocessed using edge computing nodes to generate a uniformly encoded operational data stream. A distributed message queue mechanism is used to aggregate the uniformly encoded operational data stream to a central platform, resulting in a multi-source data set. This multi-source data set is then cleaned and formatted to generate a globally consistent dataset. The globally consistent dataset is stored in a time-series database to obtain a quickly queried historical trend dataset. A neural network model is used to analyze the historical trend dataset and extract business cycle feature vectors. Global power capacity is predicted based on these business cycle feature vectors to obtain a capacity prediction result. An anomaly detection algorithm identifies bottleneck areas in the capacity prediction result, generating a global capacity status display.

[0006] Furthermore, the step of acquiring operational data from the DC power supply device to obtain a real-time collected operational data stream includes: acquiring voltage, current, and power parameters from the DC power supply device through a high-frequency sampling protocol to obtain real-time collected operational data; classifying the parameters in the operational data using a k-means clustering algorithm to identify abnormal data points; if an abnormal data point is detected, determining the abnormality type through a preset threshold to obtain an abnormality classification result; based on the abnormality classification result, using a support vector machine algorithm to predict the operational data to generate predicted data of the device's operational status; and comparing the predicted data with the real-time collected operational data stream to determine the deviation of the device's operational status and obtain a status deviation analysis result.

[0007] Furthermore, the step of preprocessing the running data stream using edge computing nodes to generate a uniformly encoded running data stream includes: preprocessing the running data stream using edge computing nodes; if the data volume of the running data stream exceeds a preset threshold, performing data compression to obtain compressed data packets; performing format standardization operations based on the compressed data packets to generate a uniformly encoded running data stream; and analyzing the changing trends of voltage, current, and power parameters through the uniformly encoded running data stream to determine the parameter change characteristics.

[0008] Furthermore, the step of using a distributed message queue mechanism to aggregate the uniformly encoded runtime data stream to a central platform to obtain a multi-source data set includes: deploying a high-throughput data bus on the central platform to acquire the uniformly encoded runtime data stream; dynamically adjusting the queue priority if the data transmission delay of the runtime data stream exceeds a preset threshold to obtain a priority-adjusted data stream; aggregating the priority-adjusted data stream through the distributed message queue mechanism to generate a multi-source data set; and extracting metadata information from each data source based on the multi-source data set to determine data source characteristics.

[0009] Furthermore, the step of cleaning and format unifying the multi-source data set to generate a globally consistent dataset includes: extracting metadata information from the multi-source data set through data source identifiers to obtain a preliminary dataset; normalizing the fields of the preliminary dataset using a data cleaning algorithm to generate a standardized dataset; if missing values ​​exist in the standardized dataset, imputing them using an interpolation algorithm to obtain a completed dataset; unifying the device data format of the completed dataset using a pre-established metadata mapping table to generate a format-unified dataset; and using a consistency verification algorithm to detect the completeness and consistency of the format-unified dataset to obtain a globally consistent dataset.

[0010] Furthermore, the step of storing the globally consistent dataset in a time-series database to obtain a quickly queryable historical trend dataset includes: storing multi-dimensional data of the globally consistent dataset in a time-series database to determine the historical trend dataset; using a partitioned index optimization mechanism to partition and store voltage, current, and power parameters according to time identifiers and device identifiers to obtain partitioned datasets; extracting parameter change trends from the partitioned datasets using a time-series analysis algorithm to obtain parameter trend features; using a clustering algorithm to group the parameter trend features into device groups to determine device status groups; if the parameters in the device status groups exceed a preset threshold, identifying abnormal device identifiers using an anomaly detection algorithm to obtain an abnormal device list.

[0011] Furthermore, the step of using a neural network model to analyze the historical trend dataset and extract business cycle feature vectors includes: using a long short-term memory neural network model to analyze the multi-dimensional data of the historical trend dataset and extract business cycle features; if the prediction error of the neural network model exceeds a preset threshold, adjusting the model parameters through online learning to obtain an optimized model; generating a business cycle feature vector based on the optimized model; and determining the periodic change pattern of the equipment operating status through the business cycle feature vector.

[0012] Furthermore, the step of predicting the global power capacity based on the service cycle feature vector to obtain the capacity prediction result includes: analyzing the service cycle feature vector using a random forest algorithm to predict the global power capacity; integrating the historical trend dataset and real-time running data stream through a weighted feature fusion mechanism to generate fused feature data; obtaining the capacity prediction result based on the fused feature data; if the deviation between the capacity prediction result and the actual capacity exceeds a preset threshold, recalculating the feature weights to obtain an optimized capacity prediction result; and determining the distribution characteristics of the global power capacity based on the optimized capacity prediction result.

[0013] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a method for predicting the capacity of distributed DC power supply equipment. It acquires operational data through high-frequency sampling, achieves efficient data aggregation using edge computing and distributed message queues, normalizes and completes heterogeneous data, stores historical trend data using a time-series database, and utilizes a long short-term memory neural network and a random forest algorithm for capacity prediction. This invention also identifies potential capacity bottlenecks through anomaly detection and generates a visualized global capacity status display. This method achieves high-precision prediction of the capacity of distributed DC power supply equipment, can promptly identify potential bottlenecks, provides strong support for power management decisions, and effectively improves the operating efficiency and reliability of distributed DC power supply systems. Attached Figure Description

[0014] Figure 1 This is a flowchart of a DC power supply monitoring system according to the present invention.

[0015] Figure 2 This is a schematic diagram of a DC power supply monitoring system according to the present invention.

[0016] Figure 3 This is another schematic diagram of a DC power supply monitoring system according to the present invention.

[0017] Figure 4 This is another schematic diagram of a DC power supply monitoring system according to the present invention.

[0018] Figure 5 This is another schematic diagram of a DC power supply monitoring system according to the present invention.

[0019] Figure 6 This is another schematic diagram of a DC power supply monitoring system according to the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0021] like Figure 1-6 This embodiment of a DC power supply monitoring system may specifically include: Step S101: Obtain operating data from distributed DC power supply equipment, use a high-frequency sampling protocol to collect voltage, current and power parameters in real time, and use edge computing nodes to perform preliminary compression and format standardization on the collected data to obtain a uniformly encoded operating data stream.

[0022] Operational data is acquired from distributed DC power supply equipment. A high-frequency sampling protocol is used to collect voltage, current, and power parameters in real time, resulting in a raw operational dataset. This raw dataset is preprocessed using edge computing nodes. If the data volume exceeds a preset threshold, data compression is performed to obtain compressed data packets. Based on the compressed data packets, format standardization is performed to generate a uniformly encoded operational data stream. Time series analysis algorithms are used to extract the changing trends of voltage, current, and power parameters from the operational data stream, yielding parameter trend characteristics. Based on these trend characteristics, a k-means clustering algorithm is used to classify voltage, current, and power parameters, identifying outlier data points. If an outlier is detected, a preset threshold is used to determine the anomaly type, resulting in an anomaly classification result. Based on the anomaly classification result, a support vector machine algorithm is used to predict the operational data stream, generating predicted data for the equipment's operating status. By comparing the predicted data with the real-time acquired operational data stream, a deviation calculation algorithm is used to analyze the deviation in the equipment's operating status, yielding a status deviation analysis result. Based on the status deviation analysis result, visualization analysis technology is used to generate an equipment status view. Heatmaps and trend curves are used to dynamically render the deviation distribution, providing an intuitive display of the equipment's status.

[0023] For example, operational data is acquired from distributed DC power supply equipment. A high-frequency sampling protocol is used to collect voltage, current, and power parameters at a frequency of 1000 times per second, resulting in a raw operational dataset containing timestamps. The raw operational dataset is preprocessed using edge computing nodes. If the data volume of a single acquisition exceeds 1MB, the LZ77 algorithm is used to perform data compression, resulting in a data packet with a compression rate of 60%. Based on the compressed data packet, JSON format standardization is performed, unifying field naming and unit conversion, generating an operational data stream conforming to the IEEE 1888 standard. Using the operational data stream, the ARIMA time series analysis algorithm is used to extract 12-dimensional parameter trend features, such as voltage fluctuation amplitude and current rise slope, over the past 5 minutes. Based on the parameter trend features, a k-means clustering algorithm is used to set cluster centers with k=3, and data points with an Euclidean distance less than 0.5 are considered outliers. If an outlier is detected, it is compared to preset thresholds of voltage exceeding ±10% and current exceeding ±15% to determine if it is an overvoltage or overcurrent anomaly. Based on the anomaly classification results, a support vector machine algorithm with the RBF kernel function was used, with 1000 historical normal data sets as the training set, to predict the device's operating status data within the next 30 seconds. By comparing the predicted data with the real-time acquired data stream, a deviation exceeding 0.2 was considered a significant status deviation using the root mean square error algorithm. Based on the status deviation analysis results, a heatmap was generated using the D3.js visualization library, with red, yellow, and green gradients used to render the deviation intensity, and a line chart overlaid to display the parameter change trend.

[0024] Step S102: For the unified encoded running data stream, a distributed message queue mechanism is used to achieve efficient data aggregation. A high-throughput data bus is deployed on the central platform. If the data transmission delay exceeds a preset threshold, the queue priority is dynamically adjusted to obtain a multi-source data set aggregated to the central platform.

[0025] Voltage, current, and power parameters are acquired from DC power supply equipment using a high-frequency sampling protocol to obtain real-time operational data. Edge computing nodes are used to preprocess the operational data; if the data volume exceeds a preset threshold, data compression is performed to obtain compressed data packets. Based on the compressed data packets, format standardization is performed to generate a uniformly encoded operational data stream. The uniformly encoded operational data stream is distributed through a distributed message queue mechanism; if the queue load exceeds a preset threshold, queue priority is dynamically adjusted to obtain the data stream distributed to the central platform. A high-throughput data bus is deployed on the central platform to aggregate the data streams transmitted by the distributed message queue, resulting in a multi-source data set aggregated to the central platform. Based on the multi-source data set aggregated to the central platform, heterogeneous data is normalized using a data cleaning algorithm to obtain a pre-cleaned data set. A pre-established metadata mapping table is used to unify the format of the pre-cleaned data set; if data is missing, it is filled in using an interpolation algorithm to obtain a consistent global dataset. The operational data stream is used to analyze the changing trends of voltage, current, and power parameters in the global dataset, and a k-means clustering algorithm is used to classify the parameters and identify outlier data points. Based on abnormal data points, the abnormality type is determined by a preset threshold, and the abnormality classification result is obtained.

[0026] For example, voltage, current, and power parameters are acquired from a DC power supply device at a frequency of 1000 times per second using a high-frequency sampling protocol to obtain real-time operational data. Edge computing nodes are used to preprocess the operational data. If the data volume exceeds a preset threshold of 1MB per second, data compression based on the LZ77 algorithm is performed to obtain compressed data packets. Based on the compressed data packets, a format standardization operation is performed, converting the data to JSON format and adding a unified timestamp and device identifier to generate a uniformly encoded operational data stream. The uniformly encoded operational data stream is distributed through a distributed message queue mechanism. If the queue load exceeds a preset threshold of 80%, the queue priority is dynamically adjusted, prioritizing the transmission of high-priority data streams, resulting in a data stream distributed to the central platform. A high-throughput data bus is deployed on the central platform to aggregate the data stream transmitted by the distributed message queue at a throughput of 10GB per second, resulting in a multi-source data set aggregated to the central platform. Based on the multi-source data set aggregated to the central platform, a data cleaning algorithm based on regular expressions is used to normalize the heterogeneous data, removing invalid characters and duplicate data, resulting in a pre-cleaned data set. A pre-established metadata mapping table is used to standardize the format of the initially cleaned dataset. If missing data is detected, linear interpolation is used to fill in the missing values, resulting in a consistent global dataset. Data stream analysis is performed on the voltage, current, and power parameters in the global dataset to analyze their changing trends. A k-means clustering algorithm is used to classify the parameters, with a cluster size of 3, to identify outlier data points. Based on these outlier data points, a preset threshold is used to determine the anomaly type. If the voltage fluctuation exceeds ±5%, it is marked as a voltage anomaly, yielding the anomaly classification results.

[0027] Step S103: Based on the multi-source data set aggregated to the central platform, the heterogeneous data is normalized using a data cleaning algorithm. A pre-established metadata mapping table is used to unify the data formats of different devices. If data is missing, it is filled in using an interpolation algorithm to obtain a consistent global dataset.

[0028] Heterogeneous datasets are acquired from a multi-source data platform. Metadata information for each dataset is extracted using data source identifiers to obtain a preliminary dataset. Based on the preliminary dataset, data fields are normalized using data cleaning algorithms to generate a standardized dataset. If missing values ​​are detected in the standardized dataset, interpolation algorithms are used to fill in the missing values, resulting in a completed dataset. The device data format of the completed dataset is standardized using a pre-established metadata mapping table, generating a format-consistent dataset. For the format-consistent dataset, a consistency verification algorithm is used to check data integrity and consistency, resulting in a verification dataset. Based on the verification dataset, a globally consistent dataset is generated and stored in the central data platform, resulting in a stored dataset. For the stored dataset, a time-series database is used to store multi-dimensional historical trend data. An index optimization mechanism is used to partition and store parameters such as voltage, current, and power according to time and device identifier, resulting in a historical trend dataset. If a query request exists in the historical trend dataset, the corresponding parameter data is extracted according to time and device identifier using an index optimization mechanism, resulting in a fast query dataset. Based on the fast query dataset, a multi-dimensional data analysis report is generated and stored in the central data platform, resulting in an analysis report dataset.

[0029] For example, heterogeneous datasets are acquired from a multi-source data platform. Metadata information for each dataset is extracted using data source identifiers; for instance, temperature data is extracted from sensor A, and humidity data from sensor B, resulting in a preliminary dataset containing fields such as temperature and humidity. Based on the preliminary dataset, data cleaning algorithms are used to normalize the data fields, for example, unifying the temperature unit to degrees Celsius and the humidity unit to percentage, generating a standardized dataset. If missing values ​​are detected in the standardized dataset, interpolation algorithms are used to fill in the missing values; for example, linear interpolation is used to fill in the missing temperature data, resulting in a completed dataset. Using a pre-established metadata mapping table, the device data format of the completed dataset is unified; for example, the timestamp format of sensor A is converted from Unix timestamps to ISO 8601 format, generating a format-unified dataset. For the format-unified dataset, a consistency check algorithm is used to check data integrity and consistency; for example, checking whether the temperature data is within the range of -50℃ to 50℃, resulting in a verification dataset. Based on the verification dataset, a globally consistent dataset is generated and stored in the central data platform, resulting in a stored dataset. For the stored dataset, a time-series database is used to store multi-dimensional historical trend data. An index optimization mechanism is used to partition and store parameters such as voltage, current, and power by time and device identifier. For example, voltage data for October 2023 is partitioned and stored by device ID, resulting in a historical trend dataset. If a query request exists in the historical trend dataset, the corresponding parameter data is extracted by time and device identifier using the index optimization mechanism. For example, voltage data for October 1, 2023, for device ID 001 is extracted, resulting in a fast query dataset. Based on the fast query dataset, a multi-dimensional data analysis report is generated, such as a voltage fluctuation trend chart for device 001, and stored on the central data platform, resulting in an analysis report dataset.

[0030] Step S104: For the global dataset after consistency processing, a time-series database is used to store multi-dimensional historical trend data. Through an index optimization mechanism, parameters such as voltage, current, and power are partitioned and stored according to time and device identifier to obtain a historical trend dataset that can be quickly queried.

[0031] Data cleaning algorithms are used to normalize the multi-source data set aggregated to the central platform, and a metadata mapping table is used to unify the heterogeneous data formats, resulting in a consistent global dataset. A time-series database is used to store multi-dimensional historical trend data from this consistent global dataset, identifying queryable historical trend datasets. An index optimization mechanism is used to partition the historical trend dataset by time and device identifier, resulting in quickly queryable partitioned datasets. If the time and device identifiers of a partitioned dataset are complete, time-series analysis algorithms are used to extract the changing trends of voltage, current, and power parameters, obtaining parameter trend features. Based on these trend features, a clustering algorithm is used to group device identifiers, determining the similarity of device operating states and identifying device state groups. If the voltage or current parameters in a device state group exceed a preset threshold, an anomaly detection algorithm is used to identify abnormal device identifiers, resulting in a list of abnormal devices. The corresponding time identifiers and parameter trend features are obtained from the list of abnormal devices, and historical trend data is queried from the time-series database to determine the time window in which the anomaly occurred. Using the historical trend data within the time window, a regression analysis algorithm is used to predict the future changing trends of voltage and current parameters, yielding the predicted trend results. Based on the predicted trend results, the trend characteristics of voltage and current parameters in the partitioned dataset are updated using a time series database to determine the optimized historical trend dataset.

[0032] For example, a data cleaning algorithm is used to normalize the multi-source data set aggregated to the central platform. A metadata mapping table is used to convert current data from different manufacturers' devices into standard units (e.g., amperes). Missing data is filled in using linear interpolation, resulting in a consistent global dataset. A time-series database (e.g., InfluxDB) is used to store historical voltage, current, and power data from the global dataset. Data is sampled and compressed at a 1-second granularity to determine a queryable historical trend dataset. A composite index is built on the historical trend data by time range (e.g., partitioned by day) and device ID (e.g., device type prefix) using an index optimization mechanism. Voltage parameters are compressed using the ZSTD algorithm, and current parameters are compressed using Gorilla encoding, resulting in a quickly queryable partitioned dataset. If the partitioned dataset contains complete device IDs and timestamps, the seasonality and trend components of the voltage parameters are extracted using the STL time series decomposition algorithm, and the root mean square value of the current parameters is calculated using a sliding window to obtain the parameter trend characteristics. Based on characteristics such as voltage fluctuation rate and current harmonic distortion rate, the DBSCAN clustering algorithm (eps=0.5, min_samples=3) is used to group equipment, determine the similarity of operating states of similar equipment, and identify equipment status groups. If the voltage parameter in a group exceeds the rated value ±10% or the current parameter continuously exceeds the safety threshold for 5 minutes, the Isolation Forest anomaly detection algorithm (contamination=0.01) is used to identify abnormal equipment and obtain a list of abnormal equipment. The timestamps and trend features of the most recent 24 hours are extracted from the list of abnormal equipment. The original sampled data for the corresponding time period is queried from a time-series database (resolution improved to 100ms) to determine the precise time window of the anomaly (e.g., 14:05:23-14:07:41). Using historical data within the time window, the voltage drop trend for the next 30 minutes is predicted using an ARIMA model (p=2, d=1, q=1), and the fluctuation range of the current parameter is predicted using an LSTM neural network to obtain the predicted trend results. Based on the voltage recovery time point and current peak value in the prediction results, the trend feature labels of the partitioned dataset are corrected using the batch update interface of the time series database, and the optimized historical trend dataset is determined.

[0033] Step S105: Based on the historical trend dataset, a long short-term memory neural network model is used to analyze the multi-dimensional historical trends. Business cycle features are extracted through a pre-trained model. If the prediction error exceeds a preset threshold, the model parameters are adjusted through online learning to obtain the business cycle feature vector.

[0034] Step S106: For the business cycle feature vector, the random forest algorithm is used to predict the global power capacity. Historical trends and real-time data are integrated through a weighted feature fusion mechanism. If the deviation between the prediction result and the actual capacity exceeds a preset threshold, the feature weights are recalculated to obtain a high-precision capacity prediction result.

[0035] Step S107: Based on the high-precision capacity prediction results, potential capacity bottlenecks are identified through anomaly detection algorithms. Density-based clustering is used to perform partition analysis on the prediction data. If anomalies are detected that exceed a preset range, they are marked as bottleneck areas, and the bottleneck identification results are obtained.

[0036] Step S108: Based on the bottleneck identification results, a global capacity view is generated using visualization analysis technology. The capacity distribution and bottleneck areas are dynamically rendered using heatmaps and trend curves to obtain an intuitive display of the global capacity status.

[0037] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A DC power supply monitoring system, characterized in that, include: The system acquires operational data from DC power supply equipment to obtain a real-time operational data stream. The running data stream is preprocessed by edge computing nodes to generate a uniformly encoded running data stream; A distributed message queue mechanism is used to aggregate the uniformly encoded runtime data stream to the central platform, resulting in a multi-source data set; The multi-source data set is cleaned and formatted to generate a globally consistent dataset; By storing the globally consistent dataset in a time-series database, a historical trend dataset that can be quickly queried can be obtained. The historical trend dataset is analyzed using a neural network model to extract business cycle feature vectors. The global power capacity is predicted based on the service cycle feature vector to obtain the capacity prediction result; The bottleneck areas in the capacity prediction results are identified by an anomaly detection algorithm, and a global capacity status display is generated.

2. The method as described in claim 1, characterized in that, The process of acquiring operational data from the DC power supply device to obtain a real-time operational data stream includes: Voltage, current, and power parameters are obtained from DC power supply equipment through a high-frequency sampling protocol to obtain real-time operational data; The k-means clustering algorithm is used to classify the parameters in the running data to identify outlier data points; If an abnormal data point is detected, the anomaly type is determined by a preset threshold to obtain an anomaly classification result; Based on the anomaly classification results, the support vector machine algorithm is used to predict the operating data and generate predicted data of the device operating status. By comparing the predicted data with the real-time collected operating data stream, the deviation of the equipment's operating status is determined, and the status deviation analysis result is obtained.

3. The method as described in claim 1, characterized in that, The step of preprocessing the runtime data stream through edge computing nodes to generate a uniformly encoded runtime data stream includes: The running data stream is preprocessed using edge computing nodes; If the amount of data in the running data stream exceeds a preset threshold, data compression is performed to obtain compressed data packets; Based on the compressed data packet, a format standardization operation is performed to generate a uniformly encoded runtime data stream; By analyzing the changing trends of voltage, current, and power parameters through the unified encoded operating data stream, the characteristics of parameter changes can be determined.

4. The method as described in claim 1, characterized in that, The method employs a distributed message queue mechanism to aggregate the uniformly encoded runtime data stream to a central platform, resulting in a multi-source data set, including: A high-throughput data bus is deployed on the central platform to acquire the unified encoded runtime data stream; If the data transmission delay of the running data stream exceeds a preset threshold, the queue priority is dynamically adjusted to obtain a data stream with adjusted priority. The priority-adjusted data streams are aggregated through the distributed message queue mechanism to generate a multi-source data set; Based on the multi-source data set, metadata information of each data source is extracted to determine the data source characteristics.

5. The method as described in claim 1, characterized in that, The step of cleaning and format unifying the multi-source data set to generate a globally consistent dataset includes: Metadata information of the multi-source data set is extracted by identifying the data source to obtain a preliminary dataset; The fields of the preliminary dataset are normalized using a data cleaning algorithm to generate a standardized dataset; If there are missing values ​​in the standardized dataset, they are filled in using an interpolation algorithm to obtain a complete dataset; The device data format of the completed dataset is unified by a pre-established metadata mapping table, generating a unified dataset. A consistency check algorithm is used to detect the integrity and consistency of the format-unified dataset, resulting in a globally consistent dataset.

6. The method as described in claim 1, characterized in that, The process of storing the globally consistent dataset in a time-series database to obtain a historical trend dataset that can be quickly queried includes: By storing multi-dimensional data of the globally consistent dataset in a time-series database, historical trend datasets are determined. A partitioned indexing optimization mechanism is adopted to partition and store voltage, current and power parameters according to time and device identifiers to obtain partitioned datasets. The parameter change trends of the partitioned dataset are extracted using time series analysis algorithms to obtain parameter trend characteristics; Clustering algorithms are used to group the device status groups based on the trend characteristics of the parameters. If the parameters in the device status group exceed the preset threshold, the abnormal device identifier is identified by the anomaly detection algorithm to obtain the abnormal device list.

7. The method as described in claim 1, characterized in that, The step of using a neural network model to analyze the historical trend dataset and extract business cycle feature vectors includes: A long short-term memory neural network model is used to analyze the multi-dimensional data of the historical trend dataset and extract business cycle characteristics; If the prediction error of the neural network model exceeds a preset threshold, the model parameters are adjusted through online learning to obtain an optimized model. Based on the optimized model, generate a business cycle feature vector; The periodic change pattern of equipment operating status is determined by the business cycle feature vector.

8. The method as described in claim 1, characterized in that, The step of predicting the global power capacity based on the service cycle feature vector to obtain the capacity prediction result includes: The random forest algorithm is used to analyze the feature vector of the service cycle to predict the global power capacity; The historical trend dataset and real-time running data stream are integrated through a weighted feature fusion mechanism to generate fused feature data; Based on the fused feature data, the capacity prediction result is obtained; If the deviation between the capacity prediction result and the actual capacity exceeds a preset threshold, the feature weights are recalculated to obtain the optimized capacity prediction result. The distribution characteristics of global power capacity are determined based on the optimized capacity prediction results.

Images