Machine learning based operational scenario quality automatic monitoring method, system and storage medium

Abstract

This invention relates to the field of power system operation and maintenance technology, and discloses a method, system, and storage medium for automatic monitoring of operation scenario quality based on machine learning. The method acquires multiple quality indicator data of a power grid operation scenario within a preset first time period; inputs the multiple quality indicator data into a pre-trained quality assessment model to obtain a comprehensive quality score for the power grid operation scenario at the current moment. The quality assessment model is trained based on multiple quality indicator data and their corresponding scenario state labels from historical time periods, and is used to learn the dynamic correlation between the multiple quality indicator data and the comprehensive quality score; based on the comprehensive quality score and a preset quality threshold, it determines whether the current quality state of the power grid operation scenario is abnormal; if so, it generates and outputs quality alarm information for the power grid operation scenario based on the comprehensive quality score and the quality threshold. Therefore, this invention can improve the accuracy and comprehensiveness of power grid operation scenario quality monitoring and assessment.

Description

Technical Field

[0001] This invention relates to the field of power system operation and maintenance technology, and in particular to a method, system and storage medium for automatic monitoring of operational scenario quality based on machine learning. Background Technology

[0002] The safe and stable operation of the power grid is of paramount importance, requiring continuous monitoring of the operational status of various scenarios, including power generation, transmission, and distribution. Currently, quality monitoring of power grid operations largely relies on manual methods.

[0003] Specifically, maintenance personnel typically need to monitor multiple different operating parameters simultaneously (such as voltage, current, and power) and rely on their personal experience and procedures to comprehensively judge whether the overall operating status of the scenario is normal. This approach has significant shortcomings: First, manual monitoring struggles to process and analyze massive amounts of high-frequency monitoring data in real time, resulting in delayed responses and failing to meet the requirement for rapid response to abnormal states. Second, human experience-based judgment is difficult to quantify precisely and cannot maintain consistent standards. For complex coupling relationships between multiple parameters, manual assessment is challenging, leading to inaccurate identification of abnormal states and potential misjudgments or missed diagnoses.

[0004] Therefore, it is particularly important to propose a technical solution to improve the accuracy and comprehensiveness of quality monitoring and evaluation in power grid operation scenarios. Summary of the Invention

[0005] This invention provides a machine learning-based automatic monitoring method, system, and storage medium for power grid operation scenario quality, which can improve the accuracy and comprehensiveness of power grid operation scenario quality monitoring and evaluation.

[0006] To address the aforementioned technical problems, the first aspect of this invention discloses a method for automatic monitoring of operational scenario quality based on machine learning, the method comprising: Acquire multiple quality indicator data for power grid operation scenarios within a preset first time period; The multiple quality index data are input into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment. The quality assessment model is trained based on the multiple quality index data and their corresponding scenario state labels over historical time periods and is used to learn the dynamic correlation between the multiple quality index data and the comprehensive quality score. Based on the comprehensive quality score and the preset quality threshold, it is determined whether the current quality status of the power grid operation scenario is abnormal. If so, then based on the comprehensive quality score and the quality threshold, quality alarm information for the power grid operation scenario is generated and output.

[0007] As an optional implementation, in the first aspect of the present invention, the step of inputting the multiple quality index data into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment includes: A sliding window sampling is performed on the multiple quality indicator data to obtain a local time series data block at the current moment. The local time series data block is used to represent the fluctuation of the multiple quality indicator data within a preset first time period. Trend features and fluctuation features are extracted from the local time-series data block. The trend features are used to represent the direction of change of the multiple quality index data within the preset first time period, and the fluctuation features are used to represent the stability of the multiple quality index data within the preset first time period. The trend features and the fluctuation features are determined as an enhanced feature set, and a unified feature representation is obtained through feature fusion operation; The unified representation of the features is input into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment.

[0008] As an optional implementation, in the first aspect of the present invention, the extraction of trend features and fluctuation features from the local time-series data block includes: For each quality index sequence in the local time series data block, the trend slope corresponding to the quality index sequence is obtained by linear fitting the quality index sequence. The trend slope is used to represent the average change of the corresponding quality index sequence in a unit time. Calculate the relative rate of change of the first and last values ​​of the quality index sequence within the local time-series data block. The relative rate of change of the first and last values ​​is used to represent the overall change range of the quality index sequence within the preset first time period. Calculate the standard deviation of the first-order difference sequence of the quality index sequence to obtain the trend fluctuation intensity, which is used to represent the stability of the rate of change of the quality index sequence; The trend slope, the relative rate of change of the first and last values, and the trend fluctuation intensity are determined as the trend characteristics of the quality index sequence. The ratio of the standard deviation to the mean of the quality index sequence within the local time-series data block is calculated to obtain the coefficient of variation, which is used to represent the relative dispersion of the quality index sequence. The difference between the maximum and minimum values ​​of the quality index sequence within the local time-series data block is calculated to obtain the range, which is used to represent the absolute fluctuation range of the quality index sequence. The coefficient of variation and the range are used to determine the fluctuation characteristics of the quality index sequence.

[0009] As an optional implementation, in the first aspect of the present invention, determining the trend features and the fluctuation features as an enhanced feature set, and obtaining a unified feature representation through feature fusion operation, includes: The trend feature and fluctuation feature of each quality index sequence are normalized to obtain standardized trend feature and standardized fluctuation feature. The normalization process is used to eliminate scale differences between features of different dimensions. Attention weights are assigned to the standardized trend feature and the standardized fluctuation feature respectively to obtain weighted trend feature and weighted fluctuation feature. The attention weights are used to characterize the degree of influence of the corresponding feature on the comprehensive quality score of the power grid operation scenario at the current moment. The weighted trend feature and the weighted fluctuation feature are concatenated to obtain a unified feature representation.

[0010] As an optional implementation, in the first aspect of the present invention, the step of inputting the unified representation of the features into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment includes: Principal component transformation is performed on the unified feature representation to obtain key feature representation. The principal component transformation is used to reduce the dimensionality and decorrelate the unified feature representation to extract core information characterizing the quality status of the power grid operation scenario. Extracting temporal correlation features from the key feature representation, the temporal correlation features are used to characterize the collaborative change pattern of different quality indicators within the preset first time period, specifically including obtaining them by calculating the correlation coefficients between different feature dimension sequences in the key feature representation; The key feature representation is fused with the temporal correlation feature to form the final feature vector, specifically by concatenating the temporal correlation feature as an additional dimension after the key feature representation; The final feature vector is input into the multilayer sensing network in the quality assessment model, and the multilayer sensing network performs nonlinear mapping to output the comprehensive quality score of the power grid operation scenario at the current moment.

[0011] As an optional implementation, in the first aspect of the present invention, determining whether the current quality status of the power grid operation scenario is abnormal based on the comprehensive quality score and a preset quality threshold includes: Based on the historical comprehensive quality scores of multiple historical moments within a preset second time period, a quality score sequence is formed. The quality score sequence is used to characterize the changing trend of the quality status of the power grid operation scenario over time. The time length of the preset second time period is longer than that of the preset first time period. Obtain a preset static quality threshold, wherein the static quality threshold is a fixed numerical reference value; A dynamic quality threshold is calculated based on the quality scoring sequence. The dynamic quality threshold is used to reflect the historical normal fluctuation range of the quality status of the power grid operation scenario within the preset second time period. Specifically, it is obtained by calculating the statistics of the quality scoring sequence. Extract the score subsequence located within the preset first time period from the quality score sequence; Linear fitting is performed on the scoring subsequence to obtain the scoring trend; Based on the comprehensive quality score, the static quality threshold, the dynamic quality threshold, and the score change trend, a multi-condition logical judgment is performed to determine whether the current quality status of the power grid operation scenario is abnormal.

[0012] As an optional implementation, in the first aspect of the present invention, the step of performing multi-condition logical judgment based on the comprehensive quality score, the static quality threshold, the dynamic quality threshold, and the score change trend to determine whether the current quality status of the power grid operation scenario is abnormal includes: Determine whether the overall quality score is less than the static quality threshold; If it is less than, then the current quality status of the power grid operation scenario is determined to be an abnormal state; If it is not less than, then determine whether the comprehensive quality score is less than the dynamic quality threshold; If it is less than, then determine whether the trend of the score change is a negative trend; If so, the current quality status of the power grid operation scenario is determined to be an abnormal state.

[0013] As an optional implementation, in the first aspect of the present invention, generating and outputting quality alarm information for the power grid operation scenario based on the comprehensive quality score and the quality threshold includes: The quality anomaly level and anomaly type are determined based on the comprehensive quality score, the static quality threshold, and the dynamic quality threshold. Based on the anomaly type, the quality score sequence, and the quality anomaly level, generate alarm content containing anomaly diagnostic information; The alarm content, the identification information of the power grid operation scenario, and the current time are combined to generate and output quality alarm information for the power grid operation scenario.

[0014] A second aspect of this invention discloses an automatic monitoring system for operational scenario quality based on machine learning, the system comprising: The acquisition module is used to acquire multiple quality indicator data of the power grid operation scenario within a preset first time period; The evaluation module is used to input the multiple quality index data into a pre-trained quality evaluation model to obtain the comprehensive quality score of the power grid operation scenario at the current moment. The quality evaluation model is trained based on the multiple quality index data and their corresponding scenario state labels over historical time periods and is used to learn the dynamic correlation between the multiple quality index data and the comprehensive quality score. The judgment module is used to determine whether the current quality status of the power grid operation scenario is abnormal based on the comprehensive quality score and the preset quality threshold. If so, the generation module is used to generate and output quality alarm information for the power grid operation scenario based on the comprehensive quality score and the quality threshold.

[0015] As an optional implementation, in the second aspect of the present invention, the evaluation module inputs the multiple quality index data into a pre-trained quality evaluation model to obtain the comprehensive quality score of the power grid operation scenario at the current moment, specifically in the following ways: A sliding window sampling is performed on the multiple quality indicator data to obtain a local time series data block at the current moment. The local time series data block is used to represent the fluctuation of the multiple quality indicator data within a preset first time period. Trend features and fluctuation features are extracted from the local time-series data block. The trend features are used to represent the direction of change of the multiple quality index data within the preset first time period, and the fluctuation features are used to represent the stability of the multiple quality index data within the preset first time period. The trend features and the fluctuation features are determined as an enhanced feature set, and a unified feature representation is obtained through feature fusion operation; The unified representation of the features is input into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment.

[0016] As an optional implementation, in a second aspect of the present invention, the evaluation module extracts trend features and fluctuation features from the local time-series data block in the following specific ways: For each quality index sequence in the local time series data block, the trend slope corresponding to the quality index sequence is obtained by linear fitting the quality index sequence. The trend slope is used to represent the average change of the corresponding quality index sequence in a unit time. Calculate the relative rate of change of the first and last values ​​of the quality index sequence within the local time-series data block. The relative rate of change of the first and last values ​​is used to represent the overall change range of the quality index sequence within the preset first time period. Calculate the standard deviation of the first-order difference sequence of the quality index sequence to obtain the trend fluctuation intensity, which is used to represent the stability of the rate of change of the quality index sequence; The trend slope, the relative rate of change of the first and last values, and the trend fluctuation intensity are determined as the trend characteristics of the quality index sequence. The ratio of the standard deviation to the mean of the quality index sequence within the local time-series data block is calculated to obtain the coefficient of variation, which is used to represent the relative dispersion of the quality index sequence. The difference between the maximum and minimum values ​​of the quality index sequence within the local time-series data block is calculated to obtain the range, which is used to represent the absolute fluctuation range of the quality index sequence. The coefficient of variation and the range are used to determine the fluctuation characteristics of the quality index sequence.

[0017] As an optional implementation, in a second aspect of the present invention, the evaluation module determines the trend features and the fluctuation features as an enhanced feature set, and obtains a unified feature representation through feature fusion operation in the following specific ways: The trend feature and fluctuation feature of each quality index sequence are normalized to obtain standardized trend feature and standardized fluctuation feature. The normalization process is used to eliminate scale differences between features of different dimensions. Attention weights are assigned to the standardized trend feature and the standardized fluctuation feature respectively to obtain weighted trend feature and weighted fluctuation feature. The attention weights are used to characterize the degree of influence of the corresponding feature on the comprehensive quality score of the power grid operation scenario at the current moment. The weighted trend feature and the weighted fluctuation feature are concatenated to obtain a unified feature representation.

[0018] As an optional implementation, in the second aspect of the present invention, the evaluation module inputs the unified representation of the features into a pre-trained quality evaluation model to obtain the comprehensive quality score of the power grid operation scenario at the current moment, specifically in the following ways: Principal component transformation is performed on the unified feature representation to obtain key feature representation. The principal component transformation is used to reduce the dimensionality and decorrelate the unified feature representation to extract core information characterizing the quality status of the power grid operation scenario. Extracting temporal correlation features from the key feature representation, the temporal correlation features are used to characterize the collaborative change pattern of different quality indicators within the preset first time period, specifically including obtaining them by calculating the correlation coefficients between different feature dimension sequences in the key feature representation; The key feature representation is fused with the temporal correlation feature to form the final feature vector, specifically by concatenating the temporal correlation feature as an additional dimension after the key feature representation; The final feature vector is input into the multilayer sensing network in the quality assessment model, and the multilayer sensing network performs nonlinear mapping to output the comprehensive quality score of the power grid operation scenario at the current moment.

[0019] As an optional implementation, in the second aspect of the present invention, the specific method by which the judgment module determines whether the current quality status of the power grid operation scenario is abnormal based on the comprehensive quality score and a preset quality threshold includes: Based on the historical comprehensive quality scores of multiple historical moments within a preset second time period, a quality score sequence is formed. The quality score sequence is used to characterize the changing trend of the quality status of the power grid operation scenario over time. The time length of the preset second time period is longer than that of the preset first time period. Obtain a preset static quality threshold, wherein the static quality threshold is a fixed numerical reference value; A dynamic quality threshold is calculated based on the quality scoring sequence. The dynamic quality threshold is used to reflect the historical normal fluctuation range of the quality status of the power grid operation scenario within the preset second time period. Specifically, it is obtained by calculating the statistics of the quality scoring sequence. Extract the score subsequence located within the preset first time period from the quality score sequence; Linear fitting is performed on the scoring subsequence to obtain the scoring trend; Based on the comprehensive quality score, the static quality threshold, the dynamic quality threshold, and the score change trend, a multi-condition logical judgment is performed to determine whether the current quality status of the power grid operation scenario is abnormal.

[0020] As an optional implementation, in a second aspect of the present invention, the specific method by which the judgment module performs multi-condition logical judgment based on the comprehensive quality score, the static quality threshold, the dynamic quality threshold, and the score change trend to determine whether the current quality status of the power grid operation scenario is abnormal includes: Determine whether the overall quality score is less than the static quality threshold; If it is less than, then the current quality status of the power grid operation scenario is determined to be an abnormal state; If it is not less than, then determine whether the comprehensive quality score is less than the dynamic quality threshold; If it is less than, then determine whether the trend of the score change is a negative trend; If so, the current quality status of the power grid operation scenario is determined to be an abnormal state.

[0021] As an optional implementation, in the second aspect of the present invention, the specific method by which the generation module generates and outputs quality alarm information for the power grid operation scenario based on the comprehensive quality score and the quality threshold includes: The quality anomaly level and anomaly type are determined based on the comprehensive quality score, the static quality threshold, and the dynamic quality threshold. Based on the anomaly type, the quality score sequence, and the quality anomaly level, generate alarm content containing anomaly diagnostic information; The alarm content, the identification information of the power grid operation scenario, and the current time are combined to generate and output quality alarm information for the power grid operation scenario.

[0022] The third aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute the machine learning-based automatic monitoring method for operational scenario quality disclosed in the first aspect of the present invention.

[0023] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: In this embodiment of the invention, multiple quality indicator data of a power grid operation scenario within a preset first time period are acquired. This data is then input into a pre-trained quality assessment model to obtain a comprehensive quality score for the power grid operation scenario at the current moment. The quality assessment model is trained based on multiple quality indicator data from historical time periods and their corresponding scenario state labels, and is used to learn the dynamic correlation between the multiple quality indicator data and the comprehensive quality score. Based on the comprehensive quality score and a preset quality threshold, it is determined whether the current quality state of the power grid operation scenario is abnormal. If so, a quality alarm message for the power grid operation scenario is generated and output based on the comprehensive quality score and the quality threshold. Therefore, implementing this invention enables automated and quantitative assessment of the overall quality of a power grid operation scenario by acquiring multiple quality indicator data and inputting it into a pre-trained quality assessment model. This improves the automation level of the monitoring process, reduces reliance on human experience, and facilitates real-time and continuous analysis of massive amounts of operational data, realizing a shift from manual qualitative judgment to automated quantitative scoring. By leveraging the dynamic correlation between model-learning indicator data and comprehensive scores, it can automatically mine and model complex multi-dimensional and nonlinear relationships, thereby improving the objectivity and consistency of status assessment. This helps overcome judgment biases caused by differences in the experience of different maintenance personnel, achieving unified and intelligent assessment standards. Furthermore, by using comprehensive scores and thresholds to identify anomalies and automatically generate alarm information, it establishes a standardized anomaly detection and alarm triggering process. This improves the accuracy and comprehensiveness of power grid operation quality monitoring and assessment, while also enhancing the timeliness and efficiency of anomaly detection. Consequently, it helps shorten the interval between fault occurrence and maintenance response, enabling rapid perception and early warning of power grid operation risks. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0025] Figure 1 This is a flowchart illustrating an automatic monitoring method for operational scenario quality based on machine learning, as disclosed in an embodiment of the present invention. Figure 2 This is a flowchart illustrating another machine learning-based automatic monitoring method for operational scenario quality disclosed in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an automatic monitoring system for operational scenario quality based on machine learning, as disclosed in an embodiment of the present invention. Figure 4This is a schematic diagram of another machine learning-based automatic monitoring system for operational scenario quality disclosed in an embodiment of the present invention. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0028] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0029] This invention discloses a machine learning-based automatic monitoring method, system, and storage medium for power grid operation scenario quality. It can acquire multiple quality indicator data and input them into a pre-trained quality assessment model to achieve automated and quantitative assessment of the overall quality of power grid operation scenarios. This improves the automation level of the monitoring process, reduces reliance on human experience, and facilitates real-time and continuous analysis of massive operational data, enabling a shift from manual qualitative judgment to automated quantitative scoring. By utilizing the model to learn the dynamic correlation between indicator data and comprehensive scores, it achieves automated mining and modeling of multi-dimensional, nonlinear, and complex correlations, thereby improving the objectivity and consistency of status assessment. This helps overcome judgment biases caused by differences in the experience of different maintenance personnel, achieving unified and intelligent assessment standards. Furthermore, by using comprehensive scores and thresholds to identify anomalies and automatically generate alarm information, it establishes a standardized anomaly detection and alarm triggering process. This improves the accuracy and comprehensiveness of power grid operation scenario quality monitoring and assessment, while also enhancing the timeliness and efficiency of anomaly detection. This helps shorten the interval between fault occurrence and maintenance response, enabling rapid perception and early warning of power grid operation risks. Detailed explanations follow.

[0030] Example 1 Please see Figure 1 , Figure 1 This is a flowchart illustrating an automatic quality monitoring method for operational scenarios based on machine learning, as disclosed in an embodiment of the present invention. Figure 1 The described machine learning-based automatic monitoring method for operational scenario quality can be applied to power system operation and maintenance equipment, and also to intelligent devices related to power system operation and maintenance equipment. These intelligent devices include, but are not limited to, one or more of the following: switching equipment, cloud equipment, edge computing equipment, relay equipment, base station equipment, urban management equipment, and intelligent connected equipment. This invention does not limit the scope of these applications. Figure 1 As shown, this machine learning-based automatic monitoring method for operational scenario quality can include the following operations: 101. Obtain multiple quality indicator data for power grid operation scenarios within a preset first time period; In this embodiment of the invention, optionally, the "power grid operation scenario" mentioned above can refer to a specific physical or logical subsystem, such as an important transmission corridor, a regional distribution network, a substation, or a power plant consisting of multiple new energy generator units. The "quality index data" mentioned above refers to quantitative parameters reflecting the operating status and performance of the scenario, such as, but not limited to, voltage deviation, frequency deviation, three-phase imbalance, load factor, equipment temperature, and the number of alarm signals. The "preset first time period" mentioned above refers to a time window used for recent status assessment; its length can be set according to the data acquisition frequency and monitoring real-time requirements, for example, it can be from a few minutes to several hours. Data can be obtained in real-time or near real-time from the historical databases of the monitoring and data acquisition system, phasor measurement unit, or various intelligent sensors corresponding to the scenario.

[0031] 102. Input multiple quality indicator data into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment. The quality assessment model is trained based on multiple quality indicator data and their corresponding scenario state labels over historical time periods, and is used to learn the dynamic correlation between multiple quality indicator data and the comprehensive quality score. In this embodiment of the invention, optionally, the aforementioned "quality assessment model" can be a function model trained by a machine learning algorithm, capable of comprehensively scoring multi-dimensional inputs. During the model training phase, a large number of historical samples need to be prepared. Each sample contains: multiple quality indicator data collected within a past time period of the same length as the "preset first time period" (as model input features), and a "scenario status label" (such as "normal," "attention," "abnormal," or a specific score, serving as the target for model training) determined based on the overall operational status of the scenario within that time period (e.g., whether a fault occurred, whether operation was stable). Using these labeled historical samples, the model is trained using supervised learning algorithms (e.g., regression algorithms if the label is a score; classification algorithms if the label is a level), enabling it to automatically learn complex, non-linear "dynamic correlations" from the data—that is, the mapping rules between different combinations and change patterns of multiple indicators and the final overall quality status of the scenario. After training, the indicator data acquired at the current moment is input into this model, and the model outputs a quantitative "comprehensive quality score," which comprehensively reflects the overall operational health of the power grid scenario at the current moment.

[0032] In this embodiment of the invention, optionally, details of sample construction are provided to further illustrate how a training sample is constructed. For example, for a historical moment t, using it as the end point, a window of length 'preset first time period' is extracted forward, and multiple quality index data within this window are used as input features X_t; simultaneously, based on the actual operation records of the power grid scenario (such as whether a fault occurred, dispatch logs) within a period of time after the end of this window (e.g., the next hour), experts or according to explicit rules determine the scenario status label Y_t at moment t (such as a score of 0-100, or a category label of 'normal' or 'abnormal'). (X_t, Y_t) is used as a training sample.

[0033] As a further optional measure to address data imbalance, a practical consideration in training is mentioned. Since abnormal power grid conditions are usually far less frequent than normal conditions, strategies such as weighted loss functions or oversampling of minority class samples can be used during model training to alleviate the problem of class imbalance in training samples.

[0034] Further optionally, for model selection and validation: the quality assessment models described above can be implemented using algorithms such as gradient boosting decision trees, support vector regression / classification, or neural networks. After the model is trained, its performance needs to be evaluated using an independent validation set, such as mean squared error (for regression) or accuracy and recall (for classification), to ensure its generalization ability.

[0035] In this embodiment of the invention, as an optional implementation, the above-mentioned input of multiple quality index data into a pre-trained quality assessment model to obtain a comprehensive quality score of the power grid operation scenario at the current moment includes: A sliding window sampling method is used to sample multiple quality indicator data to obtain a local time series data block at the current moment. The local time series data block is used to represent the fluctuation of multiple quality indicator data within a preset first time period. Trend features and fluctuation features are extracted from local time series data blocks. Trend features are used to represent the direction of change of multiple quality indicator data within a preset first time period, and fluctuation features are used to represent the stability of multiple quality indicator data within a preset first time period. Trend features and fluctuation features are identified as an enhanced feature set, and a unified feature representation is obtained through feature fusion operations. The unified representation of features is input into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment.

[0036] In this embodiment of the invention, optionally, for the first step, a sliding window sampling is performed on the aforementioned multiple quality indicator data to obtain a local time-series data block at the current moment. This is a time-series data preprocessing step. "Sliding window sampling" refers to using a fixed-length time window (i.e., a "preset first time period"), with the current moment as the end point of the window, to extract all historical data points within that window from the continuous indicator data stream, forming a multi-dimensional time-series data block. For example, if the "preset first time period" is the most recent 30 minutes, with one data point per minute, then for each quality indicator, a sequence containing 30 data points will be extracted, and the sequences of all indicators together constitute a "local time-series data block." This data block aims to capture the complete fluctuation trajectory of each indicator over a recent period, rather than just a single instantaneous value.

[0037] Optionally, in the second step, trend and volatility features are extracted from the aforementioned local time-series data blocks. This is crucial for feature engineering. Direct evaluation from the raw time-series data is difficult; therefore, more representative information needs to be extracted. "Trend features" characterize whether each indicator sequence generally increases, decreases, or remains relatively stable within a time window; this can be obtained, for example, by calculating the slope of the sequence's linear regression. "Volatility features," on the other hand, characterize the stability or dispersion of each indicator sequence within that time window; this can be obtained, for example, by calculating the sequence's standard deviation or coefficient of variation. Both types of features are extracted for each quality indicator sequence in the data block.

[0038] Optionally, in the third step, the aforementioned trend features and fluctuation features are defined as an enhanced feature set, and a unified feature representation is obtained through feature fusion. The extracted trend features and fluctuation features are independent feature vectors for each indicator. This step first aggregates them to form a higher-dimensional "enhanced feature set," which contains deeper temporal behavioral information not present in the original indicators. Then, these features are processed through "feature fusion," aiming to integrate information, unify scale, and possibly reduce dimensionality, ultimately forming a well-formatted "unified feature representation" suitable for model input. Fusion operations can include simple concatenation, as well as weighting, normalization, or transformation through a small neural network.

[0039] Alternatively, in the fourth step, the unified feature representation described above is input into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment. This step is consistent with the corresponding step described above, but the model input here is a carefully constructed "unified feature representation" rich in time-series information, rather than the original data, which is expected to make the model's assessment more accurate.

[0040] As can be seen, implementing this optional embodiment can obtain local time-series data blocks by using sliding window sampling of multiple indicator data, transforming isolated data points into serialized data units containing time-dimensional information. This improves the data's ability to represent recent operational fluctuations, thereby facilitating the capture of the dynamic behavior of indicators evolving over time and laying the foundation for subsequent time-series analysis. By extracting trend and fluctuation features separately from the time-series data blocks, two core time-series patterns characterizing the direction of change and stability are extracted from the original data. This increases the information density and interpretability of feature information, enabling a more comprehensive characterization of the dynamic and steady-state characteristics of power grid operation and providing more discriminative input for the model. Furthermore, by fusing the two types of features into a unified representation and inputting it into the model, comprehensive analysis and synergistic utilization of trend and fluctuation information are achieved. This improves the model's comprehensive judgment of operational status, enabling more accurate assessment of complex anomalies that simultaneously include different patterns such as trend deterioration and severe fluctuations, achieving information complementarity and enhancement at the feature level.

[0041] In this optional embodiment, as an optional implementation, the above-described extraction of trend features and fluctuation features from local time-series data blocks includes: For each quality index sequence in a local time series data block, the trend slope corresponding to the quality index sequence is obtained by performing linear fitting on the quality index sequence. The trend slope is used to represent the average change of the corresponding quality index sequence per unit time. Calculate the relative rate of change of the first and last values ​​of the quality index sequence within a local time-series data block. The relative rate of change of the first and last values ​​is used to represent the overall change of the quality index sequence within a preset first time period. Calculate the standard deviation of the first-order difference sequence of the quality index series to obtain the trend fluctuation intensity, which is used to represent the stability of the rate of change of the quality index series. The trend slope, the relative rate of change of the first and last values, and the intensity of trend fluctuations are defined as the trend characteristics of this quality index sequence. The coefficient of variation is obtained by calculating the ratio of the standard deviation to the mean of the quality index sequence within a local time series data block. The coefficient of variation is used to represent the relative dispersion of the quality index sequence. The difference between the maximum and minimum values ​​of the quality index sequence within a local time-series data block is calculated to obtain the range, which is used to represent the absolute fluctuation range of the quality index sequence. The coefficient of variation and range are defined as the fluctuation characteristics of this quality index series.

[0042] In this embodiment of the invention, optionally, the trend feature can be composed of the following three sub-features: Trend slope: The slope of the best-fit line is calculated by linearly fitting the numerical values ​​of the indicator sequence to the timestamps (e.g., using the least squares method). A positive slope indicates an upward trend in the indicator within the window period, while a negative slope indicates a downward trend. The absolute value reflects the average rate of change.

[0043] Relative change rate of the first and last values: This calculates the relative change of the value at the beginning of the time window compared to the value at the end (i.e., the current moment). The specific calculation method is: (End value - Beginning value) / Beginning value (or treated as an absolute value). This feature reflects the overall change of the indicator over the entire time period, unaffected by intermediate fluctuations.

[0044] Trend fluctuation strength: First, calculate the first-order difference sequence of the indicator series, that is, calculate the difference between each point in the sequence and the previous point to obtain a new sequence. This difference sequence reflects the instantaneous change of the indicator in each sampling interval. Then, calculate the standard deviation of this difference sequence. This "trend fluctuation strength" characterizes the stability of the indicator's rate of change. A small standard deviation indicates stable change; a large standard deviation indicates drastic and unstable change.

[0045] Combining the trend slope, the relative rate of change of the first and last values, and the intensity of trend fluctuations obtained from the above calculations constitutes the "trend characteristics" of this quality indicator sequence.

[0046] Alternatively, the fluctuation feature can be composed of the following two sub-features: Coefficient of variation: Calculated as the ratio of the standard deviation of the original sequence of the indicator to the absolute mean (to avoid problems caused by a mean of zero or negative values, the absolute value can be used, or the sequence can be ensured to be positive). The coefficient of variation is a dimensionless quantity that eliminates the influence of the magnitude of the indicator itself, purely reflecting the relative dispersion of the data, facilitating the comparison of volatility between indicators with different dimensions.

[0047] Range: Calculates the difference between the maximum and minimum values ​​of the original sequence of the indicator. The range reflects the magnitude of the absolute fluctuation range of the indicator within that time period.

[0048] Combining the coefficient of variation and range calculated above constitutes the "fluctuation characteristics" of this quality indicator sequence.

[0049] As can be seen, implementing this optional embodiment can quantify the trend from three dimensions—average rate of change, overall magnitude of change, and stability of change—by extracting trend slope, relative rate of change of the first and last values, and intensity of trend fluctuation as trend features. This improves the precision and robustness of the description of the change patterns of operational indicators, and helps to distinguish between different trend forms such as long-term slow deterioration and short-term sharp fluctuations, thus achieving a more accurate trend characterization. It can also improve the comprehensiveness of the stability assessment of operational indicators by extracting the coefficient of variation and range as fluctuation features, while considering both the relative dispersion of data and the absolute fluctuation range. This helps to overcome the limitations of single fluctuation measurement indicators (such as using only standard deviation) which may be affected by dimensions or base values, achieving a more robust volatility representation. Furthermore, by combining the above multi-dimensional time-series features, a comprehensive time-series behavioral profile can be constructed for each quality indicator, thereby improving the information completeness of the feature vector of a single indicator. This allows the model to simultaneously learn multiple attributes such as the intensity, speed, and stability of indicator changes, achieving a deep transformation from the original sequence to higher-order features.

[0050] In this optional embodiment, as another optional implementation, the above-mentioned determination of trend features and fluctuation features as an enhanced feature set, and the obtaining of a unified feature representation through feature fusion operation, includes: The trend and fluctuation characteristics of each quality index sequence are normalized to obtain standardized trend and standardized fluctuation characteristics. The normalization process is used to eliminate scale differences between characteristics of different dimensions. Attention weights are assigned to the standardized trend features and standardized fluctuation features respectively to obtain weighted trend features and weighted fluctuation features. The attention weights are used to characterize the degree of influence of the corresponding features on the comprehensive quality score of the power grid operation scenario at the current moment. By concatenating the weighted trend features and the weighted volatility features, a unified feature representation is obtained.

[0051] In this embodiment of the invention, optionally, the trend characteristics and fluctuation characteristics of each quality indicator sequence are first normalized. Since the trend slope and the rate of change at the beginning and end may be negative and have inconsistent dimensions, while the fluctuation intensity, coefficient of variation, and range are all positive, direct concatenation can lead to differences in numerical scale, affecting model learning. Therefore, each feature needs to be normalized, for example, by using min-max normalization to scale it to the [0,1] interval, or by using Z-score normalization to transform it into a distribution with a mean of 0 and a standard deviation of 1. The processed results are called "standardized trend characteristics" and "standardized fluctuation characteristics." This step aims to eliminate dimensions, bringing all features to the same scale, facilitating subsequent fusion and model calculation.

[0052] Optionally, attention weights are then assigned to standardized trend features and standardized fluctuation features, respectively. It is recognized that the importance of trend information and fluctuation information of indicators may differ in power grid operation assessment. For example, for voltage indicators, a sustained downward trend may be more threatening than short-term fluctuations. Therefore, this step introduces an "attention weight" mechanism. A weight value W_t can be assigned to the trend feature of each indicator, and a weight value W_v can be assigned to the fluctuation feature. These weights can be determined in several ways: one is to pre-set them based on power grid operation procedures and expert experience; another, more advanced approach is to allow the model to automatically learn the degree of importance (i.e., attention score) for different features during model training. The standardized features are then multiplied by their corresponding weights to obtain the "weighted trend feature" and "weighted fluctuation feature." A larger weight means a greater impact of that feature in the final assessment.

[0053] Optionally, the weighted trend features and weighted fluctuation features are then concatenated. The trend feature vector and fluctuation feature vector of each indicator, after weighting, are simply concatenated end-to-end according to the indicator order or feature category order to form a longer one-dimensional vector. This concatenated vector is the "unified feature representation," which comprehensively and quantitatively describes the temporal behavior patterns of all quality indicators in the power grid operation scenario within the "preset first time period."

[0054] As can be seen, implementing this optional embodiment can eliminate the scale effect caused by differences in the units and numerical ranges of different features by normalizing trend features and fluctuation features respectively. This improves the comparability and fairness of different features in subsequent fusion and model calculation, and helps to avoid the model training process being dominated by features with large numerical ranges, thus achieving standardization of the feature space. By assigning attention weights to the standardized features, an explicit modeling mechanism for differences in feature importance is introduced, thereby improving the flexibility and adaptability of the feature fusion process. This allows the model to dynamically adjust its attention to different features (such as the fluctuation of some indicators being more important than the trend) according to the current scenario, achieving focused information fusion. By concatenating the weighted features to obtain a unified representation, the final transformation from heterogeneous features to homogeneous model input is completed, thereby improving the regularity of the input data structure and the convenience of model processing. This provides a unified feature foundation with differentiated importance for downstream operations such as dimensionality reduction and correlation analysis, enabling smooth connection of the feature engineering process.

[0055] In this optional embodiment, as another optional implementation, the above-mentioned inputting the unified representation of features into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment includes: Principal component transformation is performed on the unified feature representation to obtain the key feature representation. The principal component transformation is used to reduce the dimensionality and decorrelate the unified feature representation to extract the core information characterizing the quality status of the power grid operation scenario. Extracting temporal correlation features from key feature representations, which are used to characterize the collaborative change patterns of different quality indicators within a preset first time period, specifically by calculating the correlation coefficients between different feature dimension sequences in the key feature representations; The key feature representation is fused with the temporal correlation features to form the final feature vector. Specifically, the temporal correlation features are concatenated as an additional dimension after the key feature representation. The final feature vector is input into the multilayer sensing network in the quality assessment model. The multilayer sensing network performs nonlinear mapping and outputs the comprehensive quality score of the power grid operation scenario at the current moment.

[0056] In this embodiment of the invention, optionally, in the first step, principal component transformation is performed on the unified feature representation. The "unified feature representation" obtained after the aforementioned steps may have a high dimensionality, and there may be correlations between features (for example, the trends and fluctuations of some indicators are correlated). Directly inputting it into the model is inefficient and may introduce noise. Therefore, principal component analysis is used to transform it. PCA maps the original high-dimensional features to a new set of uncorrelated low-dimensional features (i.e., "principal components") through linear transformation, and sorts them according to their variance. The top k principal components with the largest variance are selected to form the "key feature representation". This operation achieves dimensionality reduction (reducing computational load) and decorrelation (extracting independent information), while retaining the most core variation information in the original data.

[0057] Optionally, for the second step, time-series correlation features are extracted from the key feature representation. The coordinated changes among various parameters of the power grid are an important reflection of the system state. Although PCA handles the static correlations between features, this step further explores their time-series correlations. Specifically, each principal component dimension of the "key feature representation" is treated as a new time series (the length of which is the number of sampling points in a preset first time period). Then, the Pearson correlation coefficient or Spearman rank correlation coefficient is calculated between each pair of these principal component time series to obtain a correlation coefficient matrix. The elements of the upper or lower triangular portion (excluding the diagonal) of this matrix are extracted to form a vector, which is the "time-series correlation feature." It characterizes how the core information dimensions reflecting the power grid state fluctuate coordinatedly over the most recent period.

[0058] Optionally, in the third step, the key feature representations and temporal correlation features are fused. The "temporal correlation feature" vector obtained in the second step is taken as a set of additional features and concatenated with the "key feature representation" vector obtained in the first step to form a "final feature vector". In this way, the final feature vector contains both the core temporal summary information of each indicator (from PCA) and the dynamic correlation pattern information between these core information dimensions.

[0059] Optionally, in the fourth step, the final feature vector is input into a multilayer perceptron (MLP) within the quality assessment model. The core structure of this "quality assessment model" is a MLP (a type of feedforward neural network). An MLP consists of an input layer, one or more hidden layers, and an output layer, with fully connected layers and nonlinearity introduced through activation functions. The "final feature vector" is input into the MLP's input layer, where the data undergoes layer-by-layer nonlinear transformations and abstractions in the hidden layers, ultimately producing a scalar value at the output layer—the "comprehensive quality score." The MLP's powerful nonlinear mapping capabilities enable it to learn complex functional relationships between complex fused features and the final score.

[0060] Alternatively, the aforementioned multilayer perceptron can include an input layer (with the same number of neurons as the final feature vector dimension), two hidden layers (with decreasing numbers of neurons, such as 128 and 64), and an output layer (one neuron, used to output the overall quality score). The hidden layers can use the ReLU activation function, and the output layer can use either a linear activation function (for regression) or a sigmoid activation function (for binary classification). During network training, Dropout can be used to prevent overfitting.

[0061] As can be seen, implementing this optional embodiment can reduce the dimensionality and decorrelate the data by performing principal component transformation on the unified feature representation, thereby improving the simplicity and information independence of the feature representation. This helps reduce data redundancy, alleviate the curse of dimensionality, and extract the core information components that best represent the quality state of the scene, achieving purification and condensation of the feature space. It can extract temporal correlation features from key feature representations to deeply explore the collaborative change patterns of different core information dimensions over time, thereby improving the ability to capture the dynamic relationships within the system. This facilitates the discovery of hidden coupling relationships or chain reaction patterns between indicators, achieving an upgrade from single-indicator analysis to system correlation analysis. Furthermore, by fusing key feature representations with temporal correlation features to form a final feature vector, a composite feature vector containing both individual core features and group correlation features can be constructed. This improves the hierarchy and richness of the feature information input to the model, enabling the model to comprehensively learn the impact of individual states and system interactions on overall quality, achieving a more comprehensive state representation. By using a multilayer perceptron to output scores through nonlinear mapping, and leveraging the powerful nonlinear fitting and abstraction capabilities of neural networks, the model's ability to learn the mapping relationship between complex high-dimensional features and the final score can be improved. This is beneficial for accurately depicting the complex relationship between power grid operation quality and multi-dimensional, nonlinear features, and achieving precise end-to-end quality assessment.

[0062] 103. Based on the comprehensive quality score and the preset quality threshold, determine whether the current quality status of the power grid operation scenario is abnormal. In this embodiment of the invention, optionally, based on the aforementioned comprehensive quality score and a preset quality threshold, it is determined whether the current quality status of the aforementioned power grid operation scenario is an abnormal state. This step involves making a decision. The aforementioned "preset quality threshold" is one or more boundary values ​​used to classify quality levels. For example, a threshold S can be set: if the comprehensive quality score is lower than S, it is determined to be an "abnormal state"; if it is higher than or equal to S, it is determined to be a "normal state". The threshold S can be determined based on historical experience, procedural requirements, or statistical analysis of historical score data.

[0063] 104. If so, generate and output quality alarm information for the power grid operation scenario based on the comprehensive quality score and quality threshold.

[0064] In this embodiment of the invention, optionally, the system will automatically generate an alarm when an abnormal state is determined. The alarm information should at least include the scene identifier where the abnormality occurred, the time of occurrence, the current comprehensive quality score, and the threshold information exceeded. The alarm information can be further pushed to the monitoring dashboard, sent via SMS or email to relevant maintenance personnel, thereby triggering subsequent handling procedures.

[0065] As can be seen, implementing the embodiments of the present invention can achieve automated and quantitative assessment of the overall quality of power grid operation scenarios by acquiring multiple quality indicator data and inputting them into a pre-trained quality assessment model. This improves the automation level of the monitoring process, reduces reliance on human experience, and facilitates real-time and continuous analysis of massive operational data, realizing the transformation from manual qualitative judgment to automatic quantitative scoring by machines. By utilizing the model to learn the dynamic correlation between indicator data and comprehensive scores, it can achieve automated mining and modeling of multi-dimensional, nonlinear, and complex correlations, thereby improving the objectivity and consistency of status assessment. This helps overcome judgment biases caused by differences in the experience of different maintenance personnel, achieving the unification and intelligence of assessment standards. By judging anomalies based on comprehensive scores and thresholds and automatically generating alarm information, a standardized anomaly detection and alarm triggering process can be established. This improves the accuracy and comprehensiveness of power grid operation scenario quality monitoring and assessment, while also increasing the timeliness and efficiency of anomaly detection. This helps shorten the interval between fault occurrence and maintenance response, enabling rapid perception and early warning of power grid operation risks. The following provides detailed explanations.

[0066] Example 2 Please see Figure 2 , Figure 2 This is a flowchart illustrating another machine learning-based automatic quality monitoring method for operational scenarios disclosed in an embodiment of the present invention. Figure 2 The described machine learning-based automatic monitoring method for operational scenario quality can be applied to power system operation and maintenance equipment, and also to intelligent devices related to power system operation and maintenance equipment. These intelligent devices include, but are not limited to, one or more of the following: switching equipment, cloud equipment, edge computing equipment, relay equipment, base station equipment, urban management equipment, and intelligent connected equipment. This invention does not limit the scope of these applications. Figure 2 As shown, this machine learning-based automatic monitoring method for operational scenario quality can include the following operations: 201. Obtain multiple quality indicator data for power grid operation scenarios within a preset first time period; 202. Input multiple quality indicator data into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment. The quality assessment model is trained based on multiple quality indicator data and their corresponding scenario state labels over historical time periods, and is used to learn the dynamic correlation between multiple quality indicator data and the comprehensive quality score. 203. Based on the historical comprehensive quality scores of multiple historical moments within a preset second time period, a quality score sequence is formed. The quality score sequence is used to characterize the changing trend of the quality status of the power grid operation scenario over time. The preset second time period is longer than the preset first time period. 204. Obtain the preset static quality threshold, which is a fixed reference value; 205. Calculate the dynamic quality threshold based on the quality score sequence. The dynamic quality threshold is used to reflect the historical normal fluctuation range of the quality status of the power grid operation scenario within a preset second time period. Specifically, it is obtained by calculating the statistics of the quality score sequence. 206. Extract the score subsequence located within the preset first time period from the quality score sequence; 207. Perform linear fitting on the score subsequence to obtain the score change trend; 208. Based on the comprehensive quality score, static quality threshold, dynamic quality threshold, and score change trend, perform multi-condition logical judgment to determine whether the current quality status of the power grid operation scenario is abnormal.

[0067] In this embodiment of the invention, optionally, a quality score sequence is formed based on the historical comprehensive quality scores within a preset second time period. This step introduces a longer time perspective. The aforementioned "preset second time period" is much longer than the "preset first time period," for example, it could be the past 24 hours, a week, or a month. Historical comprehensive quality scores, calculated and saved at a certain frequency (e.g., every minute, every 5 minutes) within this time period, are collected and arranged chronologically to form a "quality score sequence." This sequence reflects the overall quality fluctuation of the power grid scenario over a longer historical period and serves as its "health baseline."

[0068] Optionally, a preset static quality threshold can be obtained. This is an absolute threshold value set according to power grid safety operation regulations, industry standards, or management requirements; it is a fixed value. For example, it can be set to 60 points (assuming a score range of 0-100), representing the passing grade.

[0069] Optionally, a dynamic quality threshold can be calculated based on the quality score sequence. Considering that power grid operation has peak and off-peak periods and normal days, its "normal" quality score level itself will fluctuate. Therefore, a dynamic threshold that adapts to its historical normal fluctuations needs to be calculated. Specifically, a statistic is calculated for the "quality score sequence," for example, the moving average of the sequence (such as the mean of the most recent N scores) minus K times the standard deviation (K is a coefficient set according to the tolerance for anomalies, such as 2 or 3). The resulting dynamic threshold reflects the reasonable lower limit of the quality score under normal fluctuations in the scenario within the most recent "second time period." It is more adaptable to the operational characteristics and periodic changes of the scenario itself than a static threshold. Unlike the rigid fixed thresholds in existing technologies, this step, by calculating a dynamic quality threshold, allows the judgment benchmark to be dynamically adjusted with the power grid load cycle, seasonal changes, etc., thereby effectively reducing false alarms caused by normal fluctuations.

[0070] Optionally, a subsequence of ratings within a preset first time period is extracted, and a rating trend is fitted. From the long sequence formed above, the ratings for the most recent period are extracted, the length of which is equal to the "preset first time period," resulting in a "rating subsequence." This short sequence is then linearly fitted to obtain the slope of a straight line. This slope represents the "rating trend." A negative slope indicates a recent continuous decline in quality; a positive slope indicates improvement or stabilization. This captures the recent dynamics of the quality status.

[0071] Optionally, multi-condition logical judgments can be performed. A comprehensive judgment is made by utilizing four aspects of information: instantaneous score, absolute standard, historical baseline, and recent trend. This judgment logic integrates the absolute safety line, historical operating baseline, and real-time change trend, overcoming the problems of inconsistent manual judgment standards and high false alarm / missed alarm rates of simple threshold alarms, thus achieving intelligent and accurate anomaly identification. The specific judgment logic will be further defined in subsequent embodiments.

[0072] As can be seen, implementing this optional embodiment can provide a long-term, dynamic reference baseline for current state judgment by forming a historical quality score sequence based on a longer second time period. This improves the historical context awareness of state judgment, and facilitates the evaluation of the current score within its own historical fluctuation background, enabling personalized and contextualized judgment. By simultaneously introducing static quality thresholds and dynamic quality thresholds calculated based on historical sequences, a dual judgment basis combining absolute and relative standards can be constructed, thereby improving the completeness and adaptability of the anomaly detection mechanism. This helps to prevent severe anomalies that exceed the absolute safety threshold, as well as detect relative anomalies that deviate from the normal baseline, achieving a monitoring strategy that balances leniency and strictness. By extracting recent score subsequences and fitting score change trends, the recent evolution of quality status can be captured, improving the ability to predict the development trend of the status. This helps to identify potential risks that, although currently acceptable, are continuously deteriorating, extending from post-event alerts to pre-event warnings.

[0073] In this embodiment of the invention, as an optional implementation, the above-mentioned multi-condition logical judgment based on comprehensive quality score, static quality threshold, dynamic quality threshold, and score change trend to determine whether the current quality status of the power grid operation scenario is abnormal includes: Determine whether the overall quality score is less than the static quality threshold; If it is less than, then the current quality status of the power grid operation scenario is determined to be an abnormal state; If it is not less than, then determine whether the overall quality score is less than the dynamic quality threshold; If it is less than, then determine whether the trend of the score change is a negative trend; If so, the current quality status of the power grid operation scenario is determined to be abnormal.

[0074] In this embodiment of the invention, optionally, for the first level of judgment (hard anomaly): determine whether the current comprehensive quality score is less than the static quality threshold. If so, it indicates that the score has fallen below the absolute red line for safe operation of the power grid, the situation is critical, and no other factors need to be considered; it is directly determined to be an abnormal state. This condition corresponds to the most severe anomaly.

[0075] Optionally, for the second-level judgment (dynamic baseline anomaly): if the current score is not lower than the static threshold, proceed to the second level. Determine whether the current comprehensive quality score is lower than the dynamic quality threshold. If so, it means that although the score has not broken through the absolute red line, it is significantly lower than the recent normal level of the scenario itself, which is an anomaly of "deviation from the baseline".

[0076] Optionally, for the third level of judgment (trend confirmation): when the second-level condition (score below the dynamic threshold) is met, an alarm is not immediately triggered; trend confirmation is still required. This involves determining whether the score change trend is negative (i.e., whether the fitted slope is negative). If negative, it indicates that the quality is not only below the historical baseline but is also continuing to deteriorate, which enhances the confidence of the anomaly judgment, thus confirming an abnormal state. If the trend is non-negative (stable or improving), it may be a temporary, accidental fluctuation, and is not yet considered an anomaly; continued observation is maintained.

[0077] This combined logic of "immediate alarm at static threshold" + "alarm only when below dynamic threshold and trending downward" ensures a rapid response to severe anomalies while effectively filtering out acceptable normal fluctuations that are only temporarily below historical averages but do not continue to worsen, thus significantly improving the accuracy of alarms.

[0078] As can be seen, implementing this optional embodiment can establish a high-priority fast channel for anomaly detection by first determining whether the comprehensive score is less than the static threshold, thereby improving the response speed to the most severe and clear anomalies. This helps ensure that any state that exceeds the safety threshold can be immediately captured and alerted, achieving zero-delay response to critical risks. Furthermore, by further determining whether the score is less than the dynamic threshold and the trend is negative when it is not lower than the static threshold, a refined confirmation channel for anomaly detection can be established. This improves the accuracy of identifying non-severe but continuously deteriorating anomalies, effectively filtering out occasional fluctuations that are temporarily below the historical average but do not form a worsening trend, reducing false alarms and improving alarm accuracy. By setting the condition that "below the dynamic threshold" and "negative trend" must be met simultaneously, a trend confirmation step is added to the determination of relative anomalies, thereby improving the rigor and confidence of anomaly detection. This helps avoid unnecessary alarms for temporary and recoverable fluctuations, improves the operability of alarm information, and achieves intelligent discrimination from "whether it is an anomaly" to "whether it is a continuously deteriorating anomaly."

[0079] 209. If so, generate and output quality alarm information for the power grid operation scenario based on the comprehensive quality score and quality threshold.

[0080] In this embodiment of the invention, for other supplementary explanations of steps 201, 202, and 209, please refer to the supplementary explanations of steps 101, 102, and 104 in Embodiment 1. This embodiment of the invention will not repeat these explanations.

[0081] In this optional embodiment, as an optional implementation, the above-mentioned generation and output of quality alarm information for power grid operation scenarios based on comprehensive quality scores and quality thresholds includes: The quality anomaly level and anomaly type are determined based on the comprehensive quality score, static quality threshold, and dynamic quality threshold. Based on the anomaly type, quality score sequence, and quality anomaly level, generate alarm content containing anomaly diagnostic information; The alarm content, the identification information of the power grid operation scenario, and the current time are combined to generate and output quality alarm information for the power grid operation scenario.

[0082] In this embodiment of the invention, optionally, the first step involves determining the quality anomaly level and anomaly type. The anomaly is classified according to the current range of the comprehensive quality score. For example, a score below the static threshold is classified as "urgent"; a score between the static and dynamic thresholds with a negative trend is classified as "serious". Simultaneously, based on whether the first-level judgment was triggered or whether the second or third-level judgment was passed, the anomaly type can be distinguished as "absolutely exceeding the limit" or "trend deterioration".

[0083] Optionally, for the second step, an alarm message containing anomaly diagnostic information is generated. The alarm message is not simply the word "anomaly." The system automatically generates a structured diagnostic text. For example, it may include: "Anomaly type: Deteriorating trend; Current score: 55 (dynamic threshold: 60); Recent trend: Score has decreased by approximately 10% in the past 30 minutes; Historical comparison: Score is below the lower limit of normal fluctuation in the past 24 hours." A simplified trend chart of the quality score sequence can also be attached. This information comes from the comprehensive quality score, static / dynamic thresholds, score change trends, and quality score sequences used in the judgment process, greatly helping operations personnel quickly understand the nature and severity of the anomaly.

[0084] Optionally, for the third step, the final alarm information is generated and output. The detailed "alarm content" generated in the second step, along with "identification information" that uniquely identifies the power grid scenario (e.g., "XX substation 10kV I-section busbar") and the precise "current time" of the anomaly, are packaged into a structured data object (e.g., JSON format). Then, by calling the monitoring system's alarm interface, message queue, or API, the alarm information is output to the next-level monitoring platform, notification system, or maintenance work order system, completing the entire automated monitoring and alarm process.

[0085] As can be seen, implementing this optional embodiment can classify anomalies by determining their level and type based on a comprehensive score and multiple thresholds, thereby improving the clarity of the severity and nature conveyed by alarm information. This helps operations and maintenance personnel quickly determine the priority of handling and understand the nature of the anomaly (such as "urgent limit violation" or "attention degradation"), achieving differentiation and refinement of alarms. It can generate alarm content containing diagnostic information based on anomaly type, scoring sequence, and level, providing contextual and explanatory information beyond simple "limit violation" prompts, thereby improving the guidance and value of alarm information. This helps shorten the time for operations and maintenance personnel to analyze root causes and formulate strategies, supporting faster decision-making and enabling alarm escalation from "what happened" to "why it happened and how serious it is." It can generate structured and complete standardized alarm messages by combining alarm content, scene identifiers and time information and outputting them, thereby improving the efficiency and accuracy of alarm information flow and parsing between different systems. This facilitates seamless integration with upper-level monitoring platforms, work order systems or notification channels, drives automated or semi-automated handling processes, and realizes closed-loop automation of monitoring, analysis, alarm and handling processes.

[0086] Example 3 Please see Figure 3 , Figure 3This is a schematic diagram of the structure of an automatic monitoring system for operational scenario quality based on machine learning, as disclosed in an embodiment of the present invention. This automatic monitoring system for operational scenario quality based on machine learning can be applied to power system operation and maintenance equipment, and also to intelligent devices related to power system operation and maintenance equipment. These intelligent devices include, but are not limited to, one or more of the following: switching equipment, cloud equipment, edge computing equipment, relay equipment, base station equipment, urban management equipment, and intelligent connected devices. The present invention does not impose any limitations on these applications. Figure 3 As shown, the machine learning-based automated monitoring system for operational scenario quality may include: Module 301 is used to acquire multiple quality indicator data of power grid operation scenarios within a preset first time period; The evaluation module 302 is used to input multiple quality indicator data into a pre-trained quality evaluation model to obtain the comprehensive quality score of the power grid operation scenario at the current moment. The quality evaluation model is trained based on multiple quality indicator data and their corresponding scenario state labels over historical time periods and is used to learn the dynamic correlation between multiple quality indicator data and the comprehensive quality score. The judgment module 303 is used to determine whether the current quality status of the power grid operation scenario is abnormal based on the comprehensive quality score and the preset quality threshold. If so, the generation module 304 is used to generate and output quality alarm information for power grid operation scenarios based on the comprehensive quality score and quality threshold.

[0087] As can be seen, implementing the embodiments of the present invention can achieve automated and quantitative assessment of the overall quality of power grid operation scenarios by acquiring multiple quality indicator data and inputting them into a pre-trained quality assessment model. This improves the automation level of the monitoring process, reduces reliance on human experience, and facilitates real-time and continuous analysis of massive operational data, realizing the transformation from manual qualitative judgment to automatic quantitative scoring by machines. By utilizing the model to learn the dynamic correlation between indicator data and comprehensive scores, it can achieve automated mining and modeling of multi-dimensional, nonlinear, and complex correlations, thereby improving the objectivity and consistency of status assessment. This helps overcome judgment biases caused by differences in the experience of different maintenance personnel, achieving the unification and intelligence of assessment standards. By judging anomalies based on comprehensive scores and thresholds and automatically generating alarm information, a standardized anomaly detection and alarm triggering process can be established. This improves the accuracy and comprehensiveness of power grid operation scenario quality monitoring and assessment, while also increasing the timeliness and efficiency of anomaly detection. This helps shorten the interval between fault occurrence and maintenance response, enabling rapid perception and early warning of power grid operation risks. The following provides detailed explanations.

[0088] In this embodiment of the invention, as an optional implementation, the evaluation module 302 inputs multiple quality index data into a pre-trained quality evaluation model to obtain the comprehensive quality score of the power grid operation scenario at the current moment. The specific methods include: A sliding window sampling method is used to sample multiple quality indicator data to obtain a local time series data block at the current moment. The local time series data block is used to represent the fluctuation of multiple quality indicator data within a preset first time period. Trend features and fluctuation features are extracted from local time series data blocks. Trend features are used to represent the direction of change of multiple quality indicator data within a preset first time period, and fluctuation features are used to represent the stability of multiple quality indicator data within a preset first time period. Trend features and fluctuation features are identified as an enhanced feature set, and a unified feature representation is obtained through feature fusion operations. The unified representation of features is input into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment.

[0089] As can be seen, implementing this optional embodiment can obtain local time-series data blocks by using sliding window sampling of multiple indicator data, transforming isolated data points into serialized data units containing time-dimensional information. This improves the data's ability to represent recent operational fluctuations, thereby facilitating the capture of the dynamic behavior of indicators evolving over time and laying the foundation for subsequent time-series analysis. By extracting trend and fluctuation features separately from the time-series data blocks, two core time-series patterns characterizing the direction of change and stability are extracted from the original data. This increases the information density and interpretability of feature information, enabling a more comprehensive characterization of the dynamic and steady-state characteristics of power grid operation and providing more discriminative input for the model. Furthermore, by fusing the two types of features into a unified representation and inputting it into the model, comprehensive analysis and synergistic utilization of trend and fluctuation information are achieved. This improves the model's comprehensive judgment of operational status, enabling more accurate assessment of complex anomalies that simultaneously include different patterns such as trend deterioration and severe fluctuations, achieving information complementarity and enhancement at the feature level.

[0090] In this optional embodiment, as an optional implementation, the specific method by which the evaluation module 302 extracts trend features and fluctuation features from the local time series data block includes: For each quality index sequence in a local time series data block, the trend slope corresponding to the quality index sequence is obtained by performing linear fitting on the quality index sequence. The trend slope is used to represent the average change of the corresponding quality index sequence per unit time. Calculate the relative rate of change of the first and last values ​​of the quality index sequence within a local time-series data block. The relative rate of change of the first and last values ​​is used to represent the overall change of the quality index sequence within a preset first time period. Calculate the standard deviation of the first-order difference sequence of the quality index series to obtain the trend fluctuation intensity, which is used to represent the stability of the rate of change of the quality index series. The trend slope, the relative rate of change of the first and last values, and the intensity of trend fluctuations are defined as the trend characteristics of this quality index sequence. The coefficient of variation is obtained by calculating the ratio of the standard deviation to the mean of the quality index sequence within a local time series data block. The coefficient of variation is used to represent the relative dispersion of the quality index sequence. The difference between the maximum and minimum values ​​of the quality index sequence within a local time-series data block is calculated to obtain the range, which is used to represent the absolute fluctuation range of the quality index sequence. The coefficient of variation and range are defined as the fluctuation characteristics of this quality index series.

[0091] As can be seen, implementing this optional embodiment can quantify the trend from three dimensions—average rate of change, overall magnitude of change, and stability of change—by extracting trend slope, relative rate of change of the first and last values, and intensity of trend fluctuation as trend features. This improves the precision and robustness of the description of the change patterns of operational indicators, and helps to distinguish between different trend forms such as long-term slow deterioration and short-term sharp fluctuations, thus achieving a more accurate trend characterization. It can also improve the comprehensiveness of the stability assessment of operational indicators by extracting the coefficient of variation and range as fluctuation features, while considering both the relative dispersion of data and the absolute fluctuation range. This helps to overcome the limitations of single fluctuation measurement indicators (such as using only standard deviation) which may be affected by dimensions or base values, achieving a more robust volatility representation. Furthermore, by combining the above multi-dimensional time-series features, a comprehensive time-series behavioral profile can be constructed for each quality indicator, thereby improving the information completeness of the feature vector of a single indicator. This allows the model to simultaneously learn multiple attributes such as the intensity, speed, and stability of indicator changes, achieving a deep transformation from the original sequence to higher-order features.

[0092] In this optional embodiment, as another optional implementation, the evaluation module 302 determines the trend features and fluctuation features as an enhanced feature set, and obtains a unified feature representation through feature fusion operation in the following specific ways: The trend and fluctuation characteristics of each quality index sequence are normalized to obtain standardized trend and standardized fluctuation characteristics. The normalization process is used to eliminate scale differences between characteristics of different dimensions. Attention weights are assigned to the standardized trend features and standardized fluctuation features respectively to obtain weighted trend features and weighted fluctuation features. The attention weights are used to characterize the degree of influence of the corresponding features on the comprehensive quality score of the power grid operation scenario at the current moment. By concatenating the weighted trend features and the weighted volatility features, a unified feature representation is obtained.

[0093] As can be seen, implementing this optional embodiment can eliminate the scale effect caused by differences in the units and numerical ranges of different features by normalizing trend features and fluctuation features respectively. This improves the comparability and fairness of different features in subsequent fusion and model calculation, and helps to avoid the model training process being dominated by features with large numerical ranges, thus achieving standardization of the feature space. By assigning attention weights to the standardized features, an explicit modeling mechanism for differences in feature importance is introduced, thereby improving the flexibility and adaptability of the feature fusion process. This allows the model to dynamically adjust its attention to different features (such as the fluctuation of some indicators being more important than the trend) according to the current scenario, achieving focused information fusion. By concatenating the weighted features to obtain a unified representation, the final transformation from heterogeneous features to homogeneous model input is completed, thereby improving the regularity of the input data structure and the convenience of model processing. This provides a unified feature foundation with differentiated importance for downstream operations such as dimensionality reduction and correlation analysis, enabling smooth connection of the feature engineering process.

[0094] In this optional embodiment, as another optional implementation, the evaluation module 302 inputs the unified representation of features into a pre-trained quality evaluation model to obtain the comprehensive quality score of the power grid operation scenario at the current moment in the following specific ways: Principal component transformation is performed on the unified feature representation to obtain the key feature representation. The principal component transformation is used to reduce the dimensionality and decorrelate the unified feature representation to extract the core information characterizing the quality status of the power grid operation scenario. Extracting temporal correlation features from key feature representations, which are used to characterize the collaborative change patterns of different quality indicators within a preset first time period, specifically by calculating the correlation coefficients between different feature dimension sequences in the key feature representations; The key feature representation is fused with the temporal correlation features to form the final feature vector. Specifically, the temporal correlation features are concatenated as an additional dimension after the key feature representation. The final feature vector is input into the multilayer sensing network in the quality assessment model. The multilayer sensing network performs nonlinear mapping and outputs the comprehensive quality score of the power grid operation scenario at the current moment.

[0095] As can be seen, implementing this optional embodiment can reduce the dimensionality and decorrelate the data by performing principal component transformation on the unified feature representation, thereby improving the simplicity and information independence of the feature representation. This helps reduce data redundancy, alleviate the curse of dimensionality, and extract the core information components that best represent the quality state of the scene, achieving purification and condensation of the feature space. It can extract temporal correlation features from key feature representations to deeply explore the collaborative change patterns of different core information dimensions over time, thereby improving the ability to capture the dynamic relationships within the system. This facilitates the discovery of hidden coupling relationships or chain reaction patterns between indicators, achieving an upgrade from single-indicator analysis to system correlation analysis. Furthermore, by fusing key feature representations with temporal correlation features to form a final feature vector, a composite feature vector containing both individual core features and group correlation features can be constructed. This improves the hierarchy and richness of the feature information input to the model, enabling the model to comprehensively learn the impact of individual states and system interactions on overall quality, achieving a more comprehensive state representation. By using a multilayer perceptron to output scores through nonlinear mapping, and leveraging the powerful nonlinear fitting and abstraction capabilities of neural networks, the model's ability to learn the mapping relationship between complex high-dimensional features and the final score can be improved. This is beneficial for accurately depicting the complex relationship between power grid operation quality and multi-dimensional, nonlinear features, and achieving precise end-to-end quality assessment.

[0096] In this embodiment of the invention, as another optional implementation, the specific method by which the judgment module 303 determines whether the current quality status of the power grid operation scenario is abnormal based on the comprehensive quality score and a preset quality threshold includes: Based on the historical comprehensive quality scores of multiple historical moments within a preset second time period, a quality score sequence is formed. The quality score sequence is used to characterize the changing trend of the quality status of the power grid operation scenario over time. The preset second time period is longer than the preset first time period. Obtain the preset static quality threshold, which is a fixed reference value; The dynamic quality threshold is calculated based on the quality score sequence. The dynamic quality threshold is used to reflect the historical normal fluctuation range of the quality status of the power grid operation scenario within a preset second time period. Specifically, it is obtained by calculating the statistics of the quality score sequence. Extract the score subsequence located within a preset first time period from the quality score sequence; Linear fitting is performed on the score subsequence to obtain the score change trend; Based on the comprehensive quality score, static quality threshold, dynamic quality threshold, and score change trend, a multi-condition logical judgment is made to determine whether the current quality status of the power grid operation scenario is abnormal.

[0097] As can be seen, implementing this optional embodiment can provide a long-term, dynamic reference baseline for current state judgment by forming a historical quality score sequence based on a longer second time period. This improves the historical context awareness of state judgment, and facilitates the evaluation of the current score within its own historical fluctuation background, enabling personalized and contextualized judgment. By simultaneously introducing static quality thresholds and dynamic quality thresholds calculated based on historical sequences, a dual judgment basis combining absolute and relative standards can be constructed, thereby improving the completeness and adaptability of the anomaly detection mechanism. This helps to prevent severe anomalies that exceed the absolute safety threshold, as well as detect relative anomalies that deviate from the normal baseline, achieving a monitoring strategy that balances leniency and strictness. By extracting recent score subsequences and fitting score change trends, the recent evolution of quality status can be captured, improving the ability to predict the development trend of the status. This helps to identify potential risks that, although currently acceptable, are continuously deteriorating, extending from post-event alerts to pre-event warnings.

[0098] In this optional embodiment, as an optional implementation method, the judgment module 303 performs multi-condition logical judgment based on the comprehensive quality score, static quality threshold, dynamic quality threshold, and score change trend to determine whether the current quality status of the power grid operation scenario is abnormal. The specific methods include: Determine whether the overall quality score is less than the static quality threshold; If it is less than, then the current quality status of the power grid operation scenario is determined to be an abnormal state; If it is not less than, then determine whether the overall quality score is less than the dynamic quality threshold; If it is less than, then determine whether the trend of the score change is a negative trend; If so, the current quality status of the power grid operation scenario is determined to be abnormal.

[0099] As can be seen, implementing this optional embodiment can establish a high-priority fast channel for anomaly detection by first determining whether the comprehensive score is less than the static threshold, thereby improving the response speed to the most severe and clear anomalies. This helps ensure that any state that exceeds the safety threshold can be immediately captured and alerted, achieving zero-delay response to critical risks. Furthermore, by further determining whether the score is less than the dynamic threshold and the trend is negative when it is not lower than the static threshold, a refined confirmation channel for anomaly detection can be established. This improves the accuracy of identifying non-severe but continuously deteriorating anomalies, effectively filtering out occasional fluctuations that are temporarily below the historical average but do not form a worsening trend, reducing false alarms and improving alarm accuracy. By setting the condition that "below the dynamic threshold" and "negative trend" must be met simultaneously, a trend confirmation step is added to the determination of relative anomalies, thereby improving the rigor and confidence of anomaly detection. This helps avoid unnecessary alarms for temporary and recoverable fluctuations, improves the operability of alarm information, and achieves intelligent discrimination from "whether it is an anomaly" to "whether it is a continuously deteriorating anomaly."

[0100] In this optional embodiment, as another optional implementation, the specific method by which the generation module 304 generates and outputs quality alarm information for the power grid operation scenario based on the comprehensive quality score and quality threshold includes: The quality anomaly level and anomaly type are determined based on the comprehensive quality score, static quality threshold, and dynamic quality threshold. Based on the anomaly type, quality score sequence, and quality anomaly level, generate alarm content containing anomaly diagnostic information; The alarm content, the identification information of the power grid operation scenario, and the current time are combined to generate and output quality alarm information for the power grid operation scenario.

[0101] As can be seen, implementing this optional embodiment can classify anomalies by determining their level and type based on a comprehensive score and multiple thresholds, thereby improving the clarity of the severity and nature conveyed by alarm information. This helps operations and maintenance personnel quickly determine the priority of handling and understand the nature of the anomaly (such as "urgent limit violation" or "attention degradation"), achieving differentiation and refinement of alarms. It can generate alarm content containing diagnostic information based on anomaly type, scoring sequence, and level, providing contextual and explanatory information beyond simple "limit violation" prompts, thereby improving the guidance and value of alarm information. This helps shorten the time for operations and maintenance personnel to analyze root causes and formulate strategies, supporting faster decision-making and enabling alarm escalation from "what happened" to "why it happened and how serious it is." It can generate structured and complete standardized alarm messages by combining alarm content, scene identifiers and time information and outputting them, thereby improving the efficiency and accuracy of alarm information flow and parsing between different systems. This facilitates seamless integration with upper-level monitoring platforms, work order systems or notification channels, drives automated or semi-automated handling processes, and realizes closed-loop automation of monitoring, analysis, alarm and handling processes.

[0102] Example 4 Please see Figure 4 , Figure 4 This is a schematic diagram of another machine learning-based automatic monitoring system for operational scenario quality disclosed in an embodiment of the present invention. This machine learning-based automatic monitoring system for operational scenario quality can be applied to power system operation and maintenance equipment, and also to intelligent devices related to power system operation and maintenance equipment. These intelligent devices include, but are not limited to, one or more of the following: switching equipment, cloud equipment, edge computing equipment, relay equipment, base station equipment, urban management equipment, and intelligent connected equipment. The embodiments of the present invention do not impose limitations on this. Figure 4 As shown, the machine learning-based automated monitoring system for operational scenario quality may include: Memory 401 that stores executable program code.

[0103] Processor 402 coupled to memory 401.

[0104] The processor 402 calls the executable program code stored in the memory 401 to execute the steps in the machine learning-based automatic monitoring method for operational scenario quality described in Embodiment 1 or Embodiment 2 of the present invention.

[0105] Example 5 This invention discloses a computer storage medium storing computer instructions. When these computer instructions are invoked, they are used to execute the steps in the machine learning-based automatic monitoring method for operational scenario quality described in Embodiment 1 or Embodiment 2 of this invention.

[0106] Example 6 This invention discloses a computer program product, which includes a non-transitory computer storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps in the machine learning-based automatic monitoring method for operational scenario quality described in Embodiment 1 or Embodiment 2.

[0107] The system embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0108] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0109] Finally, it should be noted that the automatic monitoring method, system, and storage medium for operational scenario quality based on machine learning disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A machine learning-based method for automatic quality monitoring in operational scenarios, characterized in that, The method includes: Acquire multiple quality indicator data for power grid operation scenarios within a preset first time period; The multiple quality index data are input into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment. The quality assessment model is trained based on the multiple quality index data and their corresponding scenario state labels over historical time periods and is used to learn the dynamic correlation between the multiple quality index data and the comprehensive quality score. Based on the comprehensive quality score and the preset quality threshold, it is determined whether the current quality status of the power grid operation scenario is abnormal. If so, then based on the comprehensive quality score and the quality threshold, quality alarm information for the power grid operation scenario is generated and output.

2. The automatic monitoring method for operational scenario quality based on machine learning according to claim 1, characterized in that, The step of inputting the multiple quality index data into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment includes: A sliding window sampling is performed on the multiple quality indicator data to obtain a local time series data block at the current moment. The local time series data block is used to represent the fluctuation of the multiple quality indicator data within a preset first time period. Trend features and fluctuation features are extracted from the local time-series data block. The trend features are used to represent the direction of change of the multiple quality index data within the preset first time period, and the fluctuation features are used to represent the stability of the multiple quality index data within the preset first time period. The trend features and the fluctuation features are determined as an enhanced feature set, and a unified feature representation is obtained through feature fusion operation; The unified representation of the features is input into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment.

3. The automatic monitoring method for operational scenario quality based on machine learning according to claim 2, characterized in that, The extraction of trend features and fluctuation features from the local time-series data block includes: For each quality index sequence in the local time series data block, the trend slope corresponding to the quality index sequence is obtained by linear fitting the quality index sequence. The trend slope is used to represent the average change of the corresponding quality index sequence in a unit time. Calculate the relative rate of change of the first and last values ​​of the quality index sequence within the local time-series data block. The relative rate of change of the first and last values ​​is used to represent the overall change range of the quality index sequence within the preset first time period. Calculate the standard deviation of the first-order difference sequence of the quality index sequence to obtain the trend fluctuation intensity, which is used to represent the stability of the rate of change of the quality index sequence; The trend slope, the relative rate of change of the first and last values, and the trend fluctuation intensity are determined as the trend characteristics of the quality index sequence. The ratio of the standard deviation to the mean of the quality index sequence within the local time-series data block is calculated to obtain the coefficient of variation, which is used to represent the relative dispersion of the quality index sequence. The difference between the maximum and minimum values ​​of the quality index sequence within the local time-series data block is calculated to obtain the range, which is used to represent the absolute fluctuation range of the quality index sequence. The coefficient of variation and the range are used to determine the fluctuation characteristics of the quality index sequence.

4. The automatic monitoring method for operational scenario quality based on machine learning according to claim 3, characterized in that, The step of determining the trend features and the fluctuation features as an enhanced feature set, and obtaining a unified feature representation through feature fusion operation, includes: The trend feature and fluctuation feature of each quality index sequence are normalized to obtain standardized trend feature and standardized fluctuation feature. The normalization process is used to eliminate scale differences between features of different dimensions. Attention weights are assigned to the standardized trend feature and the standardized fluctuation feature respectively to obtain weighted trend feature and weighted fluctuation feature. The attention weights are used to characterize the degree of influence of the corresponding feature on the comprehensive quality score of the power grid operation scenario at the current moment. The weighted trend feature and the weighted fluctuation feature are concatenated to obtain a unified feature representation.

5. The automatic monitoring method for operational scenario quality based on machine learning according to claim 4, characterized in that, The step of inputting the unified representation of the features into a pre-trained quality assessment model to obtain the comprehensive quality score of the power grid operation scenario at the current moment includes: Principal component transformation is performed on the unified feature representation to obtain key feature representation. The principal component transformation is used to reduce the dimensionality and decorrelate the unified feature representation to extract core information characterizing the quality status of the power grid operation scenario. Extracting temporal correlation features from the key feature representation, the temporal correlation features are used to characterize the collaborative change pattern of different quality indicators within the preset first time period, specifically including obtaining them by calculating the correlation coefficients between different feature dimension sequences in the key feature representation; The key feature representation is fused with the temporal correlation feature to form the final feature vector, specifically by concatenating the temporal correlation feature as an additional dimension after the key feature representation; The final feature vector is input into the multilayer sensing network in the quality assessment model, and the multilayer sensing network performs nonlinear mapping to output the comprehensive quality score of the power grid operation scenario at the current moment.

6. The automatic monitoring method for operational scenario quality based on machine learning according to any one of claims 1-5, characterized in that, The step of determining whether the current quality status of the power grid operation scenario is abnormal based on the comprehensive quality score and a preset quality threshold includes: Based on the historical comprehensive quality scores of multiple historical moments within a preset second time period, a quality score sequence is formed. The quality score sequence is used to characterize the changing trend of the quality status of the power grid operation scenario over time. The time length of the preset second time period is longer than that of the preset first time period. Obtain a preset static quality threshold, wherein the static quality threshold is a fixed numerical reference value; A dynamic quality threshold is calculated based on the quality scoring sequence. The dynamic quality threshold is used to reflect the historical normal fluctuation range of the quality status of the power grid operation scenario within the preset second time period. Specifically, it is obtained by calculating the statistics of the quality scoring sequence. Extract the score subsequence located within the preset first time period from the quality score sequence; Linear fitting is performed on the scoring subsequence to obtain the scoring trend; Based on the comprehensive quality score, the static quality threshold, the dynamic quality threshold, and the score change trend, a multi-condition logical judgment is performed to determine whether the current quality status of the power grid operation scenario is abnormal.

7. The automatic monitoring method for operational scenario quality based on machine learning according to claim 6, characterized in that, The step of performing multi-condition logical judgment based on the comprehensive quality score, the static quality threshold, the dynamic quality threshold, and the score change trend to determine whether the current quality status of the power grid operation scenario is abnormal includes: Determine whether the overall quality score is less than the static quality threshold; If it is less than, then the current quality status of the power grid operation scenario is determined to be an abnormal state; If it is not less than, then determine whether the comprehensive quality score is less than the dynamic quality threshold; If it is less than, then determine whether the trend of the score change is a negative trend; If so, the current quality status of the power grid operation scenario is determined to be an abnormal state.

8. The automatic monitoring method for operational scenario quality based on machine learning according to claim 7, characterized in that, The step of generating and outputting quality alarm information for the power grid operation scenario based on the comprehensive quality score and the quality threshold includes: The quality anomaly level and anomaly type are determined based on the comprehensive quality score, the static quality threshold, and the dynamic quality threshold. Based on the anomaly type, the quality score sequence, and the quality anomaly level, generate alarm content containing anomaly diagnostic information; The alarm content, the identification information of the power grid operation scenario, and the current time are combined to generate and output quality alarm information for the power grid operation scenario.

9. An automated quality monitoring system for operational scenarios based on machine learning, characterized in that, The system includes: The acquisition module is used to acquire multiple quality indicator data of the power grid operation scenario within a preset first time period; The evaluation module is used to input the multiple quality index data into a pre-trained quality evaluation model to obtain the comprehensive quality score of the power grid operation scenario at the current moment. The quality evaluation model is trained based on the multiple quality index data and their corresponding scenario state labels over historical time periods and is used to learn the dynamic correlation between the multiple quality index data and the comprehensive quality score. The judgment module is used to determine whether the current quality status of the power grid operation scenario is abnormal based on the comprehensive quality score and the preset quality threshold. If so, the generation module is used to generate and output quality alarm information for the power grid operation scenario based on the comprehensive quality score and the quality threshold.

10. A computer storage medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked, are used to execute the machine learning-based automatic monitoring method for operational scenario quality as described in any one of claims 1-8.

Images