A transformer area energy storage operation dynamic early warning control method and system

Abstract

The application discloses a kind of district energy storage operation dynamic early warning control method and system, it is related to abnormal detection technical field, including the following steps: obtaining the operation data of energy storage system, constructs data trend deviation vector, and extracts trend deviation feature;Trend deviation feature is dynamically matched with the micro-disturbance abnormal feature database that is constructed based on historical data in advance, and the preliminary abnormal early warning signal is output;Disturbance analysis is carried out by obtaining synchronous phasor measurement data, disturbance response feature is identified and is combined with trend similarity, and the preliminary abnormal early warning signal is corrected;The target abnormal feature corresponding to the corrected abnormal early warning signal is extracted, and the preliminary abnormal early warning signal correction result is combined, and the early warning credibility score is output;The application realizes the dynamic perception of energy storage system operation state and abnormal early warning credibility evaluation, effectively improves the early warning accuracy and system response capability, solves the problem of identification lag and high false alarm rate.

Description

Technical Field

[0001] This invention relates to the field of anomaly detection technology, and more specifically, to a dynamic early warning control method and system for the operation of energy storage in a distribution area. Background Technology

[0002] With the rapid development of new energy and distributed energy, energy storage systems, as an important technological means to enhance grid regulation capabilities, ensure power supply reliability, and promote energy transformation, are gradually becoming a key component of modern power systems. Especially at the distribution substation level, energy storage systems can effectively mitigate load fluctuations, optimize energy management, improve power quality, and contribute to the construction of smart grids. However, energy storage devices experience many complex dynamic changes during operation, such as fluctuations in operating parameters, sudden disturbances, and equipment aging. These factors can all lead to performance degradation and even safety hazards in energy storage systems, seriously affecting grid stability and operational safety. Therefore, how to monitor and accurately identify abnormal states of energy storage systems in real time and issue timely early warning signals has become a core technical challenge to ensure the safe and efficient operation of energy storage systems.

[0003] For example, the invention patent with announcement number CN114995208A discloses an energy storage safety system and a control method for the energy storage safety system. This control method acquires the real-time temperature inside the energy storage tank of the battery energy storage system and the gas concentration of combustible gas released during battery thermal runaway. Then, when the real-time temperature exceeds a first warning temperature and / or the gas concentration exceeds a first warning concentration, a first warning signal is generated and finally sent to an alarm. This method, starting from the gas concentration of combustible gas and the real-time temperature inside the energy storage tank, solves problems such as untimely battery thermal runaway warnings, inaccurate judgment of battery thermal runaway, inability to prevent battery thermal runaway, and excessively long alarm signal chains.

[0004] For example, the invention patent with announcement number CN115328009A discloses a monitoring and management system for gravity energy storage systems, relating to the field of gravity energy storage technology. This invention addresses the problem of the lack of a comprehensive monitoring and control management system for gravity energy storage systems in the prior art. The invention includes a data acquisition unit for collecting data from the gravity energy storage side, grid-side operation data, and environmental data; a monitoring and early warning unit for making early warning judgments based on the collected data and generating early warning information when an early warning occurs; a data analysis unit for performing performance analysis on the gravity energy storage side, analysis of the relationship between energy storage and grid operation, and protection risk analysis based on the indicators obtained from the collected data and early warning information, generating analysis results and decision instructions; and a comprehensive display and reporting unit for displaying the real-time data collected by the data acquisition unit, the early warning information generated by the monitoring and early warning unit, and the analysis results and decision instructions output by the data analysis unit. This invention enables real-time monitoring, management, and control of gravity energy storage systems, ensuring stable grid operation and improving energy conversion and utilization efficiency.

[0005] The above-disclosed technical solutions have at least the following technical problems:

[0006] Existing technologies are not comprehensive enough in extracting time-series features from energy storage system operation data. Most focus only on static indicators or short-term fluctuations, lacking the extraction of dynamic trend shift features such as adaptive segmentation, trend change intensity, and significance scoring, making it difficult to capture the complex evolution patterns of energy storage operation. To address these issues, this invention proposes a solution. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a dynamic early warning control method and system for energy storage operation in a distribution area. By constructing a dynamic matching mechanism between trend offset features and micro-perturbation templates, and combining phasor perturbation analysis and multimodal data fusion, the present invention achieves dynamic perception of the operating status of the energy storage system and reliability assessment of abnormal early warning, effectively improving the accuracy of early warning and the system response capability, and solving the problems of identification lag and high false alarm rate of traditional methods.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A dynamic early warning control method for energy storage operation in a distribution area includes the following steps: acquiring the operating data of the energy storage system, constructing a data trend offset vector, and extracting the trend offset features of the data trend offset vector; performing similarity matching between the trend offset features and a dynamically updated historical anomaly feature database, and outputting a preliminary anomaly early warning signal when the matching degree exceeds a preset threshold; and performing disturbance analysis by acquiring synchronous phasor measurement data, identifying disturbance response features, and correcting the preliminary anomaly early warning signal by combining trend similarity.

[0010] Extract the target anomaly features corresponding to the corrected anomaly warning signal, and combine them with the correction results of the initial anomaly warning signal to output the warning credibility score.

[0011] In a preferred embodiment, the method for extracting trend shift features is as follows: Based on the preprocessed time series data, the rate of change and slope change of the data points are extracted, and the time series data is divided into several trend candidate segments using an adaptive segmentation algorithm; each trend candidate segment is processed using a time series clustering algorithm, and the trend feature vector of each segment is extracted; the difference between the trend feature vectors of any two adjacent segments is measured to generate a trend change intensity value, and a significance score is output in combination with the trend features; highly significant trend shift points are selected based on the significance score, adjacent shift points are connected to generate the trend shift principal axis, and the trend shift features corresponding to each shift point are extracted.

[0012] In a preferred embodiment, the trend shift features are matched with a dynamically updated historical anomaly feature database for similarity. When the matching degree exceeds a preset threshold, a preliminary anomaly warning signal is output. Specifically, along the trend shift axis, the highly significant trend shift points within the current sliding time window are sorted by time series, the trend shift features of each point are extracted and formed into a trend shift sequence; the trend shift sequence is compared with the anomaly sequences in the anomaly feature database to generate a matching confidence score, and a preliminary anomaly warning signal is output according to the warning matching threshold.

[0013] In a preferred embodiment, the trend offset sequence is compared with the abnormal sequences in the abnormal feature database to generate a matching confidence score, specifically as follows: within a sliding time window, points with significant scores are selected to form a trend offset point set; the trend offset point set is arranged chronologically to generate the current trend offset path, and the relative increment of the principal axis coordinates of adjacent points is calculated to form a relative increment path; the baseline relative path in the pre-stored historical abnormal database is read; the path direction consistency score and principal axis increment similarity score between the current relative increment path and the baseline relative path are calculated; a weighted average of the direction consistency score and the increment similarity score is calculated to output the final matching confidence score.

[0014] In a preferred embodiment, disturbance analysis is performed by acquiring synchronous phasor measurement data, disturbance response characteristics are identified, and trend similarity is combined to correct the preliminary anomaly warning signal. Specifically, the following steps are taken: phasor measurement data of the energy storage node and its adjacent topology nodes are acquired within the preliminary anomaly warning window, and the trend of grid boundary entropy change is collected synchronously; disturbance synchronization analysis is performed on the phasor measurement data of adjacent topology nodes, abrupt change characteristic quantities are extracted, and disturbance response characteristics are identified; the current boundary entropy change trend is matched with the boundary entropy trajectory of the historical anomaly evolution sample set using a dynamic time warping method, and the trend similarity is output; the preliminary anomaly warning signal is corrected by fusing disturbance response characteristics and trend similarity, wherein the correction includes boosting, maintaining, marking, or downgrading the preliminary anomaly warning signal.

[0015] In a preferred embodiment, the current boundary entropy change trend is matched with the boundary entropy trajectory of the historical anomaly evolution sample set using a dynamic time warping method, and the trend similarity is output. Specifically, during the warning period, the boundary entropy evolution sequence under the sliding window is extracted based on the key operating parameters of the energy storage system and recorded as the current boundary entropy sequence; the historical anomaly evolution sample set is read, where each sample contains a historical boundary entropy change sequence; the current boundary entropy sequence and the historical boundary entropy change sequence are temporally elastically registered using a dynamic time warping algorithm, and the corresponding matching distance is output; each matching distance is converted into a trend similarity score using an exponential decay function, and the maximum trend similarity score is selected; the maximum trend similarity score is compared with a set threshold, and the trend similarity is output.

[0016] In a preferred embodiment, the target anomaly features corresponding to the corrected anomaly warning signal are extracted, and the warning confidence score is output by combining the correction results of the preliminary anomaly warning signal. This includes: acquiring multimodal time-series data within the time window corresponding to the corrected anomaly warning signal and constructing a multimodal time-series data input set; performing cross-combination processing on the multimodal time-series data to generate a cross-feature space; extracting key cross-features from the cross-feature space based on statistical analysis methods; performing embedding and reconstruction analysis on the key cross-features within the anomaly window through kernel principal component analysis and outputting a residual score; extracting disturbance behavior features from phasor measurement data and generating a disturbance confidence score through discrete counting index mapping; and fusing the residual score and the disturbance confidence score to output a warning confidence score.

[0017] In a preferred embodiment, the residual score and perturbation confidence score are fused to output the early warning credibility score. Specifically, the residual score, trend similarity score, and perturbation confidence score are mapped to fuzzy linguistic variables through fuzzy membership functions; a fuzzy logic rule set is designed based on domain knowledge and used as a prior knowledge constraint; under the prior knowledge constraint, a conditional probability model is constructed using a decision tree algorithm; the conditional probability weight of each rule is calculated based on historical data, and the triggering rules are fused using Bayesian weighted fusion to output the final early warning credibility score.

[0018] A dynamic early warning control system for energy storage operation in a distribution area includes a feature extraction module, a matching module, a correction module, and an output module, with connections between the modules. The feature extraction module acquires the operating data of the energy storage system, constructs a data trend offset vector, and extracts trend offset features. The matching module dynamically matches the trend offset features with a pre-constructed database of perturbation anomaly features based on historical data and outputs a preliminary anomaly early warning signal. The correction module performs perturbation analysis by acquiring synchronous phasor measurement data, identifies perturbation response features, and corrects the preliminary anomaly early warning signal by combining trend similarity. The output module extracts the target anomaly features corresponding to the corrected anomaly early warning signal and outputs an early warning credibility score based on the correction result of the preliminary anomaly early warning signal.

[0019] The technical effects and advantages of the dynamic early warning control method and system for energy storage operation in transformer substations as described in this invention are as follows:

[0020] 1. This invention integrates multi-source heterogeneous data analysis with time-series intelligent algorithms to achieve dynamic, accurate, and reliable anomaly early warning and control of energy storage system operation status. It possesses numerous technical advantages, including forward-looking capabilities, sensitivity, and robustness, and has significant engineering application value and broad prospects for widespread application. Secondly, through adaptive segmentation and time-series clustering algorithms, it structures the time-series data, accurately capturing significant trend shifts in data changes, and then constructing a trend shift axis. This approach effectively uncovers the deep dynamic change characteristics of energy storage system operation, overcoming the limitations of traditional static threshold monitoring methods in identifying complex abnormal behaviors, and providing a solid dynamic perception foundation for the early warning mechanism.

[0021] 2. This invention achieves real-time identification of anomaly patterns through dynamic matching with a pre-built perturbation anomaly template library. It employs dual indicators—path direction consistency and principal axis increment similarity—to perform high-dimensional similarity measurement between trend offset sequences and historical anomaly trajectories. Combined with a weighted average algorithm, the matching confidence score is output, improving the accuracy and real-time performance of anomaly detection. This mechanism has strong scalability and can adapt to different forms of anomaly behavior in various operating scenarios. Based on the initial warning, the system further introduces phasor measurement data from energy storage nodes and their topological neighborhoods for perturbation synchronization analysis. This is combined with dynamic time warping matching of boundary entropy evolution trends and historical evolution samples to double-verify the credibility of the initial warning signal. This stage, through the fusion of perturbation response and entropy evolution as dual evidence mechanisms, effectively improves the robustness and reliability of the warning results, avoiding false alarms and missed alarms. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the process of a dynamic early warning and control method for energy storage operation in a distribution area according to the present invention.

[0023] Figure 2 This is a schematic diagram of the structure of a dynamic early warning control system for energy storage operation in a distribution area according to the present invention. Detailed Implementation

[0024] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1, Figure 1 This invention provides a dynamic early warning and control method for the operation of energy storage in a distribution area, comprising the following steps:

[0026] S1. Obtain the operating data of the energy storage system, construct the data trend offset vector, and extract the trend offset features of the data trend offset vector.

[0027] In this embodiment, the operating data of the energy storage system is acquired, a data trend offset vector is constructed, and the trend offset features of the data trend offset vector are extracted, as follows:

[0028] S11: Obtain the raw multi-dimensional parameter time-series data of the energy storage system operation, including parameters such as voltage, current, SOC, and temperature;

[0029] Preprocessing of multidimensional parameter time series data includes time alignment, missing value imputation, and noise filtering.

[0030] Based on the preprocessed data, the rate of change and slope change of the data are extracted. Based on the adaptive segmentation algorithm (such as the sliding window-based change point detection method, which calculates the rate of change and slope change of statistical data within the sliding window and divides the boundary when the change amplitude exceeds the set threshold), the entire time series is divided into several trend candidate segments. Each segment corresponds to a relatively stable or continuously changing trend segment. The segment length is automatically adjusted according to the change intensity, forming unequal length divisions.

[0031] For each segment, a time series clustering algorithm is applied to extract its representative trend features and form a set of trend feature vectors. The trend features include the average slope, the change in the sign of the average slope, and the jump in the standard deviation of the slope.

[0032] The difference between the cluster center vectors of any two adjacent sub-segments is measured, and the intensity of the trend change is calculated to quantify the magnitude of the trend jump and preliminarily characterize the degree of trend shift.

[0033] Based on the intensity of trend abrupt change, the change in the sign of the average slope (used to judge the reversal of trend direction), and the jump in the standard deviation of the slope (used to judge volatility), a significance evaluation function for trend abrupt change is constructed as a cognitive salience function. The significance score of each potential trend deviation point is calculated to quantify its importance to the system's attention.

[0034] The cognitive salience function is applied to the full-time trend feature sequence to extract highly significant trend shift points and form the trend shift axis. For example, a significance threshold is set, and shift points with significance scores higher than the threshold are selected and organized in chronological order to form the trend shift axis. The axis represents an ordered set of key evolutionary jump points in the time series, which is used for subsequent key feature extraction and event focusing.

[0035] For each offset point in the trend offset principal axis, extract its associated trend offset features. These features include offset time (obtained directly from the principal axis), offset direction (determined by the change in the sign of the slope of the preceding and following segments), offset magnitude (quantified by the difference in feature values ​​of the segments), local volatility (obtained by the standard deviation of the slope of the preceding and following segments), offset duration (estimated by the segment width window), and significance level (inherited from the salience function score). Construct a trend offset vector.

[0036] The specific cognitive salience function is as follows:

[0037]

[0038] In the formula: Normalize the jumps in the standard deviation of the slope to between 0 and 1. The change in the standard deviation of the slope at trend deviation point i reflects the degree of trend instability. It is a trend reversal indicator at point i. It is 1 if a shift from positive to negative trend occurs, and 0 otherwise. and These are weighting coefficients. It is the trend jump magnitude at trend deviation point i. It is the base of the natural logarithm. It is the Softmax exponential scaling factor. It is the comprehensive significance score of the potential trend offset point i. It is an index variable that represents each point among all trend offset points.

[0039] S2 performs similarity matching between the trend offset features and the dynamically updated historical anomaly feature database. When the matching degree exceeds a preset threshold, a preliminary anomaly warning signal is output.

[0040] In this embodiment, the trend offset feature is matched with a dynamically updated historical anomaly feature database. When the matching degree exceeds a preset threshold, a preliminary anomaly warning signal is output, as follows:

[0041] From a large amount of historical energy storage system operation data, various typical abnormal evolution processes (such as abnormal power oscillation, abnormal SOC rise, frequent charge-discharge switching, etc.) are screened out, and key parameter evolution segments are extracted in the form of time series slices to form a "perturbation anomaly feature database". Each template contains key feature point sequence, amplitude gradient features, change rate curve and feature jump point label in the trend offset main axis direction.

[0042] For real-time running data, along the extracted trend offset main axis, the highly significant trend offset points within the current sliding time window are sorted by time series, and their trend offset features are extracted to form a trend offset sequence.

[0043] The current trend offset sequence is compared with each abnormal template sequence in the template library to generate a matching confidence score;

[0044] Set a warning matching threshold. When the matching confidence score of any abnormal template exceeds the warning matching threshold, it is considered that the current window trend has typical abnormal evolution characteristics, and a preliminary abnormal warning signal is output.

[0045] The similarity comparison between the current trend offset sequence and each abnormal template sequence in the template library is performed as follows:

[0046] Within the sliding time window, points with scores higher than a set threshold are selected based on the calculated significance score to form a set of trend offset points. Each point contains: a timestamp, significance score, direction of change, and principal axis position coordinates (representing the projection position of the point on the trend principal axis).

[0047] Arranging the sequence of offset points in chronological order forms the current trend offset path, which can be regarded as an "event flow trajectory" in two-dimensional space.

[0048] Based on the current trend offset path, the relative increment of the principal axis coordinates (the difference in principal axis position between adjacent projected coordinates in the trend offset path) is output to form a relative increment path, thereby eliminating the absolute numerical difference.

[0049] Read the relative paths saved in the pre-built historical anomaly templates. Each template contains a fixed-length relative offset increment and direction sequence.

[0050] Based on the current relative incremental path and each template path, calculate the path similarity and output the path direction consistency score and the principal axis incremental similarity score (obtained using the weighted cosine similarity algorithm).

[0051] Based on the path direction consistency score and the main axis incremental similarity score, the final matching confidence score is output using a weighted average algorithm.

[0052] The path direction consistency score is as follows:

[0053]

[0054] The specific spindle incremental similarity score is as follows:

[0055]

[0056] The specific matching confidence score is as follows:

[0057]

[0058] In the formula: It is the path direction consistency score. It is the total number of trend deviation points. This is the trend jump direction label, indicating the trend jump direction of the offset point. It is usually set to +1 (upward) or -1 (downward). It is the direction of the i-th trend change point in the j-th template path. This is a consistency indicator function. It takes a value of 1 if the current direction matches the template direction, and 0 otherwise. It is used to calculate average consistency. It is the principal axis incremental similarity score. It is the increment of the i-th position in the j-th template path, compared with the current path. Correspondingly, it is used to calculate the inner product and norm. This is the principal axis offset increment, representing the position difference of the i-th offset point relative to the previous offset point along the principal axis, used to construct the path vector. and Let be the Euclidean norm of the sequence, and be the lengths of the current path and template path increment vectors, respectively. It is the confidence score of the match between the current path and the j-th template path. and It is the weighting coefficient.

[0059] S3 performs disturbance analysis by acquiring synchronous phasor measurement data, identifies disturbance response characteristics, and combines trend similarity to correct the preliminary anomaly warning signal.

[0060] In this embodiment, disturbance analysis is performed by acquiring synchronization phasor measurement data, disturbance response characteristics are identified, and trend similarity is combined to correct the preliminary anomaly warning signal, as follows:

[0061] Get the corresponding preliminary anomaly warning window , The following data: Synchronous phasor measurement data of the energy storage node and its neighboring nodes, including voltage phase angle, frequency, and voltage amplitude (sampling frequency ≥ 10Hz); Current operating parameter sequence in , Real-time boundary entropy evolution trend within the segment; boundary entropy evolution sample set formed by historically confirmed anomalous evolution paths;

[0062] Phasor measurement data of adjacent topology nodes of the energy storage node in , Disturbance synchronization analysis is performed within a time period to extract abrupt change characteristics, including voltage phase angle change rate, system frequency jump rate, and voltage amplitude change rate.

[0063] During the disturbance synchronization analysis, if any of the mutation characteristics changes and highly overlaps with the warning time period (e.g., the time difference of the mutation is less than 200ms and the direction of the jump is consistent), it is considered that there is a "system-level consistent disturbance". The preliminary warning signal is marked as a high-confidence warning and enters the subsequent processing flow. If the neighboring nodes do not have significant synchronous response behavior, it is judged that the disturbance may be a local isolated event and the system-level confidence is reduced.

[0064] The boundary entropy change trend of the current running parameters is obtained, and the boundary entropy change trend of the current running parameters is matched with the boundary entropy trajectory of multiple historical abnormal evolution samples through dynamic time warping method. The trend similarity is output, and it is determined whether the boundary entropy trend is consistent with the historical abnormal evolution sample set. If it is consistent, it is marked as "having an abnormal evolution trend".

[0065] If the similarity between the current boundary entropy trend and the historical typical abnormal evolution process trend is higher than the set threshold, the current disturbance is considered to have a continuous evolution trend and is marked as a trend development warning; otherwise, it is regarded as an isolated anomaly or noise disturbance.

[0066] The results of phasor mutation judgment and boundary entropy comparison are combined, and the preliminary anomaly warning signal is corrected, as follows:

[0067] If both the PMU phasor mutation and the boundary entropy evolution are satisfied, the alert level is upgraded to a high-confidence warning.

[0068] If only the PMU mutation is met, the initial warning level is maintained;

[0069] If only the boundary entropy consistency is satisfied, it is marked as a "trend warning" state, pending confirmation by subsequent observations;

[0070] If none of the conditions are met, the current warning signal will be downgraded or suppressed.

[0071] The method of dynamic time warping is used to perform similarity matching between the trend of the boundary entropy change of the current operating parameters and the boundary entropy trajectories of multiple historical anomalous evolution samples, as follows:

[0072] The boundary entropy evolution sequence of any window length is extracted from the key operating parameters of the energy storage system (such as SOC, current, and voltage) during the warning period and denoted as the current boundary entropy sequence. This sequence reflects the entropy structure evolution of the current parameter change trend.

[0073] Construct a historical abnormal evolution sample set. The sample data comes from the operating trajectories of energy storage systems that have been confirmed as faults or abnormal events in the past. Each sample contains a historical boundary entropy change sequence.

[0074] The current boundary entropy evolution sequence and the historical boundary entropy change sequence are normalized, and the minimum-maximum normalization method is used to map each entropy value to the [0,1] interval to eliminate the amplitude difference between different samples and maintain the consistency of the evolution trend structure.

[0075] The Dynamic Time Warping (DTW) algorithm is used to perform temporal elastic registration between the current boundary entropy sequence and each historical boundary entropy change sequence in the sample set, and the matching distance is output.

[0076] Each matching distance is converted into a trend similarity score using an exponential decay function;

[0077] Calculate the similarity scores for all trends, obtain the maximum similarity value, and compare it with a set threshold;

[0078] If the maximum similarity value exceeds the set threshold, the current trend is considered to be consistent with the historical abnormal evolution path and is judged to have an abnormal evolution trend.

[0079] If the maximum similarity value is lower than the set threshold, the current trend does not have consistent development and is regarded as an isolated disturbance or short-term fluctuation.

[0080] The specific exponential decay function is as follows:

[0081]

[0082] In the formula: This represents the trend similarity score for the k-th match distance transformation. The parameter used to adjust trend sensitivity typically ranges from 0.5 to 1.5. This represents the registration distance between the current boundary entropy change sequence and the k-th historical boundary entropy change sequence.

[0083] S4: Extract the target anomaly features corresponding to the corrected anomaly warning signal, and output the warning credibility score by combining the correction results of the preliminary anomaly warning signal.

[0084] In this embodiment, the target anomaly features corresponding to the corrected anomaly warning signal are extracted, and combined with the initial anomaly warning signal correction results, a warning confidence score is output, as follows:

[0085] Obtain the operating parameter data within the time window corresponding to the corrected abnormal warning signal, including but not limited to the voltage, current, state of charge (SOC), power and its rate of change of the energy storage node, and combine them with the synchronous phasor measurement parameters (frequency, voltage phase angle) and system event records (such as load changes, equipment switching signals, etc.) to construct a multimodal time series data input set;

[0086] The multimodal data is cross-combined to generate higher-order derived features, forming a cross-feature space;

[0087] By utilizing the constructed cross-feature space, statistical analysis or dimensionality reduction methods (such as kernel principal component analysis, isolated forest, or Shapley value method) are used to identify the feature dimensions most sensitive to changes in system behavior within the anomaly window. Key cross-features (such as SOC change rate × current change rate, power change slope × voltage phase angle difference, and voltage phase angle difference between energy storage nodes and adjacent nodes × phase angle jump synchronization) are extracted as the core input for subsequent embedding reconstruction and disturbance feature extraction, thereby improving the model's ability to distinguish under weak signal or multi-source disturbance backgrounds.

[0088] Based on kernel principal component analysis, key cross features within the anomaly window are embedded and reconstructed. The reconstruction error can quantify the degree of structural consistency disruption of the system state and output residual scores.

[0089] The disturbance behavior features in the collected phasor measurement data are extracted and a disturbance confidence score is generated by mapping through discrete counting index. The disturbance behavior features include the number of short-time abrupt changes in voltage phase angle, the duration of continuous oscillation of system frequency, and the frequency of sharp swings in voltage amplitude.

[0090] The residual score, trend similarity score, and perturbation confidence score are mapped to fuzzy linguistic variables by using fuzzy membership functions, which are defined as "low", "medium", and "high" levels. The fuzzy membership functions can be constructed using trigonometric functions, trapezoidal functions, or Gaussian functions to ensure the smooth transition of scores near the critical value.

[0091] Several fuzzy logic rules are designed based on fuzzy linguistic variables to express the reliable judgment conditions for early warnings based on expert experience, including:

[0092] If the trend consistency score is "high" and the perturbation confidence score is "high", then the confidence level is "high".

[0093] If the disturbance confidence score is "low" and the residual score is "low", then the confidence level is "low".

[0094] If the trend consistency score is "medium" and the perturbation confidence score is "low", then the confidence level is "medium".

[0095] Using fuzzy logic rules as a priori judgment criteria, a conditional probability model is constructed through a decision tree algorithm to evaluate the credible probability of abnormal events developing under the triggering of each rule;

[0096] The conditional probability weights for each rule are obtained using historical data or statistical learning methods based on maximum likelihood estimation. The judgment results of all triggering rules are then fused using a Bayesian weighting method to output the final warning credibility score.

[0097] The specific calculation formula for the fuzzy membership function is as follows:

[0098]

[0099] In the formula, The membership function value. For the current input value (such as perturbation confidence score, residual score, trend similarity score). This represents the peak position of the membership function. The standard deviation is denoted as .

[0100] It should be noted that discrete counting indexes refer to discretizable features such as the frequency or duration of system disturbances within a specific time window after event-level identification.

[0101] Example 2, Figure 2This invention presents a dynamic early warning control system for energy storage operation in a distribution area, comprising a feature extraction module, a matching module, a correction module, and an output module, with interconnections between the modules. The feature extraction module acquires the operating data of the energy storage system, constructs a data trend offset vector, and extracts trend offset features. The matching module dynamically matches the trend offset features with a pre-constructed database of perturbation anomaly features based on historical data, and outputs a preliminary anomaly early warning signal. The correction module performs perturbation analysis by acquiring synchronous phasor measurement data, identifies perturbation response features, and corrects the preliminary anomaly early warning signal by combining trend similarity. The output module extracts the target anomaly features corresponding to the corrected anomaly early warning signal, and outputs an early warning credibility score based on the correction results of the preliminary anomaly early warning signal.

[0102] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0103] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0104] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0105] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0106] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0107] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic early warning and control method for the operation of energy storage in a distribution area, characterized in that: Acquire operational data of the energy storage system, construct a data trend offset vector, and extract the trend offset features of the data trend offset vector; Specifically: Extract the rate of change and slope change of time series data, and divide the time series data into several trend candidate segments through adaptive segmentation; Time series clustering is used to process each trend candidate segment, and the trend feature vector of each segment is extracted; The difference between the trend feature vectors of any two adjacent sub-segments is measured to generate a trend change intensity value, and a significance score is output by combining the trend features. Based on the score, highly significant trend offset points are selected, adjacent offset points are connected to generate the trend offset principal axis, and the trend offset features of each offset point are extracted. The trend deviation feature is matched with a dynamically updated historical anomaly feature database. When the matching degree exceeds a preset threshold, a preliminary anomaly warning signal is output. This includes: sorting the highly significant trend shift points within the sliding time window by time along the trend shift main axis, extracting the trend shift features of each point and forming a trend shift sequence, comparing it with the abnormal sequences in the abnormal feature database, and generating a matching confidence score. Specifically, within a sliding time window, points with significance scores higher than a set threshold are selected to form a set of trend offset points; The trend offset path is generated by arranging the points in time sequence, and the relative increment of the principal axis coordinates of adjacent points is calculated to form a relative increment path. Read the baseline relative path from the historical anomaly database; Calculate the directional consistency score and principal axis increment similarity score between the relative incremental path and the baseline relative path, and output the final matching confidence score after weighted averaging. By acquiring synchronous phasor measurement data for disturbance analysis, identifying disturbance response characteristics, and combining trend similarity, the preliminary anomaly warning signal is corrected. Extract the target anomaly features corresponding to the corrected anomaly warning signal, and combine them with the correction results of the initial anomaly warning signal to output the warning credibility score.

2. The dynamic early warning and control method for the operation of energy storage in a distribution area according to claim 1, characterized in that, The process involves acquiring synchronous phasor measurement data to perform disturbance analysis, identifying disturbance response characteristics, and combining trend similarity to correct the initial anomaly warning signal, as detailed below: Acquire phasor measurement data of energy storage nodes and their adjacent topology nodes within the initial anomaly warning window, and simultaneously collect the trend of power grid boundary entropy change; Perturbation synchronization analysis is performed on the phasor measurement data of adjacent topological nodes to extract abrupt change characteristics and identify perturbation response features; The current trend of boundary entropy change is matched with the boundary entropy trajectory of historical abnormal evolution sample set by the dynamic time warping method, and the trend similarity is output. The initial anomaly warning signal is corrected by integrating disturbance response characteristics and trend similarity. The correction includes upgrading, maintaining, marking or downgrading the initial anomaly warning signal.

3. The dynamic early warning and control method for the operation of energy storage in a distribution area according to claim 2, characterized in that, The dynamic time warping method is used to match the current boundary entropy change trend with the boundary entropy trajectory of the historical anomalous evolution sample set, and output the trend similarity, as follows: During the warning period, the boundary entropy evolution sequence under the sliding window is extracted based on the key operating parameters of the energy storage system and denoted as the current boundary entropy sequence. Read the historical abnormal evolution sample set, where each sample contains a historical boundary entropy change sequence; Based on the dynamic time warping algorithm, the current boundary entropy sequence is elastically registered with the historical boundary entropy change sequence, and the corresponding matching distance is output. Each matching distance is converted into a trend similarity score using an exponential decay function, and the score with the highest trend similarity is selected. The maximum trend similarity score is compared with a set threshold to output the trend similarity.

4. The dynamic early warning and control method for the operation of energy storage in a distribution area according to claim 3, characterized in that, The extracted and corrected anomaly warning signal corresponds to the target anomaly features. Combined with the preliminary anomaly warning signal correction results, an early warning credibility score is output, including: Obtain multimodal time series data within the time window corresponding to the corrected abnormal warning signal, and construct a multimodal time series data input set; Multimodal time-series data are cross-combined to generate a cross-feature space; Key cross features are extracted from the cross feature space based on statistical analysis methods. Kernel principal component analysis is used to embed and reconstruct key cross features within the anomaly window, and residual scores are output. Perturbation behavior features are extracted from phasor measurement data and generated perturbation confidence scores by mapping discrete counting indices. By integrating residual scores and disturbance confidence scores, an early warning confidence score is output.

5. The dynamic early warning and control method for the operation of energy storage in a distribution area according to claim 4, characterized in that, The fusion of residual score and perturbation confidence score outputs an early warning confidence score, specifically: The residual score, trend similarity score, and perturbation confidence score are mapped to fuzzy linguistic variables using fuzzy membership functions; Design a fuzzy logic rule set based on domain knowledge and use it as a prior knowledge constraint; Under the constraint of prior knowledge, a conditional probability model is constructed using the decision tree algorithm; The conditional probability weight of each rule is calculated based on historical data, and the triggering rules are fused using Bayesian weighted fusion to output the final warning credibility score.

6. The dynamic early warning and control method for the operation of energy storage in a distribution area according to claim 5, characterized in that, The specific directional consistency score of the path is as follows: The specific spindle incremental similarity score is as follows: The specific matching confidence score is as follows: In the formula: It is the path direction consistency score. It is the total number of trend deviation points. This is a trend reversal direction label, indicating the direction of the trend reversal at the offset point. It is the direction of the i-th trend change point in the j-th template path. This is a consistency indicator function; it takes a value of 1 if the current direction is consistent with the template direction, and 0 otherwise. It is the principal axis incremental similarity score. It is the increment at the i-th position in the j-th template path. This is the principal axis offset increment, representing the position difference of the i-th offset point relative to the previous offset point along the principal axis. and Let be the Euclidean norm of the sequence, and be the lengths of the current path and template path increment vectors, respectively. It is the confidence score of the match between the current path and the j-th template path. and It is the weighting coefficient.

7. A system using the dynamic early warning and control method for substation energy storage operation as described in any one of claims 1-6, characterized in that, It includes a feature extraction module, a matching module, a correction module, and an output module, and the modules are interconnected. The feature extraction module is used to acquire the operating data of the energy storage system, construct the data trend offset vector, and extract the trend offset features of the data trend offset vector. The matching module is used to perform similarity matching between trend offset features and a dynamically updated historical anomaly feature database. When the matching degree exceeds a preset threshold, a preliminary anomaly warning signal is output. The correction module is used to perform disturbance analysis by acquiring synchronous phasor measurement data, identify disturbance response characteristics, and combine trend similarity to correct the preliminary anomaly warning signal; The output module is used to extract the target anomaly features corresponding to the corrected anomaly warning signal, and output the warning credibility score by combining the correction results of the initial anomaly warning signal.

Images