Station-end AC / DC integrated power supply data management and control platform

Abstract

The invention discloses a station end AC / DC integrated power supply data management and control platform. The platform comprises a data acquisition module, an edge preprocessing module, a distributed storage module, an intelligent analysis module, a life prediction module, an alarm and visualization module and a model monitoring and iteration module. The platform obtains key data such as voltage and current through the data acquisition module, improves data quality through the edge preprocessing module, realizes efficient storage and integration by means of the distributed storage module, accurately detects abnormity and locates faults by means of the intelligent analysis module, deduces residual life through the life prediction module, and formulates a maintenance plan. Intelligent alarm and visual display are realized by combining an alarm and visualization module, and performance is optimized through a model monitoring and iteration module. The system operation reliability is remarkably improved, the operation and maintenance cost is reduced, support is provided for intelligent operation and maintenance of a transformer substation, and digital and intelligent development of an electric power system is promoted.

Description

Technical Field

[0001] This invention relates to the field of power system automation and data management technology, and in particular to a station-end AC / DC integrated power data management platform. Background Technology

[0002] AC / DC power supply systems are a core component of substation secondary equipment, bearing the crucial responsibility of providing stable power and backup power for protection devices, monitoring and control equipment, and other systems. Currently, data display for AC / DC power supply systems relies on the substation integrated automation system. However, this third-party platform suffers from issues such as incomplete data display, poor real-time performance, low level of intelligence, and unprofessional display of core parameters.

[0003] Specifically, the shortcomings of existing technologies are mainly reflected in the following aspects: First, the data acquisition dimensions are limited, and it is impossible to cover high-frequency monitoring of key parameters such as voltage, current, temperature, and insulation resistance; Second, there is a lack of professional edge preprocessing capabilities, and noise and outliers mixed in the raw data directly affect the accuracy of analysis; Third, the data storage architecture is not optimized for power time-series data, resulting in high-frequency data writing delays and low efficiency in historical data query; Fourth, fault detection relies on traditional threshold methods, which cannot identify gradual faults and complex associated faults, and lacks predictive maintenance capabilities; Fifth, operation and maintenance decisions lack data closed-loop support, and model iteration lags behind the aging trend of equipment.

[0004] The aforementioned problems result in low operation and maintenance efficiency of AC / DC power supply systems in substations, delayed fault early warning, and an inability to meet national requirements for the digitalization and intelligentization of power equipment. Therefore, there is an urgent need for a professional management and control platform that integrates data acquisition, intelligent analysis, lifespan prediction, and operation and maintenance decision-making. Summary of the Invention

[0005] In view of the above situation and to overcome the defects of the prior art, the purpose of this invention is to provide a station-end AC / DC integrated power data management and control platform.

[0006] The technical solution is: a station-side integrated AC / DC power data management and control platform, including: The data acquisition module is used to collect key data of the AC / DC power supply system, including voltage, current, temperature, and insulation resistance. An edge preprocessing module, connected to the data acquisition module, is used to preprocess the acquired key data; A distributed storage module, connected to the edge preprocessing module, is used to store preprocessed data according to data type and to build a data lake to achieve multi-source data integration and hierarchical management; The intelligent analysis module, connected to the distributed storage module, is used to perform intelligent analysis on the data lake to realize equipment anomaly detection and fault root cause analysis. The lifespan prediction module is connected to the intelligent analysis module and is used to generate equipment remaining lifespan prediction results and maintenance plans based on the intelligent analysis results. The alarm and visualization module is connected to the life prediction module and is used to generate operation and maintenance reports based on prediction results and maintenance plans through intelligent alarm mechanisms and visualization technology. The model monitoring and iteration module is connected to the intelligent analysis module, the lifetime prediction module, and the alarm and visualization module, respectively. It is used to monitor the model performance in the intelligent analysis, prediction, and decision-making process, and to achieve model iterative optimization through data closed-loop feedback.

[0007] Preferably, the preprocessing operations performed by the edge preprocessing module include: A sliding window mechanism is used to remove outliers from the collected data; wavelet transform is used to reduce noise in the collected data containing high-frequency noise; and the Min-Max method is used to normalize the collected data to eliminate dimensional differences.

[0008] Preferably, the intelligent analysis module includes a gradual fault detection unit, which is configured as follows: The time-series data of voltage and current are input into the LSTM neural network to calculate the error between the real-time predicted value and the actual value. When the error exceeds the threshold set based on the error distribution of historical normal data, it is determined to be a gradual fault.

[0009] Preferably, the intelligent analysis module further includes a current waveform distortion recognition unit, which is configured as follows: 1DCNN is used to process current waveform data. Features are extracted through multi-layer convolution operations. Each convolution layer performs convolution operations with the kernel weights and features of the previous layer and adjusts the bias. The output features are then processed by the ReLU activation function to generate output features. When the matching degree between the output features and the preset harmonic and interharmonic feature library reaches 90% or more, it is determined to be current waveform distortion.

[0010] Preferably, the intelligent analysis module further includes an insulation resistance anomaly marking unit, which is configured as follows: The insulation resistance data is processed by an autoencoder, and the second norm error between the original data and the data reconstructed by the encoder and decoder is calculated. When the error exceeds the reconstruction error threshold calculated based on historical normal insulation resistance data, it is marked as an anomaly.

[0011] Preferably, the intelligent analysis module further includes a fault root cause analysis unit, which is configured as follows: The XGBoost multi-classification model is adopted, with temperature and charge / discharge cycles as input features, and output fault types. The objective function of the model includes a cross-entropy loss term and a regularization term. The regularization term consists of the number of tree nodes, leaf node weights, and regularization coefficients. Fault types include rectifier module breakdown and battery sulfation.

[0012] Preferably, the lifetime prediction module includes a remaining lifetime prediction unit, which is configured as follows: The Cox proportional hazards model is adopted, with operating time and ambient temperature as covariates. The equipment risk rate is calculated by multiplying the baseline risk rate by the exponential function of the covariates, and the remaining life is calculated based on the trend of risk rate changes.

[0013] Preferably, the alarm and visualization module includes an intelligent alarm unit, which is configured as follows: A Gaussian mixture model is used to process the real-time monitoring data. The model is composed of multiple normally distributed components weighted together, with the sum of the weights of each component being 1. When the probability density value of the real-time monitoring data in the model is lower than the threshold set based on historical alarm data, an alarm is triggered. The real-time monitoring data includes voltage, current, and temperature.

[0014] Preferably, the model monitoring and iteration module includes a performance monitoring unit, which is configured as follows: The false positive rate and false negative rate of the model are calculated using a confusion matrix. The false positive rate is the ratio of the number of false positive data to the sum of the number of false positive and true negative data, and the false negative rate is the ratio of the number of false negative data to the sum of the number of true positive and false negative data. Among them, a false positive is when normal data is mistakenly identified as abnormal, a true negative is when normal data is correctly identified, a false negative is when abnormal data is mistakenly identified as normal, and a true positive is when abnormal data is correctly identified.

[0015] Preferably, the model monitoring and iteration module further includes an iterative optimization unit, which is configured as follows: Samples with a false positive rate exceeding 5% or a false negative rate exceeding 3% are labeled and added to the training set. The model learning rate is optimized using AutoML technology, and the learning rate is dynamically adjusted based on the initial learning rate, decay coefficient, and number of iterations.

[0016] Through the above technical solutions, the beneficial effects of this invention are as follows: This invention, through multi-stage collaborative operation, accurately acquires and processes key data from AC / DC power systems, achieving efficient storage and intelligent analysis. It accurately detects anomalies, locates the root causes of faults, predicts the remaining lifespan of equipment, and formulates optimized maintenance plans. Combined with intelligent alarms and intuitive displays, and through dynamic model optimization, it ensures long-term accuracy. This not only solves the problems of incomplete data display and poor real-time performance associated with traditional third-party platforms, but also realizes a shift from passive maintenance to proactive prevention, significantly improving system reliability and stability, reducing operation and maintenance costs, providing comprehensive support for intelligent operation and maintenance of substations, and promoting the efficient development of power systems towards digitalization and intelligence. Attached Figure Description

[0017] Figure 1 This is a system module structure diagram of the AC / DC integrated power data management and control platform at the station end of the present invention.

[0018] Figure 2 This is a structural block diagram of the intelligent analysis module of the present invention.

[0019] Figure 3 This is a schematic diagram of the risk model for the lifetime prediction module of this invention. Detailed Implementation

[0020] The foregoing and other technical contents, features and effects of the present invention are described in conjunction with the appendix below. Figure 1 To be continued Figure 3 The detailed description of the embodiments will make this clear. All structural details mentioned in the following embodiments are based on the accompanying drawings.

[0021] Exemplary embodiments of the present invention will now be described with reference to the accompanying drawings.

[0022] like Figure 1 As shown, this invention provides a station-side integrated AC / DC power supply data management and control platform, including a data acquisition module, an edge preprocessing module, a distributed storage module, an intelligent analysis module, a lifespan prediction module, an alarm and visualization module, and a model monitoring and iteration module. The functions of each module are as follows: The data acquisition module serves as the fundamental input unit of the station-side integrated AC / DC power supply data management and control platform. Its core function is to collect key data from the AC / DC power supply system, specifically including voltage. Current ,temperature Insulation resistance .

[0023] To ensure reliable data acquisition, this module establishes real-time communication links with various sensors and devices in the AC / DC power supply system of the substation by adapting to common power system protocols (Modbus, IEC61850). Specifically, voltage data is acquired through voltage sensors, current data through current transformers, temperature data is monitored by temperature sensors, and insulation resistance data is acquired through insulation resistance detectors, ensuring that all key parameters are accurately captured.

[0024] The edge preprocessing module connects to the data acquisition module and is responsible for preprocessing the acquired raw data to eliminate noise, outliers, and standardize units, thereby improving data quality. The preprocessing process of this module includes the following operations: First, a sliding window mechanism is used to remove outliers. This is achieved by setting a window that includes... A sliding window continuously collects data, and the data sequence within the window is... ,in Represents the first in the window First, calculate the mean of the data within the window. The calculation formula is:

[0025] Then calculate the standard deviation of the data within the window based on the mean. The formula is:

[0026] When any data in the window satisfy If the data is found to be outlier, it is immediately removed. This mechanism uses statistical methods to identify transient interference data that deviates from the normal distribution, effectively preventing erroneous data caused by sudden sensor malfunctions or electromagnetic interference from entering subsequent processes.

[0027] Then, wavelet transform is used to denoise the acquired data containing high-frequency noise. For the high-frequency noise mixed in the original data, continuous wavelet transform is used to filter the noise; the transform formula is as follows:

[0028] in, This represents the raw data containing high-frequency noise. These are translation parameters used to control the position of the wavelet function on the time axis. The scaling parameter is used to adjust the frequency response of the wavelet function. For the pre-selected mother wavelet function, This represents the conjugate operation. The translation parameters are adjusted accordingly. and scale parameters This allows the wavelet function to effectively match the high-frequency noise components in the original data, and then filters out the noise through inverse transform, retaining the effective features in the data that reflect the operating status of the equipment.

[0029] Finally, the Min-Max method was used to normalize the collected data. To eliminate the influence of numerical differences in voltage, current, temperature, and insulation resistance due to different dimensions on subsequent algorithm analysis, the formula was used... Map the original data to a uniform numerical range. The raw data to be processed. This represents the minimum value of this type of data within the collection period. This refers to the maximum value of this type of data within the collection period. This is the data after normalization. This process maps different types of original data to the same numerical range, ensuring that subsequent intelligent analysis algorithms can perform collaborative calculations and feature extraction on various parameters.

[0030] Through the above operations, the edge preprocessing module can transform the raw collected data into standardized data that is free of anomalies, low in noise, and dimensionless, providing a high-quality data foundation for the data hierarchical management of the distributed storage module and the accurate modeling of the intelligent analysis module.

[0031] The distributed storage module connects to the edge preprocessing module, receives standardized data output from the edge preprocessing module, and uses a differentiated storage architecture according to data type to achieve efficient storage. Based on this, a data lake is built, and a hierarchical management strategy is used to integrate and flexibly access multi-source data, providing stable data access support for the intelligent analysis module.

[0032] To address the characteristics of different data types, differentiated storage architectures are designed for each module to match its read and write requirements. Voltage and current time-series data are characterized by high-frequency generation, continuous writing, and the need for fast time-series queries; therefore, they are stored in an InfluxDB cluster. The write frequency of the time-series data is defined as... The query response time is Through InfluxDB's time-series optimization engine, it can be ensured that Still maintain This system meets the requirements for real-time writing of high-frequency data and historical trend querying. Equipment metadata (including sensor model and detection equipment parameters) and maintenance records are structured data and need to support multi-condition join queries; therefore, they are stored in PostgreSQL. Through PostgreSQL's relational data model, the association mapping between equipment metadata and maintenance records can be realized, and join query efficiency metrics can be defined. =Actual query time / Theoretical optimal time, ensuring ≤1.5, ensuring efficient management and related retrieval of structured data.

[0033] Simultaneously, the module builds a data lake based on the Hadoop ecosystem, enabling unified integration and batch / stream analysis of multi-source data. The data lake logically connects data from InfluxDB and PostgreSQL through a distributed file system, defining data integration integrity metrics. =Successfully integrated data volume / Total data to be integrated, ensuring A resolution of ≥0.99 enables seamless integration of various data types. Based on this, a tiered storage strategy is employed to balance storage costs and access efficiency: Frequently accessed data from the past three months (such as frequently updated real-time voltage and current data) is stored in a Redis in-memory database, with defined access frequencies for this data. ,satisfy For applications requiring rapid access, its memory storage response time Warm data from the past year is stored in Parquet columnar storage, with a defined frequency for accessing warm data. ,satisfy For real-time query needs, columnar compression storage reduces data footprint to 30%–50% of its original size, balancing storage efficiency and query speed. Historical cold data exceeding one year (such as maintenance records and long-term trend data) is stored in compatible object storage, with defined access frequencies for cold data. ,satisfy For low-frequency access needs, its storage cost is only 10% to 20% of that of hot data storage, significantly reducing the cost of long-term data preservation.

[0034] Through the aforementioned differentiated storage architecture and hierarchical management strategy, the distributed storage module achieves efficient management of data of different types and access frequencies. This ensures fast read and write of high-frequency real-time data while controlling overall costs through hierarchical storage, providing flexible, efficient, and low-cost support for data retrieval by the subsequent intelligent analysis module.

[0035] The intelligent analysis module connects to the distributed storage module and uses multiple algorithms to perform intelligent analysis of data in the data lake, enabling equipment anomaly detection and root cause analysis, such as... Figure 2 As shown, it contains four core units: The gradual fault detection unit is configured to identify potential faults that change slowly during equipment operation. Its operating logic is as follows: voltage and current timing data stored in the distributed storage module are input into an LSTM neural network. The model learns the voltage and current variation patterns under normal equipment operation and outputs the results in real time. Predicted value at time Simultaneously, obtain the actual voltage or current value at that moment. Calculate the prediction error of both. Preset threshold Based on the error distribution characteristics determined from historical normal operation data, when real-time calculations... When a fault is detected, the device is identified as having a gradual fault. This mechanism can effectively detect faults that develop slowly over time, such as battery capacity decay, and prevent system downtime caused by sudden faults.

[0036] The current waveform distortion recognition unit is used to detect abnormal distortions such as harmonics and interharmonics in current waveforms. It extracts features from the current waveform data using a 1D-CNN, specifically through multi-layer convolutional operations: [The text abruptly ends here, so the translation stops as well.] The formula for the output of a layer convolution is: ,in Representing the Layer The weights of each convolutional kernel, For the first The current waveform characteristics of the layer input. For the first Layer bias parameters, It is the ReLU activation function. This indicates a convolution operation. After extracting and enhancing the local features of the current waveform layer by layer through multi-layer convolution, the final output features are matched with a preset harmonic and interharmonic feature library. When the matching degree reaches 90% or above, it is determined that the current waveform is distorted, which can promptly detect waveform pollution problems caused by rectifier module abnormalities.

[0037] The insulation resistance anomaly marking unit identifies anomalies in the insulation resistance data. It processes the insulation resistance data using a self-encoder: the encoder reads the input insulation resistance data... Mapped to a low-dimensional feature space, the decoder then reconstructs the low-dimensional features into output data with the same dimension as the original data; the original data is then computed. With the second norm error of the reconstructed data ,in For encoder, For decoder. Preset threshold. The reconstruction error distribution based on historical normal insulation resistance data is calculated as follows: When an abnormal point is detected, the insulation resistance data is marked as an anomaly, thereby accurately capturing potential risks such as insulation degradation.

[0038] The root cause analysis unit is responsible for locating the specific fault type and cause after detecting an anomaly. It employs an XGBoost multi-classification model, using temperature and charge / discharge cycles as input features, and outputs the specific fault type, including rectifier module breakdown and battery sulfation. The model's objective function is... ,in The cross-entropy loss function is used to measure the actual fault type label. and Wheel Prediction Fault Type The differences between them; For regularization terms, The number of tree nodes. The weights of the leaf nodes are... , This is the regularization coefficient. By constraining the model complexity through the regularization term, the risk of overfitting is reduced, ensuring accurate output of fault types when given features such as temperature and charge / discharge cycles, thus providing maintenance personnel with clear guidance on the root causes of faults.

[0039] Through the coordinated operation of the above four units, the intelligent analysis module realizes a complete analysis process from anomaly detection to root cause location of faults, providing core technical support for the platform's intelligent operation and maintenance decision-making.

[0040] The lifespan prediction module connects to the intelligent analysis module. Based on the equipment anomaly detection and root cause analysis results output by the intelligent analysis module, it calculates the remaining lifespan of the equipment using a professional model and formulates differentiated maintenance plans based on equipment characteristics, providing a forward-looking basis for operation and maintenance decisions. The core component of this module is the remaining lifespan prediction unit, whose operating logic revolves around the Cox proportional hazards model to achieve accurate prediction of equipment lifespan.

[0041] like Figure 3 As shown, the remaining life prediction unit uses the equipment status characteristics identified by the intelligent analysis module as a basis to select covariates that significantly affect equipment aging to construct an analysis model. Specifically, the covariate vector... Runtime and ambient temperature Composition, that is ,in Reflecting the impact of cumulative operating time on aging of equipment, This demonstrates the accelerating effect of ambient temperature fluctuations on equipment lifespan. The Cox proportional hazards model is used to calculate the impact of these fluctuations on equipment lifespan at any given time. risk rate The model formula is:

[0042] In the formula, The baseline risk rate represents the inherent risk rate of the equipment when the covariate value is zero. This is a vector of covariate coefficients used to quantify the impact of runtime and ambient temperature on the risk rate. Its components correspond to... and Influence weight The risk ratio reflects the multiple of the equipment risk rate relative to the benchmark risk rate under the current covariate combination.

[0043] Based on the trend of risk rate changes over time, the remaining life prediction unit derives the remaining life of the equipment. Specifically, this is achieved by continuously monitoring the risk rate. The dynamic changes, when the risk rate reaches a preset critical threshold. The difference between the corresponding point in time and the current moment is the remaining lifespan of the equipment. ,Right now ,in To achieve the risk rate At that moment, This is the current moment in the analysis.

[0044] After obtaining the remaining life prediction results, the module combines them with equipment importance parameters. ( The value ranges from 0 to 1, with higher values ​​indicating greater criticality of the equipment to system stability. A differentiated maintenance plan is then developed. For the remaining lifespan... And importance For equipment with high remaining lifespan or low importance, priority is given to including it in the near-term maintenance plan; for equipment with a long remaining lifespan or low importance, periodic inspection plans are developed. Through this data-driven, precise planning, the module achieves a shift from traditional passive maintenance to proactive preventative maintenance, effectively reducing the risk of sudden equipment failures.

[0045] In summary, the lifespan prediction module uses the Cox proportional hazards model to correlate key factors such as equipment runtime and ambient temperature with lifespan, and combines intelligent analysis results to output remaining lifespan and maintenance plans, providing core decision support for the platform's intelligent operation and maintenance.

[0046] The alarm and visualization module, connected to the lifespan prediction module, is the core link in the platform's status awareness and decision communication. Its core function is to capture abnormal equipment status in real time through the intelligent alarm mechanism based on the remaining lifespan prediction results and maintenance plans output by the lifespan prediction module, and to transform complex data into intuitive information with the help of visualization technology, providing operation and maintenance personnel with accurate and easy-to-understand decision support.

[0047] The core component of this module is the intelligent alarm unit, which is responsible for identifying abnormal states during equipment operation in real time. Its core technology relies on a Gaussian mixture model to evaluate the probability density of real-time monitoring data. The Gaussian mixture model is a weighted combination of multiple normally distributed components, and its probability density calculation formula is as follows:

[0048] In the formula, For the first The weights of each normally distributed component, and satisfying This is used to measure the contribution of the component to the overall probability density; For the first There are normally distributed components, among which This is the mean of the component. The variance of this component describes the central location and dispersion of the distribution. This model allows for the quantification of real-time monitoring data. Probability distribution characteristics under normal operating conditions.

[0049] Preset threshold The probability density distribution based on historical alarm data is used to determine the lowest probability boundary representing the normal operating state. When the probability density value is calculated in real time... When the device deviates from its normal operating state, an alarm is triggered. Simultaneously, the unit traces the root cause of the fault using a Bayesian network: based on historical fault data, it constructs a correlation between faults and characteristics. When an alarm is triggered, it uses conditional probability reasoning to locate the most likely root cause of the current anomaly, providing maintenance personnel with direct troubleshooting directions.

[0050] The visualization function presents equipment status and prediction results in an intuitive way, including two types of core charts: a heat map of equipment health status drawn based on D3.js technology, with temperature as the horizontal axis and insulation resistance as the vertical axis, and quantifies the health index through color depth (the darker the color, the worse the health status), intuitively presenting the health differences of different equipment; and a fault prediction trend chart drawn based on the Prophet and ARIMA hybrid model, which shows the changing trend of equipment fault risk over a period of time through the correspondence between the time axis and the fault risk value.

[0051] Furthermore, the module automatically generates maintenance reports using natural language processing technology. These reports clearly include the equipment's remaining lifespan, potential failure risk levels, and targeted maintenance recommendations. The remaining lifespan is directly presented based on the calculation results from the lifespan prediction module, the potential failure risk levels are categorized according to the anomaly detection results from the intelligent analysis module, and the maintenance recommendations are generated in conjunction with the maintenance plan, ensuring that maintenance personnel can quickly understand the equipment status and take appropriate action.

[0052] Through the real-time monitoring and intuitive visualization of the intelligent alarm unit, the alarm and visualization module achieves efficient transformation from data to decision-making, providing clear and timely support for the refined operation and maintenance of AC / DC power supply systems in substations.

[0053] The model monitoring and iteration module is connected to the intelligent analysis module, lifespan prediction module, and alarm and visualization module, respectively. Through full lifecycle performance monitoring and dynamic iteration, it ensures that the algorithm models used by each module can continuously adapt to equipment aging trends and changes in operating conditions. This module includes a performance monitoring unit and an iterative optimization unit, and its specific operation is as follows: The core function of the performance monitoring unit is to evaluate the model's detection accuracy in real time, quantifying the model's misclassification rate through a confusion matrix. Four basic metrics are defined: the number of normal data points misclassified as anomalies (FP); the number of normal data points correctly classified as normal (TN); the number of anomaly data points misclassified as normal (FN); and the number of anomaly data points correctly classified as anomalies (TP). Based on these four metrics, two key performance parameters are calculated: the false positive rate (FPR) and the false negative rate (FNR). The false positive rate is calculated as FPR = FP / (FP + TN), measuring the proportion of normal data incorrectly labeled as anomalies; the false negative rate is calculated as FNR = FN / (TP + FN), measuring the proportion of anomaly data incorrectly ignored. By calculating and tracking FPR and FNR in real time, the performance monitoring unit can accurately capture changes in model accuracy at different operational stages, providing a clear basis for subsequent optimization.

[0054] The iterative optimization unit dynamically updates the model based on the evaluation results of the performance monitoring unit. When the false positive rate exceeds 5% or the false negative rate exceeds 3%, the unit manually labels these misclassified samples (including normal data samples corresponding to FP and abnormal data samples corresponding to FN) and adds them to the model training set to enhance the model's ability to identify edge cases. Simultaneously, the learning rate of the model is optimized using AutoML technology; the dynamic adjustment formula for the learning rate is as follows: In the formula, The initial learning rate is the initial step size for model training. This is the decay coefficient, used to control the rate at which the learning rate decreases; The number of iterations reflects the number of training rounds of the model. This formula shows that the learning rate gradually decreases as the number of iterations increases, allowing the model to converge quickly in the early stages of training and then fine-tuning the parameters in later stages to improve generalization ability.

[0055] Through real-time evaluation by the performance monitoring unit and dynamic adjustment by the iterative optimization unit, the model monitoring and iteration module ensures that the algorithm models used by modules such as intelligent analysis and life prediction can continuously adapt to changes in equipment operating status, maintain high analysis accuracy over a long period of time, and provide technical support for the stable and reliable operation of the platform.

[0056] In summary, this invention accurately acquires key data such as voltage, current, temperature, and insulation resistance of AC / DC power systems through a data acquisition module. The edge preprocessing module removes outliers, reduces noise, and standardizes the data to improve quality. The distributed storage module's differentiated architecture and hierarchical management enable efficient data storage and integration. The intelligent analysis module's multi-algorithm collaboration accurately detects equipment anomalies and pinpoints root causes. The lifespan prediction module derives remaining lifespan based on a risk model and develops differentiated maintenance plans. The alarm and visualization module provides intelligent alarms and intuitive displays. The model monitoring and iteration module dynamically optimizes model performance. This effectively improves the accuracy of data acquisition, the professionalism of analysis, and the forward-looking nature of operation and maintenance. It solves the problems of incomplete data display and poor real-time performance caused by traditional reliance on third-party platforms, and achieves a shift from passive maintenance to proactive prevention. This significantly improves the operational reliability and stability of AC / DC power systems, reduces operation and maintenance costs, provides comprehensive data support and decision-making assurance for intelligent operation and maintenance of substations, and promotes the efficient development of power systems towards digitalization and intelligence.

[0057] The above is a further detailed description of the present invention in conjunction with specific embodiments. It should not be considered that the specific implementation of the present invention is limited to this. For those skilled in the art to which the present invention pertains and related fields, any extensions, operation methods, and data substitutions made based on the technical solution concept of the present invention should fall within the protection scope of the present invention.

Claims

1. A station-end AC-DC integrated power supply data management and control platform, characterized in that, include: The data acquisition module is used to collect key data of the AC / DC power supply system, including voltage, current, temperature, and insulation resistance. An edge preprocessing module, connected to the data acquisition module, is used to preprocess the acquired key data; A distributed storage module, connected to the edge preprocessing module, is used to store preprocessed data according to data type and to build a data lake to achieve multi-source data integration and hierarchical management; The intelligent analysis module, connected to the distributed storage module, is used to perform intelligent analysis on the data lake to realize equipment anomaly detection and fault root cause analysis. The lifespan prediction module is connected to the intelligent analysis module and is used to generate equipment remaining lifespan prediction results and maintenance plans based on the intelligent analysis results. The alarm and visualization module is connected to the life prediction module and is used to generate operation and maintenance reports based on prediction results and maintenance plans through intelligent alarm mechanisms and visualization technology. The model monitoring and iteration module is connected to the intelligent analysis module, the lifetime prediction module, and the alarm and visualization module, respectively. It is used to monitor the model performance in the intelligent analysis, prediction, and decision-making process, and to achieve model iterative optimization through data closed-loop feedback.

2. The station-side AC / DC integrated power supply data management and control platform as described in claim 1, characterized in that, The preprocessing operations performed by the edge preprocessing module include: A sliding window mechanism is used to remove outliers from the collected data; wavelet transform is used to reduce noise in the collected data containing high-frequency noise; and the Min-Max method is used to normalize the collected data to eliminate dimensional differences.

3. The station-side AC / DC integrated power supply data management and control platform as described in claim 2, characterized in that, The intelligent analysis module includes a gradual fault detection unit, which is configured as follows: The time-series data of voltage and current are input into the LSTM neural network to calculate the error between the real-time predicted value and the actual value. When the error exceeds the threshold set based on the error distribution of historical normal data, it is determined to be a gradual fault.

4. The station-side AC / DC integrated power supply data management and control platform as described in claim 3, characterized in that, The intelligent analysis module also includes a current waveform distortion recognition unit, which is configured as follows: 1DCNN is used to process current waveform data. Features are extracted through multi-layer convolution operations. Each convolution layer performs convolution operations with the kernel weights and features of the previous layer and adjusts the bias. The output features are then processed by the ReLU activation function to generate output features. When the matching degree between the output features and the preset harmonic and interharmonic feature library reaches 90% or more, it is determined to be current waveform distortion.

5. The station-side AC / DC integrated power data management and control platform as described in claim 4, characterized in that, The intelligent analysis module also includes an insulation resistance anomaly marking unit, which is configured as follows: The insulation resistance data is processed by an autoencoder, and the second norm error between the original data and the data reconstructed by the encoder and decoder is calculated. When the error exceeds the reconstruction error threshold calculated based on historical normal insulation resistance data, it is marked as an anomaly.

6. The station-side AC / DC integrated power data management and control platform as described in claim 5, characterized in that, The intelligent analysis module also includes a fault root cause analysis unit, which is configured as follows: The XGBoost multi-classification model is adopted, with temperature and charge / discharge cycles as input features, and output fault types. The objective function of the model includes a cross-entropy loss term and a regularization term. The regularization term consists of the number of tree nodes, leaf node weights, and regularization coefficients. Fault types include rectifier module breakdown and battery sulfation.

7. The station-side AC / DC integrated power data management and control platform as described in claim 6, characterized in that, The lifetime prediction module includes a remaining lifetime prediction unit, which is configured as follows: The Cox proportional hazards model is adopted, with operating time and ambient temperature as covariates. The equipment risk rate is calculated by multiplying the baseline risk rate by the exponential function of the covariates, and the remaining life is calculated based on the trend of risk rate changes.

8. The station-side AC / DC integrated power supply data management and control platform as described in claim 7, characterized in that, The alarm and visualization module includes an intelligent alarm unit, which is configured as follows: A Gaussian mixture model is used to process the real-time monitoring data. The model is composed of multiple normally distributed components weighted together, with the sum of the weights of each component being 1. When the probability density value of the real-time monitoring data in the model is lower than the threshold set based on historical alarm data, an alarm is triggered. The real-time monitoring data includes voltage, current, and temperature.

9. The station-side AC / DC integrated power supply data management and control platform as described in claim 8, characterized in that, The model monitoring and iteration module includes a performance monitoring unit, which is configured as follows: The false positive rate and false negative rate of the model are calculated using a confusion matrix. The false positive rate is the ratio of the number of false positive data to the sum of the number of false positive and true negative data, and the false negative rate is the ratio of the number of false negative data to the sum of the number of true positive and false negative data. False positives are normal data that is mistakenly identified as abnormal, true negatives are normal data that is correctly identified, false negatives are abnormal data that is mistakenly identified as normal, and true positives are abnormal data that is correctly identified.

10. The station-side AC / DC integrated power supply data management and control platform as described in claim 9, characterized in that, The model monitoring and iteration module also includes an iterative optimization unit, which is configured as follows: Samples with a false positive rate exceeding 5% or a false negative rate exceeding 3% are labeled and added to the training set. The model learning rate is optimized using AutoML technology, and the learning rate is dynamically adjusted based on the initial learning rate, decay coefficient, and number of iterations.

Images