A power grid operation risk prediction method, device, equipment and system

Abstract

The present application relates to the technical field of data processing, and particularly relates to a power grid operation risk prediction method, device, equipment and system, which is used for reducing the influence of abnormal fluctuation in historical load data on the accuracy of a prediction model. The method comprises the following steps: obtaining load data of power grid equipment in multiple sub-areas in a target area in a historical operation time period; the load data is used to represent the power grid equipment operation load state; dividing the load data into multiple segments, and determining an abnormal segment from the multiple segments; the abnormal segment is a segment with abnormal fluctuation of the load data; reducing the weight of the load data in the abnormal segment in the training process; training a power grid risk prediction model based on the load data with the adjusted weight, and predicting the power grid operation risk in the target area based on the trained power grid risk prediction model.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a method, apparatus, equipment and system for predicting power grid operation risks. Background Technology

[0002] With the continuous growth of electricity demand and the increasing complexity of power grid structures, the safe and stable operation of the power grid faces severe challenges. Especially during peak electricity consumption periods or under extreme weather conditions, sharp load fluctuations can trigger operational risks. Therefore, power grid dispatch centers urgently need efficient and accurate risk prediction methods to ensure power supply reliability. In existing technologies, time series prediction models such as the Autoregressive Integrated Moving Average (ARIMA) model are commonly used to model and analyze power grid load data, thereby achieving early warning and assessment of operational risks. However, historical load data often contains abnormal fluctuations caused by unforeseen factors such as equipment failures and sudden weather changes. These abnormal data interfere with the model training process, weakening its ability to capture normal operating patterns, resulting in generally poor accuracy and reliability of prediction results. Therefore, how to reduce the impact of abnormal fluctuations in historical load data on the accuracy of prediction models has become an urgent technical problem to be solved. Summary of the Invention

[0003] To mitigate the impact of abnormal fluctuations in historical load data on the accuracy of prediction models, this invention aims to provide a method for predicting power grid operation risks. The specific technical solution adopted is as follows:

[0004] Obtain load data of power grid equipment in multiple areas within the target region during historical operating periods; the load data is used to characterize the operating load status of power grid equipment.

[0005] The load data is divided into multiple segments, and abnormal segments are identified from these segments; abnormal segments are those where the load data fluctuates abnormally.

[0006] Reduce the weight of load data within outlier segments during the training process;

[0007] A power grid risk prediction model is trained based on load data with adjusted weights, and the power grid operation risk in the target area is predicted based on the trained power grid risk prediction model.

[0008] In one possible implementation, the method further includes: constructing a load curve based on load data and corresponding running time; determining the starting point of the load curve as the first segmentation point; determining the segmentation evaluation index for each data point in the load curve sequentially based on time order, and determining data points whose segmentation evaluation index is greater than a first preset threshold as segmentation points; wherein, the segmentation evaluation index is determined based on the difference between the maximum and minimum values ​​of the load data in the current interval, and the average absolute difference between the load value of the current data point and the load value of each other data point in the current interval; the current interval is the load data interval between the current data point and the previous segmentation point; and dividing the load data into multiple segments based on the segmentation points.

[0009] In one possible implementation, the method further includes: determining the change value and fluctuation density of each segment; the change value is used to characterize the trend and magnitude of load data change in the segment; the fluctuation density is used to characterize the frequency and severity of load data fluctuations within the segment; traversing all segments to determine multiple candidate segment sets, wherein the segments in each candidate segment set are continuous segments, the number of segments is greater than a second preset threshold, and the change values ​​of each segment have the same sign; and the segments in the candidate segment set with the largest average fluctuation density among the multiple candidate segment sets are taken as abnormal segments.

[0010] In one possible implementation, the method further includes: determining a first difference in load data at the two ends of a segment, the average amplitude of each peak in the segment, and the average amplitude of each trough in the segment; determining the sign of the first difference as the sign of the change value of the segment, and determining the absolute value of the change value of the segment as the product of the absolute value of the first difference and the average amplitude; wherein, when the first difference is positive, the average amplitude is the average amplitude of each peak in the segment; when the first difference is negative, the average amplitude is the average amplitude of each trough in the segment. The average amplitude; when the first difference is positive, the fluctuation density of a segment is determined based on the number of peaks, duration, and absolute value of change in a segment; the fluctuation density is positively correlated with the number of peaks in a segment, negatively correlated with the duration of a segment, and positively correlated with the absolute value of change; when the first difference is negative, the fluctuation density of a segment is determined based on the number of troughs, duration, and absolute value of change in a segment; the fluctuation density is positively correlated with the number of troughs in a segment, negatively correlated with the duration of a segment, and positively correlated with the absolute value of change.

[0011] In one possible implementation, the method further includes: if the time periods corresponding to the abnormal segments of power grid equipment in multiple areas under a power supply station overlap, then calculate the difference between the overlap time length and the fluctuation density of the overlapping abnormal segments; based on the difference between the overlap time length and the fluctuation density, determine a first comprehensive score; if the first comprehensive score is greater than a third threshold, then determine that a power supply station has an anomaly, and adjust the fluctuation density of the overlapping abnormal segments to a target fluctuation density, where the target fluctuation density is the average value of the fluctuation density of the overlapping abnormal segments.

[0012] In one possible implementation, the method further includes: for one of the overlapping anomalous segments, calculating a first parameter between the anomalous segment and other anomalous segments in the overlapping anomalous segments; the first parameter is the ratio of the overlap time length between the two anomalous segments to a first value, and the first value is the absolute value of the difference in fluctuation density between the two anomalous segments plus one; summing the first parameter between the anomalous segment and other anomalous segments in the overlapping anomalous segments to obtain a first sum; normalizing the first sum to obtain the anomalous evaluation values ​​of the remaining areas of the anomalous segment; and taking the average of the anomalous evaluation values ​​of the remaining areas of all anomalous segments in the overlapping anomalous segments as the first comprehensive score.

[0013] In one possible implementation, the method further includes: calculating the ratio of the volatility density of an outlier segment to the maximum volatility density of the region to which the outlier segment belongs; determining a weight adjustment value for an outlier segment based on the ratio, wherein the weight adjustment value is negatively correlated with the volatility density of the outlier segment and the weight adjustment value is between 0 and 1; and adjusting the weights of the load data within an outlier segment during the training process based on the weight adjustment value.

[0014] The present invention also provides a power grid operation risk prediction device, which includes: an acquisition unit and a processing unit.

[0015] The acquisition unit is used to acquire load data of power grid equipment in multiple areas within the target region during historical operating time periods; the load data is used to characterize the operating load status of power grid equipment.

[0016] The processing unit is used to divide the load data into multiple segments and identify the abnormal segments from these segments; the abnormal segments are those with abnormal fluctuations in the load data.

[0017] The processing unit is also used to reduce the weight of load data in abnormal segments during the training process;

[0018] The processing unit is also used to train a power grid risk prediction model based on the weighted load data, and to predict the power grid operation risk in the target area based on the trained power grid risk prediction model.

[0019] The present invention also provides a power grid operation risk prediction device, which includes: a memory and a processor; the memory is used to store computer execution instructions, and the processor is connected to the memory via a bus; when the power grid operation risk prediction device is running, the processor executes the computer execution instructions stored in the memory, so that the power grid operation risk prediction device performs the above-mentioned power grid operation risk prediction method.

[0020] The present invention also provides a power grid operation risk prediction system, the system comprising: power grid equipment for supplying power to multiple areas within a target area, and power grid operation risk prediction equipment for performing the above-described power grid operation risk prediction method.

[0021] This invention offers the following advantages: By acquiring historical load data, identifying abnormal segments and adjusting their weights accordingly, and then using the optimized data to train a prediction model, this application constructs a systematic power grid operation risk prediction mechanism. This mechanism effectively purifies the dataset used for model training, significantly reducing the interference of abnormal data caused by occasional factors such as equipment failures and sudden weather changes on the prediction model training process. This allows the model to focus more on learning the inherent laws and steady-state characteristics of the power grid under normal operating conditions, thereby fundamentally improving the accuracy and reliability of the final risk prediction results and providing more trustworthy data support for power grid dispatching decisions. Attached Figure Description

[0022] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating a power grid operation risk prediction method provided in one embodiment of the present invention. Figure 1 ;

[0024] Figure 2 A flowchart illustrating a power grid operation risk prediction method provided in one embodiment of the present invention. Figure 2 ;

[0025] Figure 3 A flowchart illustrating a power grid operation risk prediction method provided in one embodiment of the present invention. Figure 3 ;

[0026] Figure 4 A flowchart illustrating a power grid operation risk prediction method provided in one embodiment of the present invention. Figure 4 ;

[0027] Figure 5 A flowchart illustrating a power grid operation risk prediction method provided in one embodiment of the present invention. Figure 5 . Detailed Implementation

[0028] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following detailed description, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a power grid operation risk prediction method proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0030] The specific scheme of the power grid operation risk prediction method provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0031] Please see Figure 1 It illustrates a flowchart of a power grid operation risk prediction method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes the following steps:

[0032] Step 101: Obtain load data of power grid equipment in multiple areas within the target region during historical operating periods.

[0033] Among them, load data is used to characterize the operating load status of power grid equipment.

[0034] As one possible implementation, this application utilizes the automated system of the power grid dispatch center and remote terminal units or smart meters deployed in various substations to continuously and synchronously collect load data from all power grid equipment within the target area at preset fixed time intervals. This load data is time-varying sequential data, used to quantify the operating load status of the corresponding power grid equipment at various historical moments. The load data can be stored on a corresponding data platform, such as a cloud server, for subsequent intelligent analysis and model training.

[0035] Step 102: Divide the load data into multiple segments.

[0036] In this step, based on the fluctuation characteristics of the load data, the continuous time series data is divided into multiple segments with different operating state characteristics.

[0037] As one possible implementation, this application sorts the load data of each power grid device in chronological order; determines the fluctuation characteristics of the load data in chronological order; and then divides the load data with large fluctuation characteristics into different segments of load data.

[0038] It should be noted that in the daily operation of power grid equipment, there are precise monitoring systems and power distribution processes. Furthermore, the daily power demand (load) of the power grid is relatively stable, and power is allocated to various electrical devices according to a pre-set demand plan. This quantitative allocation mechanism can maintain the stability of the power grid load. Therefore, ideally, if power distribution is stable and load demand is constant, the load data curve of the power grid equipment should show a stable state with minimal fluctuations. This stability is an important indicator for evaluating power distribution efficiency and equipment operating status. When relatively large fluctuations occur after a period of stability, it may be due to external factors or anomalies in the power grid operation. Further analysis of the specific manifestations of the fluctuations is needed to determine the possibility of the anomaly and its causes, and to address these issues to ensure the normal operation of the power grid. Therefore, this application can divide the load data curve of the power grid equipment into several segments based on its characteristic performance. Each segment may represent a specific power grid operating state or load condition.

[0039] Step 103: Identify the abnormal segments from multiple segments.

[0040] Among them, the abnormal segment is the segment of load data with abnormal fluctuations.

[0041] In this step, by using the dynamic fluctuation characteristics of load data, data segments that characterize abnormal power grid operation are identified from all data segments, thereby accurately locating the source of risk.

[0042] In one possible implementation, this application may first calculate the fluctuation characteristics of each segment, and then compare the difference between the fluctuation characteristics of each segment and the adjacent segments. If the difference is large, it indicates that the segment does not conform to the normal fluctuation of the load data, and the segment is regarded as an abnormal segment.

[0043] Step 104: Reduce the weight of load data in abnormal segments during the training process.

[0044] It should be noted that after identifying the abnormal segments, this application adjusts the weight of load data within the lower abnormal segments during the training process. This reduces the impact of load data within the abnormal segments on the training process, enabling the trained model to more accurately predict power grid operation risks under normal conditions and avoiding interference from abnormal data. Furthermore, this application lowers the weight of load data within the abnormal segments during the training process, rather than directly deleting the data from these abnormal segments. This allows for better utilization of all the information contained in the historical data and protects the temporal integrity of the dataset.

[0045] Step 105: Train the power grid risk prediction model based on the load data after weight adjustment.

[0046] Optionally, in this step, the historical load data sequence with adjusted weights is used as training samples and input into the prediction model. The prediction model adopts a modeling architecture that considers time-series dependencies, capturing the dynamic characteristics of the load data in the time dimension through its internal parameters. During the model training phase, a weighted loss function is constructed as the optimization objective. This loss function incorporates the adjusted weights as the importance coefficient of each data point when calculating the difference between the model's predicted values ​​and the actual observed values. For data points with lower weights, their contribution to the loss function is reduced accordingly; while for data points with higher weights, their contribution remains larger. Based on this weighted loss function, a gradient descent-like optimization algorithm is used to iteratively update the model parameters, allowing the model to focus more on the operational patterns reflected in the normal segmented data with higher weights during training, while reducing the impact of abnormal segmented data on model parameter updates.

[0047] Step 106: Predict the power grid operation risk in the target area based on the trained power grid risk prediction model.

[0048] Optionally, after training the power grid risk prediction model, this application can obtain real-time power grid equipment load data within the target area and input it into the trained power grid risk prediction model so that the model can predict the power grid operation risk of the target area in the future.

[0049] Based on the above technical solution, this application constructs a systematic power grid operation risk prediction mechanism through a complete process: acquiring historical load data, identifying abnormal segments and selectively reducing their weights, and then using the optimized data to train the prediction model. This mechanism can comprehensively purify the dataset used for model training, significantly reducing the interference of abnormal data caused by occasional factors such as equipment failures and sudden weather changes on the prediction model training process. In this way, the model can focus more on learning the inherent laws and steady-state characteristics of the power grid under normal operating conditions, reducing the impact of abnormal fluctuations in historical load data on the accuracy of the prediction model. This fundamentally improves the accuracy and reliability of the final risk prediction results, providing more reliable data support for power grid dispatching decisions.

[0050] like Figure 2 As shown, in one possible implementation, the process of dividing the load data into multiple segments in step 102 above can be specifically implemented through the following steps:

[0051] Step 201: Construct a load curve based on load data and corresponding running time.

[0052] It should be noted that the load data obtained in this application is data continuously and synchronously collected at preset fixed time intervals, therefore the load data has significant time characteristics. Optionally, this application can construct a coordinate system with time as the horizontal axis and the load data value as the vertical axis based on the collection time of each load data, and map each load data to the coordinate system according to the corresponding collection time and fit it to obtain the load curve.

[0053] Step 202: Determine the starting point of the load curve as the first segmentation point.

[0054] Step 203: Determine the segmented evaluation index for each data point in the load curve in chronological order, and determine the data points whose segmented evaluation index is greater than the first preset threshold as segment points.

[0055] The segmented evaluation index is determined based on the difference between the maximum and minimum values ​​of the load data in the current interval, and the average absolute difference between the load value of the current data point and the load value of every other data point in the current interval; the current interval is the load data interval between the current data point and the previous segment point.

[0056] Optionally, for each data point on the load data curve of the power grid equipment in each region (i.e., the data point corresponding to the load data in the load curve), an evaluation index for the curve is formed by all data points between it and the previous segment point. When traversing to this data point, if the evaluation index between it and the previous segment point is greater than a preset threshold, then the current data point is considered to be a segment point to separate the curves on both sides. That is, the two sides of this data point represent two different operating stages of the power grid equipment. The segmentation evaluation index for each data point is determined based on the changing trend and fluctuation amplitude of the data points on the curve.

[0057] For example, segmented evaluation indicators Satisfy the following formula:

[0058]

[0059] In the formula, Indicates the first Segmented evaluation indicators for each data point; They represent the first The maximum and minimum values ​​of the load data of all data points between each data point and the previous segment point; Indicates the first The number of all data points between each data point and the previous segmentation point. Indicates the first Among all data points between the nth data point and the previous segmentation point, the nth data point... Load data for each data point. It should be noted that, in order to avoid... The value is 0. In this application, after determining a segmentation point, the segmentation evaluation index is not calculated for the first data point after that segmentation point.

[0060] Based on the above formula, when the first The difference between the maximum and minimum load data of all data points between each data point and the previous segment point. The larger the value, the higher the segmented evaluation index for that data point; Indicates the first The average of the differences in load data between the current data point and the previous segment point is the mean of the load data for all data points. The larger the value, the better. The greater the fluctuation range of the curve between a data point and the previous segmentation point, the more unstable it is, and the higher the segmentation evaluation index of that data point.

[0061] As an example, the first preset threshold is 0.7, when When it exceeds this threshold, it is considered that the first... Each data point represents a segmentation point in the load data curve of the power grid equipment in the current area.

[0062] Step 204: Divide the load data into multiple segments based on the segmentation points.

[0063] Optionally, after determining the segmentation points, the load data between every two adjacent segmentation points can be considered as a segment, resulting in multiple segments.

[0064] Based on the above technical solution, this application constructs a load curve and uses a segmented evaluation index that integrates the global range and local mean absolute difference to automatically divide the data into segments. This intelligently and accurately identifies key inflection points in the load curve that represent significant changes in operating status, thereby dividing continuous load data into a series of physically meaningful segments with different fluctuation characteristics. This not only lays a solid foundation for subsequent anomaly analysis but also overcomes the subjectivity and inadequacy of manual or fixed-window segmentation, improving the automation and robustness of the entire method under different operating scenarios.

[0065] like Figure 3 As shown, in one possible implementation, the process of determining the abnormal segment from multiple segments in step 103 above can be achieved through the following steps:

[0066] Step 301: Determine the change value and fluctuation density of each segment.

[0067] Among them, the change value is used to characterize the trend and magnitude of load data changes in segments; the fluctuation density is used to characterize the frequency and severity of load data fluctuations within segments.

[0068] It should be noted that for each segment of the load data curve of the power grid equipment, due to the significant differences in electricity demand at different times during daily electricity consumption, each segment may exhibit different trends in load data, specifically showing an increase or decrease in load, i.e., peak or trough electricity consumption. Therefore, the change value of each segment is introduced to more accurately grasp the dynamic characteristics of the power grid load at different times. The change value is distinguished by positive or negative values: a positive change value indicates an upward trend in the power grid load during that time period, i.e., an increase in electricity demand; while a negative change value indicates a downward trend in the power grid load, meaning a decrease in electricity demand. Furthermore, the absolute value of the change value also reflects the amplitude of load fluctuations, i.e., the magnitude of the fluctuation. The larger the amplitude, the more significant the load change during that time period, potentially corresponding to more drastic fluctuations in electricity demand. In addition, when the power grid equipment in a region is operating normally, the load data curve of the power grid equipment should show a relatively stable upward or downward trend. This means that within each segment, load data fluctuations should be relatively small and the trend should be gradual, reflecting a stable and orderly increase or decrease in electricity demand in that area under normal conditions. When the load data curve fluctuates abnormally frequently within a certain segment, this is often a warning signal, indicating that the load in that area may be abnormal. Such frequent fluctuations can be caused by various factors, including equipment failure, abnormal weather conditions, and sudden changes in electricity consumption behavior. Therefore, based on the degree of fluctuation within each segment, its fluctuation density is determined to reflect a comprehensive measure of the frequency and amplitude of load data fluctuations within a certain segment.

[0069] Optionally, the process of determining the change value includes: determining the first difference in load data at the two ends of a segment, the average amplitude of each peak in a segment, and the average amplitude of each trough in a segment; determining the sign of the first difference as the sign of the change value of a segment, and determining the absolute value of the change value of a segment as the product of the absolute value of the first difference and the average amplitude. Wherein, when the first difference is positive, the average amplitude is the average amplitude of each peak in a segment; when the first difference is negative, the average amplitude is the average amplitude of each trough in a segment.

[0070] Optionally, the process of determining the fluctuation density includes: when the first difference is positive, determining the fluctuation density of a segment based on the number of peaks, the duration of the segment, and the absolute value of the change value; the fluctuation density is positively correlated with the number of peaks in a segment, negatively correlated with the duration of the segment, and positively correlated with the absolute value of the change value; when the first difference is negative, determining the fluctuation density of a segment based on the number of troughs, the duration of the segment, and the absolute value of the change value; the fluctuation density is positively correlated with the number of troughs in a segment, negatively correlated with the duration of the segment, and positively correlated with the absolute value of the change value.

[0071] For example, the first The change value of each segment Satisfy the following formula:

[0072]

[0073] in, Indicates the first Load data for the last data point in each segment. Indicates the first Load data for the first data point of each segment The sign of the value indicates whether the segment is rising or falling; when When greater than zero, Indicates the first The average amplitude of the rise in all peaks of a segment indicates that the larger the average amplitude of all rising curves in that segment, the larger the absolute value of the change. When less than zero, Indicates the first The average amplitude of the decline in all troughs of a segment indicates that the larger the average amplitude of all declining curves in a segment, the larger the absolute value of the change in that segment.

[0074] For example, taking the first difference as a positive value, the... Fluctuation density of each segment Satisfy the following formula:

[0075]

[0076] in, Indicates the first The time length of each segment; Indicates the first The number of peaks in each segment; Indicates the first The absolute value of the change in each segment.

[0077] Based on fluctuation density The calculation formula, when the first The shorter the time length of each segment and the more peaks, the more it indicates that in the first segment... The more frequent the fluctuations within a segment, the higher the fluctuation density; and the larger the fluctuation amplitude, that is, the higher the fluctuation density. The larger the absolute value of the change value of each segment, the higher the fluctuation density.

[0078] Step 302: Traverse all segments and determine multiple candidate segment sets.

[0079] Among them, the segments in each candidate segment set are continuous segments, the number of segments is greater than the second preset threshold, and the change values ​​of each segment have the same sign.

[0080] It should be noted that, under normal circumstances, power grid load data exhibits natural periodic fluctuations in response to electricity demand at different times of the day. This fluctuation manifests as a regular rise and fall in load, reflected in data segments where adjacent segments show alternating positive and negative values ​​– a normal state for a healthy power grid operation. However, when persistent abnormal events occur (such as certain types of equipment failures), this natural alternation may be disrupted, causing the load to continuously change in the same direction across multiple segments, forming an out-of-synchronization anomaly. This application uses the identification of this continuous unidirectional change pattern as the primary criterion for screening potential anomalous segment sets.

[0081] Optionally, to determine candidate segments, this application uses the following method: A change factor s (initially zero) is introduced to record cases where the signs of changes in adjacent segments are opposite. All segments are traversed. When the sign of the change value of each segment is opposite to that of the previous segment, it indicates a significant change in the direction of load fluctuation between the two segments, and no action is taken. If the signs are the same, s is incremented by 1, and the traversal continues to the next segment. During the traversal, a threshold T is set (T is determined to be 4 based on power grid operation experience and data characteristics). When the change factor s reaches or exceeds T, the traversal ends, and the change factor s is reset to zero, indicating that there are abnormal segments in the currently analyzed segment set (not directly judging all segments as abnormal).

[0082] Step 303: Select the segments from the candidate segment set with the largest average fluctuation density among multiple candidate segment sets as abnormal segments.

[0083] Optionally, to further identify anomalous segments, the volatility density of all consecutive segments in the current set of segments where anomalous segments are possible can be calculated. The average value is used to identify several consecutive segments with the highest average fluctuation density, which are then designated as abnormal segments (this may be one or more segments). This process identifies the abnormal segments in the load data curves of the power grid equipment in each region.

[0084] Based on the above technical solution, this application quantifies the dynamic behavior of each segment by introducing two key features: change value and fluctuation density. It also filters abnormal segments based on the continuity of the change trend and the level of fluctuation density. This effectively distinguishes between normal periodic fluctuations of the power grid load and true abnormal fluctuations. The filtering process does not simply rely on a single threshold, but combines trend anomalies (continuous unidirectional changes) and fluctuation anomalies (high fluctuation density) for comprehensive judgment. This achieves more accurate and reliable identification of abnormal segments, significantly reduces the probability of false alarms and missed alarms, and enhances the pertinence of risk warning.

[0085] It should be noted that when all areas under the same power supply station are simultaneously identified as abnormal segments within a certain period of time, it indicates that the power grid failure in all areas may be caused by an abnormal operation of the power supply station. In this case, we need to uniformly correct the fluctuation density of all areas during the abnormal period of the power supply station to obtain more accurate weights in the ARIMA model, which can more realistically reflect the actual operation of the power grid and improve the accuracy of prediction.

[0086] like Figure 4 As shown, in one possible implementation, after identifying the abnormal segments from multiple segments in step 103 above, this application can further determine whether there is a power supply station anomaly, and uniformly update the fluctuation density of power grid equipment in multiple areas under the power supply station in the event of a power supply station anomaly. This can be achieved through the following steps:

[0087] Step 401: If the time periods corresponding to the abnormal segments of the power grid equipment in multiple areas under the power supply station overlap, calculate the difference between the length of the overlapping time and the fluctuation density of the overlapping abnormal segments.

[0088] Step 402: Determine the first comprehensive score based on the difference between the overlap time length and the fluctuation density.

[0089] Optionally, the process of determining the first comprehensive score includes: for one of the overlapping abnormal segments, calculating a first parameter between the abnormal segment and other abnormal segments in the overlapping abnormal segments; the first parameter is the ratio of the overlap time length between the two abnormal segments to a first value, and the first value is the absolute value of the difference in fluctuation density between the two abnormal segments plus one; summing the first parameters between the abnormal segment and other abnormal segments in the overlapping abnormal segments to obtain a first sum; normalizing the first sum to obtain the abnormal evaluation values ​​of the remaining areas of the abnormal segment; and taking the average of the abnormal evaluation values ​​of the remaining areas of all abnormal segments in the overlapping abnormal segments as the first comprehensive score.

[0090] As an example, the first The first comprehensive score for each abnormal segment Satisfy the following formula:

[0091]

[0092] in, This indicates the number of all zones under the same power supply station; Indicates the first The first abnormal segment and the remaining segments under the same power supply station The duration of abnormal segment overlap in each region; Indicates the first Fluctuation density of each abnormal segment; Indicates the remaining ones under the same power supply station The fluctuation density of the corresponding abnormal segments in each region.

[0093] Based on this formula, when the... If there are more and more abnormal segments in other areas under the same power supply station, and the more time overlaps between them, and the smaller the difference in their fluctuation density, then the higher the anomaly evaluation of the other areas of that abnormal segment, and the more likely it is an anomaly of the power supply station.

[0094] Step 403: If the first comprehensive score is greater than the third threshold, it is determined that there is an anomaly in the power supply station, and the fluctuation density of the overlapping abnormal segments is adjusted to the target fluctuation density.

[0095] The target volatility density is the average volatility density of the overlapping abnormal segments.

[0096] When the average of the anomaly evaluations for the remaining areas corresponding to the anomaly segments across all overlapping time zones exceeds the third threshold, a power station anomaly is considered to have occurred, and the corresponding anomaly segments across all zones are also attributed to power station anomalies. The fluctuation density of all anomaly segments across all zones caused by power station anomalies is uniformly adjusted to its average value to more consistently assess the impact of power station anomalies on the overall power grid operation and eliminate data biases that may be caused by power station anomalies. For example, the third threshold is set to 0.6.

[0097] Based on the above technical solution, this application detects cluster anomalies—multiple areas under the same power supply station experiencing anomalies at the same time—and accordingly corrects the fluctuation density of related anomaly segments. This effectively identifies global faults or disturbances originating from the power supply station itself (rather than individual area equipment). By uniformly correcting the fluctuation density, these data anomalies caused by common reasons are consistently handled in subsequent model training. This helps eliminate data bias at the system level, preventing multiple related anomalies from having a cumulative negative impact on the model, thereby improving the robustness of the prediction model in dealing with systemic risks.

[0098] like Figure 5 As shown, in one possible implementation, reducing the weight of the load data within the abnormal segments during the training process in step 104 above can be achieved through the following steps:

[0099] Step 501: Calculate the fluctuation density of the abnormal segment and the ratio of it to the maximum fluctuation density of the area to which the abnormal segment belongs.

[0100] Step 502: Based on the ratio, determine the weight adjustment value for the abnormal segment.

[0101] Among them, the weight adjustment value is negatively correlated with the fluctuation density of the abnormal segment, and the weight adjustment value ranges from 0 to 1.

[0102] As an example, the weight adjustment value for outlier segments. Satisfy the following formula:

[0103]

[0104] In the formula, Indicates the first The fluctuation density of each abnormal segment Indicates the first The maximum fluctuation density of the region where each anomaly segment is located. The higher the fluctuation density, the more significant the anomaly of the segment in the scene, and the smaller its weight, and the smaller its impact on the prediction result.

[0105] Step 503: Adjust the weights of the load data in the abnormal segment during the training process based on the weight adjustment value.

[0106] It should be noted that in the ARIMA model, the weights reflect the degree of influence of historical data and historical errors on the current value. This application uses the ARIMA model to predict the load data of power grid equipment based on all historical data points and their weights, obtaining highly accurate prediction results for power dispatch and resource allocation. Simultaneously, a feedback mechanism is established to continuously optimize the prediction model based on actual load conditions. Furthermore, load forecasting is a dynamic process; as new data is continuously generated and the model is continuously optimized, the prediction results will change. Therefore, it is necessary to regularly update the model and data to maintain the accuracy of the predictions, enabling a more scientific and reasonable response to the operational risks of power grid equipment and ensuring the safe and stable operation of the power grid.

[0107] Based on the above technical solution, this application determines the weight adjustment value by comparing the fluctuation density of the abnormal segment with the historical maximum fluctuation density of its corresponding region, achieving a dynamic and adaptive weight adjustment effect. This method directly correlates the magnitude of the weight adjustment with the "significance of the abnormality" of the abnormal segment in the current context; the more significant the abnormality (i.e., the closer it is to the historical extreme value), the greater the reduction in its weight. This differentiated approach, compared to a "one-size-fits-all" weight allocation strategy, can more precisely control the impact of abnormal data on the model, thereby achieving a better balance between noise suppression and preservation of effective information, further optimizing the model's training performance.

[0108] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0109] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for predicting power grid operation risks, characterized in that, The method includes: Obtain load data of power grid equipment in multiple areas within a target region during historical operating periods; the load data is used to characterize the operating load status of the power grid equipment. The load data is divided into multiple segments, and abnormal segments are identified from these segments; the abnormal segments are those where the load data fluctuates abnormally. Reduce the weight of the load data within the abnormal segments during the training process; A power grid risk prediction model is trained based on the load data with adjusted weights, and the power grid operation risk in the target area is predicted based on the trained power grid risk prediction model. The reduction of the weight of the load data within the abnormal segments during the training process includes: Calculate the ratio of the fluctuation density of an abnormal segment to the maximum fluctuation density of the area to which the abnormal segment belongs; Based on the ratio, a weight adjustment value for the abnormal segment is determined. The weight adjustment value is negatively correlated with the fluctuation density of the abnormal segment, and the value of the weight adjustment value is between 0 and 1. The weights of the load data within the abnormal segment are adjusted during the training process based on the aforementioned weight adjustment value. The step of identifying abnormal segments from the plurality of segments includes: Determine the change value and fluctuation intensity for each segment; the change value is used to characterize the trend and magnitude of load data changes in the segment; the fluctuation intensity is used to characterize the frequency and severity of load data fluctuations within the segment. Traverse all segments to determine multiple candidate segment sets. Each candidate segment set contains segments that are consecutive, have a number of segments greater than the second preset threshold, and have the same sign for the change value of each segment. The segment with the largest average fluctuation density among the multiple candidate segment sets is taken as the abnormal segment; The determination of the change value and fluctuation density of each segment includes: Determine the first difference in load data at the two ends of a segment, the average amplitude of each peak in the segment, and the average amplitude of each trough in the segment; The sign of the first difference is determined to be the sign of the change value of the segment, and the absolute value of the change value of the segment is determined to be the product of the absolute value of the first difference and the average value of the amplitude; wherein, when the first difference is positive, the average value of the amplitude is the average value of the amplitude of each peak of the segment; when the first difference is negative, the average value of the amplitude is the average value of the amplitude of each trough of the segment. When the first difference is positive, the fluctuation density of the segment is determined based on the number of peaks, the duration, and the absolute value of the change value of the segment; the fluctuation density is positively correlated with the number of peaks in the segment, negatively correlated with the duration of the segment, and positively correlated with the absolute value of the change value. When the first difference is negative, the fluctuation density of the segment is determined based on the number of troughs, the duration, and the absolute value of the change value of the segment. The fluctuation density is positively correlated with the number of troughs in the segment, negatively correlated with the duration of the segment, and positively correlated with the absolute value of the change value.

2. The power grid operation risk prediction method according to claim 1, characterized in that, The process of dividing the load data into multiple segments includes: Based on load data and corresponding runtime, construct a load curve; The starting point of the load curve is determined as the first segmentation point; Based on the time sequence, segmented evaluation indicators for each data point in the load curve are determined sequentially, and data points whose segmented evaluation indicators are greater than a first preset threshold are identified as segment points. The segmented evaluation indicators are determined based on the difference between the maximum and minimum values ​​of the load data in the current interval, and the average absolute difference between the load value of the current data point and the load value of every other data point within the current interval. The current interval is the load data interval between the current data point and the previous segment point. The load data is divided into multiple segments based on the segmentation points.

3. The power grid operation risk prediction method according to claim 1, characterized in that, After identifying the anomalous segments from the plurality of segments, the method further includes: If the time periods corresponding to the abnormal segments of the power grid equipment in multiple areas under a power supply station overlap, then calculate the difference between the length of the overlapping time and the fluctuation density of the overlapping abnormal segments. A first comprehensive score is determined based on the difference between the overlap time length and the fluctuation density. If the first comprehensive score is greater than the third threshold, it is determined that there is an anomaly in the power supply station, and the fluctuation density of the overlapping abnormal segments is adjusted to the target fluctuation density, which is the average value of the fluctuation density of the overlapping abnormal segments.

4. The power grid operation risk prediction method according to claim 3, characterized in that, The determination of the first comprehensive score based on the difference between the overlap time length and the fluctuation density includes: For one of the overlapping abnormal segments, calculate a first parameter between the one abnormal segment and the other abnormal segments in the overlapping abnormal segments; the first parameter is the ratio of the overlap time length between the two abnormal segments to a first value, and the first value is the absolute value of the difference in fluctuation density between the two abnormal segments plus one. The first parameter between the one abnormal segment and other abnormal segments in the overlapping abnormal segments is summed to obtain a first sum; The first sum is normalized to obtain the anomaly evaluation values ​​of the remaining areas of the anomaly segment. The average of the remaining area anomaly evaluation values ​​of all anomaly segments in the overlapping anomaly segments is taken as the first comprehensive score.

5. A power grid operation risk prediction device, characterized in that, The device includes: an acquisition unit and a processing unit. The acquisition unit is used to acquire load data of power grid equipment in multiple areas within the target region during historical operating time periods; the load data is used to characterize the operating load status of the power grid equipment. The processing unit is used to divide the load data into multiple segments and identify abnormal segments from the multiple segments; the abnormal segments are segments where the load data fluctuates abnormally. The processing unit is also used to reduce the weight of the load data in the abnormal segment during the training process; The processing unit is also used to train a power grid risk prediction model based on the load data after weight adjustment, and to predict the power grid operation risk in the target area based on the trained power grid risk prediction model. Specifically, the processing unit is further configured to: calculate the ratio of the fluctuation density of an abnormal segment to the maximum fluctuation density of the region to which the abnormal segment belongs; Based on the ratio, a weight adjustment value for the abnormal segment is determined. The weight adjustment value is negatively correlated with the fluctuation density of the abnormal segment, and the value of the weight adjustment value is between 0 and 1. The weights of the load data within the abnormal segment are adjusted during the training process based on the aforementioned weight adjustment value. The processing unit is further configured to: determine the change value and fluctuation density of each segment; the change value is used to characterize the load data change trend and amplitude of the segment; the fluctuation density is used to characterize the frequency and severity of load data fluctuations within the segment; traverse all segments to determine multiple candidate segment sets, wherein the segments in each candidate segment set are continuous segments, the number of segments is greater than a second preset threshold, and the change values ​​of each segment have the same sign; and select the segment in the candidate segment set with the largest average fluctuation density among the multiple candidate segment sets as the abnormal segment. The processing unit is further configured to: determine a first difference in load data at the two ends of a segment, the average amplitude of each peak in the segment, and the average amplitude of each trough in the segment; determine that the sign of the first difference is the sign of the change value of the segment, and determine that the absolute value of the change value of the segment is the product of the absolute value of the first difference and the average amplitude; wherein, when the first difference is positive, the average amplitude is the average amplitude of each peak in the segment; and when the first difference is negative, the average amplitude is the average amplitude of each trough in the segment. When the first difference is positive, the fluctuation density of the segment is determined based on the number of peaks, the duration, and the absolute value of the change value of the segment; the fluctuation density is positively correlated with the number of peaks in the segment, negatively correlated with the duration, and positively correlated with the absolute value of the change value. When the first difference is negative, the fluctuation density of the segment is determined based on the number of troughs, the duration, and the absolute value of the change value of the segment; the fluctuation density is positively correlated with the number of troughs in the segment, negatively correlated with the duration, and positively correlated with the absolute value of the change value.

6. A power grid operation risk prediction device, characterized in that, The device includes: a memory and a processor; the memory is used to store computer execution instructions, and the processor is connected to the memory via a bus; when the power grid operation risk prediction device is running, the processor executes the computer execution instructions stored in the memory, so that the power grid operation risk prediction device performs the power grid operation risk prediction method as described in any one of claims 1-4.

7. A power grid operation risk prediction system, characterized in that, The system includes: power grid equipment for supplying power to multiple areas within a target area, and power grid operation risk prediction equipment for performing the power grid operation risk prediction method as described in any one of claims 1-4.

Images