Distribution transformer short-term overload early warning method and device based on low-voltage meter data

Abstract

The invention discloses a distribution transformer short-term overload early warning method and device based on low-voltage meter data, and the method comprises the steps: obtaining historical electricity data which is obtained by collecting the reading of a low-voltage meter according to a preset time period; extracting load data, meteorological association features and event features of the historical electricity data; calculating a risk prediction value of the target time period according to the load data, the meteorological correlation features and the event features; obtaining a dynamic risk threshold according to the scene type corresponding to the target time period; and comparing the risk prediction value with a dynamic risk threshold value to obtain a risk level, and generating early warning information according to the risk level. The load fluctuation trend can be reflected more accurately based on the electric data acquired by the low-voltage meter, the dynamic risk threshold is acquired according to the scene type corresponding to the target time period, and after different dynamic risk thresholds are acquired in different scenes, the occurrence of false alarm and missing alarm can be reduced, so that the accuracy of power grid fluctuation peak early warning is improved.

Description

Technical Field

[0001] This invention relates to the field of power grid operation and maintenance technology, and in particular to a method and device for early warning of short-term overload of distribution transformers based on low-voltage meter data. Background Technology

[0002] During power grid operation, seasonal load surges such as the Spring Festival travel rush, agricultural irrigation, and extreme high temperatures can lead to short-term overloads of distribution transformers. These short-term transformer overloads pose a serious threat to power system stability and equipment lifespan. Related technologies, such as SCADA systems, struggle to capture instantaneous load fluctuations, resulting in frequent "peak load missed detection" problems. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method and device for early warning of short-term overload of distribution transformers based on low-voltage meter data, so as to improve the accuracy of early warning of power grid fluctuation peaks.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for early warning of short-term overload of distribution transformers based on low-voltage meter data includes: Historical electricity data is acquired by collecting readings from low-voltage meters at preset time intervals. Extract load data, meteorological correlation features, and event features from the historical electricity data; The risk prediction value for the target period is calculated based on the load data, meteorological correlation characteristics, and event characteristics. A dynamic risk threshold is obtained based on the scenario type corresponding to the target time period; The risk level is obtained by comparing the predicted risk value with the dynamic risk threshold, and an early warning message is generated based on the risk level.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: An electronic device, further comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the aforementioned method for early warning of short-term overload of distribution transformers based on low-pressure meter data.

[0006] The beneficial effects of this invention are as follows: compared with the SCADA system's 30-minute data sampling method, the electricity data collected by low-voltage meters can more accurately reflect the load fluctuation trend; when processing historical electricity data, after extracting the load data, meteorological correlation characteristics, and event characteristics of the historical electricity data to calculate the risk prediction value for the target period, a dynamic risk threshold is obtained according to the scenario type corresponding to the target period. This allows for the acquisition of different dynamic risk thresholds for different scenarios. By comparing the risk prediction value with the dynamic risk threshold, the occurrence of false alarms and missed alarms is reduced, thereby improving the accuracy of early warning of power grid fluctuation peaks. Attached Figure Description

[0007] Figure 1 This is a flowchart illustrating the steps of a short-term overload early warning method for distribution transformers based on low-pressure meter data, as described in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the specific application of a short-term overload early warning method for distribution transformers based on low-pressure meter data in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of a distribution transformer short-term overload early warning system according to an embodiment of the present invention; Figure 4 This is a load prediction output diagram of a distribution transformer short-term overload early warning method based on low-pressure meter data in an embodiment of the present invention. Figure 5 This is a comparison chart of model performance under different early warning thresholds in the embodiments of the present invention; Figure 6 This is a schematic diagram of the warning interface output by a method for short-term overload warning of distribution transformers based on low-pressure meter data in an embodiment of the present invention. Figure 7 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0008] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0009] Definitions:

[0010] In related technologies, traditional SCADA systems use 30-minute data sampling, which makes it difficult to capture instantaneous load fluctuations, leading to frequent "peak load missed detection" problems. Studies have shown that during peak summer electricity consumption, 30% of the load peak duration for residential distribution transformers is less than 10 minutes, but such instantaneous overloads still accelerate transformer insulation aging. Furthermore, traditional short-term overload early warning mechanisms for distribution transformers are usually based on annual maximum load forecasts plus fixed redundancy coefficients for capacity planning, resulting in long cycles and slow responses; and the use of fixed threshold alarms cannot adapt to seasonal load changes, easily leading to false alarms or missed alarms.

[0011] To address the aforementioned technical problems, this invention provides a method and device for early warning of short-term overload of distribution transformers based on low-voltage meter data. This method and device can be widely applied in power system distribution network operation monitoring, load forecasting, and equipment maintenance, and is particularly suitable for rural power grids, urban-rural fringe areas, and other regions with significant seasonal load fluctuations. Specifically: Please refer to Figure 1 A method for early warning of short-term overload of distribution transformers based on low-voltage meter data, comprising: Historical electricity data is acquired by collecting readings from low-voltage meters at preset time intervals. Extract load data, meteorological correlation features, and event features from the historical electricity data; The risk prediction value for the target period is calculated based on the load data, meteorological correlation characteristics, and event characteristics. A dynamic risk threshold is obtained based on the scenario type corresponding to the target time period; The risk level is obtained by comparing the predicted risk value with the dynamic risk threshold, and an early warning message is generated based on the risk level.

[0012] As described above, the beneficial effects of this invention are as follows: Based on low-voltage meter data collection, compared to the 30-minute data sampling method used in SCADA systems, it can more accurately reflect load fluctuation trends; when processing historical electricity data, by extracting load data, meteorological correlation features, and event features from historical electricity data to calculate the risk prediction value for the target time period, and then obtaining a dynamic risk threshold based on the scenario type corresponding to the target time period, it can adapt to different scenarios and obtain different dynamic risk thresholds. Comparing the risk prediction value with the dynamic risk threshold reduces false alarms and missed alarms, thereby improving the accuracy of early warning for power grid fluctuation peaks.

[0013] In one embodiment of this application, the extraction of load data, meteorological correlation features, and event features from the historical electricity data includes: The historical electricity data is classified using a clustering algorithm to obtain target-classified load data, as well as the target meteorological correlation features and target event features corresponding to the target-classified load data; The target load data includes residential load data, agricultural load data, and commercial load data.

[0014] As described above, by using clustering algorithms to classify historical electricity data, relevant data under different categories such as residential, agricultural, and commercial load data can be obtained. Based on the load proportion of different categories, the main loads in different time periods can be obtained, and the sources of load growth can be located, thereby enabling targeted early warnings based on load classification.

[0015] In one embodiment of this application, it further includes: Calculate the predicted classification risk value corresponding to each of the target classification load data; Obtain the classification dynamic risk threshold corresponding to each of the target classification load data; The classification risk level is obtained by comparing each of the predicted risk values ​​for the classification with its corresponding dynamic risk threshold. Comprehensive early warning information is generated based on the classification risk level corresponding to each of the target classification load data.

[0016] As described above, by obtaining the predicted risk value and dynamic risk threshold for each category and comparing them, the risk situation for each category can be obtained. Based on the risk levels of all categories, comprehensive early warning information can be generated, thereby enabling corresponding early warning operations and comprehensive load adjustments to be made according to the risk situation of different categories.

[0017] In one embodiment of this application, the step of calculating the risk prediction value for the target time period based on the load data, meteorological correlation characteristics, and event characteristics includes: ; in, This is the risk prediction value; the average load rate is obtained from the load data. and load growth rate The temperature correction coefficient is obtained through the meteorological correlation characteristics. Holiday factors are obtained through the event characteristics. .

[0018] As described above, by obtaining the average load rate and load growth rate from the load data, and combining the temperature correction coefficient obtained through meteorological correlation characteristics, as well as the holiday factor obtained through event characteristics, a risk prediction value can be obtained for the target period.

[0019] In one embodiment of this application, the temperature correction coefficient is obtained through the meteorological correlation features. include: ; in, The elastic coefficient; Real-time ambient temperature; This is the reference temperature.

[0020] As described above, when obtaining the temperature correction coefficient, the calculation is based on the real-time ambient temperature and the reference temperature, and the temperature correction coefficient is also adjusted by the elasticity coefficient so that the temperature correction coefficient can better reflect the impact of temperature on the risk prediction value in the current period.

[0021] In one embodiment of this application, obtaining the dynamic risk threshold based on the scenario type corresponding to the target time period includes: The basic risk threshold is obtained based on the scenario type corresponding to the target time period; Determine whether there is a sudden rise in temperature or / and a sudden change in load growth rate during the target period. If so, reduce the basic risk threshold to obtain the dynamic risk threshold.

[0022] As described above, when obtaining the dynamic risk threshold, the basic risk threshold is first obtained based on the scenario corresponding to the target time period. Then, the basic risk threshold is adjusted in combination with the temperature change and load growth of the target time period, so that the dynamic risk threshold can better reflect the actual risk situation of the target time period.

[0023] In one embodiment of this application, after acquiring historical electrical data, the method further includes: The historical electrical data is processed by removing outliers, filling in missing values, and normalizing.

[0024] As described above, removing outliers, filling in missing values, and normalizing the data after acquiring historical electrical data can ensure the validity of the electrical data.

[0025] In one embodiment of this application, after generating the early warning information based on the risk level, the method further includes: Obtain the execution result of the aforementioned warning information; The dynamic risk threshold is optimized based on the execution results.

[0026] As described above, by optimizing the dynamic risk threshold based on the execution results after the early warning information is executed, the subsequent early warning strategy can be dynamically adjusted.

[0027] In one embodiment of this application, the dynamic risk threshold includes at least two different levels of sub-thresholds; The step of comparing the predicted risk value with the dynamic risk threshold to obtain the risk level includes: The predicted risk value is compared with each of the sub-thresholds in turn to obtain the highest risk as the risk level.

[0028] As described above, by setting the dynamic risk threshold to multiple sub-thresholds at different levels and comparing the predicted risk value with each sub-threshold in turn, the corresponding risk level can be output when the dynamic risk threshold reaches different sub-thresholds.

[0029] Another embodiment of the present invention provides an electronic device, which further includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the distribution transformer short-term overload early warning method based on low-pressure meter data as described above.

[0030] Embodiment 1 of the present invention is as follows: Please refer to Figure 1 as well as Figure 2 A method for early warning of short-term overload in distribution transformers based on low-voltage meter data is proposed and applied to a short-term overload early warning system for distribution transformers. Please refer to [reference needed]. Figure 3 The system comprises a data source layer, a communication layer, a data processing layer, and an application layer. The data source layer collects residential / agricultural / commercial electricity load data based on 15-minute low-voltage meter data (smart meters, concentrators, etc.), offering a 200% higher resolution than traditional SCADA systems. Data sources include voltage / current waveform sampling (128 points / cycle), active / reactive power (0.2S accuracy), and temperature sensor data (±1℃ error). The communication layer employs a two-tier architecture of "concentrator-collector": the collector layer uploads data at a rate of 50kbps via HPLC; the concentrator layer includes dual-channel redundant transmission (4G / fiber optic) with a latency of <500ms. The data processing layer processes the collected data, i.e., executes subsequent S2. The application layer performs calculations and analysis on the data processed by the data processing layer, outputting early warning information and operational decisions, i.e., executes subsequent S3-S5.

[0031] The method includes the following steps: S1. Acquire historical electricity data, which is obtained by collecting readings from low-voltage meters at preset time intervals; for example, collecting 15-minute data from commercial smart meters, residential smart meters, and agricultural smart meters. It can collect operating data such as active power, reactive power, voltage, and current from all low-voltage user smart meters in the distribution area, with a collection frequency of 15 minutes / time; simultaneously, it collects ambient temperature data (from the nearest weather station or local sensor) for subsequent temperature correction; obtaining the original load data package (including timestamp, user ID, power value, voltage value, and temperature value), which is uploaded to the concentrator via HPLC power line carrier, and then transmitted to the main station server via 4G / fiber optic channel, with transmission latency controlled within 500ms. This step is the data source for the entire early warning process, and its data quality and completeness directly affect the accuracy of subsequent model calculations; therefore, it is necessary to ensure a data acquisition success rate ≥99% and a missing data ratio <1%.

[0032] S2. Extract load data, meteorological correlation features, and event features from the historical electricity data. Before processing the data, outliers need to be removed, missing values ​​filled, and the data normalized.

[0033] S21. Data Cleaning and Anomaly Handling: First, the ±3σ criterion is used to identify and remove abnormal load jump points. Combined with box plotting (IQR=1.5), historical data is validated a second time. Load data exceeding the threshold ±15% are marked and removed. This must meet the IEEE 1159-2019 standard validation requirement, with an anomaly detection accuracy ≥97%. For data with no more than 3 consecutive missing sampling points (45 minutes), linear interpolation is used to fill in the gaps, with interpolation errors controlled within 2%. Finally, all data undergoes Min-Max normalization, mapping to the [0,1] interval to ensure comparability of data with different dimensions. This step is a prerequisite for subsequent model input; the cleaned data will serve as the sole input for load splitting and feature extraction.

[0034] S22. Load Splitting: The historical electricity data is classified using a clustering algorithm to obtain target-classified load data and the corresponding target meteorological correlation features and target event features; for example, three scenarios (residential / agricultural / commercial) are divided using K-means clustering, with a silhouette coefficient ≥ 0.65 (referring to the "good clustering" standard in GB / T 33607-2017 "Technical Requirements for Electricity Load Clustering Analysis"), to obtain residential load data, agricultural load data, and commercial load data.

[0035] Clustering parameters include: number of clusters k=3 (corresponding to three types of electricity consumption: residential, agricultural, and commercial); distance metric: Euclidean distance; centroid initialization: K-means++ (to reduce the risk of local optima); iteration termination condition: rate of change of the sum of squared errors <0.1% or maximum number of iterations max_iter=300; input features: electricity consumption every 15 minutes, time period (morning peak / evening peak / night), and voltage fluctuation rate.

[0036] S23. Feature Extraction: The STL decomposition method is used to extract the periodic, trend, and residual terms corresponding to the load. The proportion of the periodic term is used to determine load stability, with a periodic term proportion of 35.7% ± 2.1%. Sliding window optimization can be combined with this method. STL parameter settings include: Periodic term period: period=96 (corresponding to 24 hours, with a sampling interval of 15 minutes); Seasonal smoothing parameter: s_period=12 (smoothing periodic terms); Trend term smoothing parameter: s_trend=48 (smoothing trend terms); Residual term retention: used for anomaly detection (such as sudden increases or decreases).

[0037] The temperature-load elasticity coefficient (0.78-1.32) was calculated using a multivariate coupling model based on measured data from multiple provinces and cities. The scenario matching algorithm used the State Grid's Spring Festival load report data, setting holiday factors (e.g., a Spring Festival multiple of 1.8) as event features. Ultimately, the algorithm outputs three load components (residential / agricultural / commercial) and their corresponding time-series characteristics, meteorological correlation characteristics, and event features. Further analysis can be performed to obtain the Spring Festival return-home growth rate, the utilization rate of irrigation equipment during busy farming seasons, and the high-temperature air conditioning load index.

[0038] S3. Based on the load data, meteorological correlation characteristics, and event characteristics, the risk prediction value for the target time period is calculated. The specific calculation method is as follows: ; ; in, This is the risk prediction value; the average load rate is obtained from the load data. and load growth rate ,For example The historical 7-day moving average load factor (calculated daily, in %). The load is predicted by the ARIMA model based on historical data, with MAPE ≤ 5% and a value range of 0.01-0.3, corresponding to a load increase of 1%-30%. The ARIMA parameters are set as follows: difference order d=1 (to eliminate data trends); autoregressive term p=2 (to consider the influence of 2 time steps); moving average term q=1; seasonal parameter (if needed): seasonal period s=24 (daily period); training data is based on 7 days of historical 15-minute load rate data, and the parameter combination is optimized through the AIC criterion.

[0039] The temperature correction coefficient is obtained through the meteorological correlation characteristics. (For example, a change of 0.8%-1.2% per 1℃, verified by Pearson correlation analysis r=0.82); The elastic coefficient; Real-time ambient temperature; The baseline temperature is set to 25°C; holiday factors are obtained based on the event characteristics. (Values ​​range from 1.0 to 2.0, dynamically assigned based on event type, e.g., Spring Festival: 1.8, busy irrigation season: 1.5, other regular periods: 1.0). The calculation model can output a predicted load rate sequence every 15 minutes for the next 72 hours, along with the corresponding warning level (Level I / II / III). Based on the above calculation method, the predicted risk values ​​for the three categories of residential, agricultural, and commercial can be calculated. For example... Figure 4 As shown, the predicted load corresponding to a 3-day prediction window can be output based on historical load forecasts.

[0040] S4. Obtain a dynamic risk threshold based on the scenario type corresponding to the target time period. Specifically, the dynamic risk threshold includes at least two different levels of sub-thresholds. The predicted risk value is compared sequentially with each of the sub-thresholds to obtain the highest risk as the risk level. For example, sub-thresholds include 70%, 80%, and 90%.

[0041] Please refer to Figure 5 The performance metrics obtained under different thresholds are shown in Table 1: Table 1. Performance indicators for different thresholds

[0042] Table 1 shows that the 70% threshold scheme presents a technical and economic dilemma. High sensitivity advantage: Employing a wide threshold strategy, the recall rate reaches 92%, effectively capturing instantaneous overload events such as sudden increases in irrigation load during busy farming seasons, making it particularly suitable for scenarios with drastic load fluctuations (such as the overlap of commercial and residential loads during peak summer electricity consumption). Resource consumption issue: A 15% false alarm rate results in 1.5 false alarms for every 10 warnings, increasing the cost of ineffective inspections by approximately 230,000 yuan per year (based on the power grid operation and maintenance costs of a certain province).

[0043] The 80% threshold scheme is relatively balanced. Comprehensive performance verification: An F1 value of 86% reflects a balance between accuracy and coverage; a false alarm rate of 8% meets the requirements for Level II early warning in the "Guidelines for Power System Safety and Stability," achieving Pareto optimality between preventative maintenance and emergency costs. Engineering demonstration results: In a provincial power grid application, this threshold reduced operation and maintenance costs by 23% while keeping equipment failure rates below 0.5 times / unit / year and reducing insulation life loss by 37%.

[0044] 90% Threshold Solution: High Specificity: 90% accuracy is suitable for areas with limited operational resources, but the 78% recall rate may miss 22% of early overload risks. It requires the use of emergency measures such as mobile energy storage. Its applicability is highly dependent on the feature weight update frequency of the DAENs model (weekly adjustment is recommended). The dynamic risk threshold adjustment strategy is as follows: Threshold selection requires establishing a multi-dimensional decision matrix and dynamically optimizing it based on scene characteristics:

[0045] Dynamic update of feature weights: Based on the sparse autoencoder structure of the DAENs model, the feature weights are optimized layer by layer through the KL divergence loss function, and the temperature factor is adjusted from the initial 0.82 to 0.79, etc.

[0046] Threshold elasticity mechanism: Set a ±5% floating range, and automatically trigger the threshold to be lowered when the following conditions occur: 1. Sudden temperature rise ≥3℃ / day (verified by Pearson correlation analysis r=0.82); 2. Sudden change in load growth rate (ARIMA prediction residual exceeds 2σ).

[0047] Closed-loop verification system: The accuracy of early warning is evaluated quarterly, and data quality audits are conducted in accordance with the IEEE 1159-2019 standard to ensure that the anomaly detection accuracy is maintained above 97.3%.

[0048] Among them, the threshold optimization of the distribution transformer overload early warning model needs to comprehensively consider the balance between technical performance and economic efficiency. For example, based on simulation experiments and actual engineering verification data of the IEEE 33-bus system, different threshold schemes show significant differences in operation and maintenance efficiency and risk control. Their selection should be dynamically adjusted according to seasonal characteristics and resource allocation.

[0049] S5. Compare the predicted risk value with the dynamic risk threshold to obtain the risk level, and generate early warning information based on the risk level. For example... Figure 2As shown, a yellow alert (Level III) is triggered when the predicted value is ≥70%, an orange alert (Level II) is triggered when the predicted value is ≥80%, and a red alert (Level I) is triggered when the predicted value is ≥90%. The model prediction accuracy must meet the requirements of MAPE ≤5% and F1 value ≥86%. Simultaneously, for different categories, the dynamic risk threshold corresponding to the target category load data can be used, and the predicted risk value of each category can be compared with its corresponding dynamic risk threshold to obtain the category risk level. Based on the category risk level corresponding to each target category load data, comprehensive early warning information is generated, and corresponding recommended measures are generated based on the early warning information. For example, based on information such as the early warning level, transformer number, predicted load rate curve, and list of recommended measures, the system automatically pushes the information to the corresponding responsible person according to the early warning level. For example, for a Level I early warning, an emergency work order is generated simultaneously, and recommended measures include: dispatching mobile energy storage, initiating load transfer, and temporary capacity expansion. Maintenance personnel must respond within 2 hours and backfill the handling results in the system to form a closed-loop record.

[0050] like Figure 6 The image shows the final output warning interface, which includes the transformer area number, current date, forecast time window, load composition analysis, and suggested measures. Simultaneously, after generating the warning information, the execution result of the warning information is obtained, and the dynamic risk threshold is optimized based on the execution result. That is, this step is the end point of the process, and its feedback data (such as whether an actual overload occurred and whether the response was timely) will be used for model backtesting and threshold optimization, supporting the dynamic adjustment of subsequent warning strategies.

[0051] Simultaneously, after outputting a status warning, the system can perform a lifespan assessment of the equipment based on the load rate. The lifespan assessment uses FEQA (Equivalent Accelerated Aging Factor) and FAA (Accelerated Aging Factor) to convert short-term temperature changes into long-term effects of insulation aging. For example, when the load rate is consistently 90%, FEQA = 3.2 × normal value → daily lifespan loss = 3.2 times that of normal operation → annual lifespan loss increases by 220% → remaining insulation lifespan is shortened by 67% (i.e., equipment with a lifespan of 20 years will fail within 6.6 years). When the predicted FEQA for the next 3 days is >2.5, the system automatically recommends "temporary capacity expansion" instead of "routine inspection." At the same time, lifespan loss data feeds back into the temperature model iteration (e.g., correcting the time constant, reducing the aging prediction error from ±15% to ±8%).

[0052] In this embodiment, the IEEE 33-node system is used as the standard test platform for the model constructed above. Its topology includes 33 nodes and 32 branches, configured with an HPLC+4G dual-channel communication architecture, achieving a transmission latency of <500ms. Against the backdrop of increasing distributed power penetration, the system provides a standardized scenario for the collaborative optimization of photovoltaic energy storage.

[0053] The distribution transformer load change trend (historical 24h + predicted 48h) shows a typical "double peak" characteristic during the summer peak electricity consumption period: The load factor during the morning peak (08:00-10:00) was 82%, mainly affected by the start-up of commercial electricity consumption; The evening peak (18:00-20:00) reaches 95% of the peak value (36th hour), with a significant superposition effect with residential electricity consumption; A critical point of 70% is reached within 60 hours, corresponding to the nighttime off-peak period.

[0054] The mechanism by which temperature affects load fluctuations is as follows: for every 1°C increase in temperature, the load rate increases by 0.8% to 1.2% (Pearson correlation coefficient r = 0.82). This correlation is verified by the DAENs model, which predicts a MAE of only 1.28%, which is more than 50% higher than the traditional BPNNs method.

[0055] The temperature model uses an iterative algorithm to convert electrical parameters (load rate) into thermal parameters (hot spot temperature), overcoming the limitations of traditional early warning systems that rely solely on current thresholds. For example, when the load rate increases from 80% to 90% (an increase of 10%), the hot spot temperature rises by 12°C. Although overcurrent protection (typically 120% of rated current) is not triggered at this point, it is already close to the critical withstand value of the insulation material (approximately 105°C for Class A insulation). The 24-hour iterative calculation simulates the dynamic impact of diurnal temperature variations and load fluctuations on temperature, avoiding misjudgments caused by static thresholds (such as the difference in risk between short-term overload in winter and continuous high-temperature overload in summer).

[0056] Please refer to Figure 7 This embodiment provides an electronic device, which further includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the distribution transformer short-term overload early warning method based on low-pressure meter data as described above.

[0057] In summary, this invention discloses a method and device for short-term overload early warning of distribution transformers based on low-voltage meter data. Compared to the 30-minute data sampling method used in SCADA systems, the method based on low-voltage meter data collection can more accurately reflect load fluctuation trends, avoid the peak missed detection problem of traditional 30-minute data, and improve the prediction accuracy to over 85%. When processing historical electricity data, after calculating the risk prediction value for the target period by extracting load data, meteorological correlation characteristics, and event characteristics from historical electricity data, a dynamic risk threshold is obtained according to the scenario type corresponding to the target period. This allows for the acquisition of different dynamic risk thresholds for different scenarios. The risk prediction value is then compared with the dynamic risk threshold, reducing false alarms and missed alarms, thereby improving the accuracy of early warning of power grid fluctuation peaks. Utilizing only existing meter data requires no additional hardware investment, reducing deployment costs and achieving a highly efficient early warning system with zero hardware investment. The 3-day advance warning function enables the operation and maintenance team to take timely preventive measures, such as dispatching mobile energy storage devices or initiating load transfer, reducing the risk of equipment damage and power outages. Furthermore, load composition analysis provides targeted suggestions for operation and maintenance, improving operation and maintenance efficiency and resource utilization. Meanwhile, the overload early warning model analyzes "how dangerous the overload is", and the life loss assessment analyzes "how much resource needs to be invested to solve the danger". Together, these two aspects upgrade the invention from a "simple alarm tool" to a "transformer health management platform".

[0058] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for early warning of short-term overload of distribution transformers based on low-voltage meter data, characterized in that, include: Historical electricity data is acquired by collecting readings from low-voltage meters at preset time intervals. Extract load data, meteorological correlation features, and event features from the historical electricity data; The risk prediction value for the target period is calculated based on the load data, meteorological correlation characteristics, and event characteristics. A dynamic risk threshold is obtained based on the scenario type corresponding to the target time period; The risk level is obtained by comparing the predicted risk value with the dynamic risk threshold, and an early warning message is generated based on the risk level.

2. The method for early warning of short-term overload of distribution transformers based on low-voltage meter data according to claim 1, characterized in that, The extraction of load data, meteorological correlation features, and event features from the historical electricity data includes: The historical electricity data is classified using a clustering algorithm to obtain target-classified load data, as well as the target meteorological correlation features and target event features corresponding to the target-classified load data; The target load data includes residential load data, agricultural load data, and commercial load data.

3. The method for early warning of short-term overload of distribution transformers based on low-voltage meter data according to claim 2, characterized in that, Also includes: Calculate the predicted classification risk value corresponding to each of the target classification load data; Obtain the classification dynamic risk threshold corresponding to each of the target classification load data; The classification risk level is obtained by comparing each of the predicted risk values ​​for the classification with its corresponding dynamic risk threshold. Comprehensive early warning information is generated based on the classification risk level corresponding to each of the target classification load data.

4. The method for early warning of short-term overload of distribution transformers based on low-voltage meter data according to claim 1, characterized in that, The risk prediction value for the target time period calculated based on the load data, meteorological correlation characteristics, and event characteristics includes: ； in, This is the risk prediction value; the average load rate is obtained from the load data. and load growth rate The temperature correction coefficient is obtained through the meteorological correlation characteristics. Holiday factors are obtained through the event characteristics. .

5. The method for early warning of short-term overload of distribution transformers based on low-voltage meter data according to claim 4, characterized in that, The temperature correction coefficient is obtained through the meteorological correlation features. include: ； in, The elastic coefficient; Real-time ambient temperature; This is the reference temperature.

6. The method for early warning of short-term overload of distribution transformers based on low-voltage meter data according to claim 4, characterized in that, The step of obtaining the dynamic risk threshold based on the scenario type corresponding to the target time period includes: The basic risk threshold is obtained based on the scenario type corresponding to the target time period; Determine whether there is a sudden rise in temperature or / and a sudden change in load growth rate during the target period. If so, reduce the basic risk threshold to obtain the dynamic risk threshold.

7. The method for early warning of short-term overload of distribution transformers based on low-voltage meter data according to claim 1, characterized in that, After acquiring historical electricity data, the process also includes: The historical electrical data is processed by removing outliers, filling in missing values, and normalizing.

8. The method for early warning of short-term overload of distribution transformers based on low-voltage meter data according to claim 1, characterized in that, After generating the early warning information based on the risk level, the process also includes: Obtain the execution result of the aforementioned warning information; The dynamic risk threshold is optimized based on the execution results.

9. A method for early warning of short-term overload of distribution transformers based on low-voltage meter data according to claim 1, characterized in that, The dynamic risk threshold includes at least two different levels of sub-thresholds; The step of comparing the predicted risk value with the dynamic risk threshold to obtain the risk level includes: The predicted risk value is compared with each of the sub-thresholds in turn to obtain the highest risk as the risk level.

10. An electronic device, the electronic device further comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements each step of the distribution transformer short-term overload early warning method based on low-pressure meter data as described in any one of claims 1-9.

Images