Energy storage power station time sequence data outlier detection method and system

Abstract

This application discloses a method and system for detecting outliers in time-series data of energy storage power stations, belonging to the field of data processing technology. The method includes: preprocessing the raw time-series data to obtain standard time-series data; performing a normality test on the standard time-series data to obtain data distribution characteristics; selecting two or three detection methods from IQR interquartile range method, criteria, and inflection point capture + IQR combination method according to the data distribution characteristics, and outputting anomaly marker candidate sets for each detection method; fusing the anomaly marker candidate sets using a weighted voting mechanism to obtain anomaly judgment results; and outputting anomaly details based on the anomaly judgment structure. This avoids the impact of misjudgment caused by a single method and improves detection reliability.

Description

Technical Field

[0001] This application belongs to the field of data processing technology, and specifically relates to a method and system for detecting outliers in time-series data of energy storage power stations. Background Technology

[0002] Time-series data (voltage, temperature, current, SOC, etc.) from energy storage power stations are the core basis for equipment health assessment, strategy optimization, and safety early warning. However, due to sensor errors, communication interference, and dynamic switching of charging and discharging conditions, the data is prone to outliers. If these outliers are not effectively removed, they will interfere with subsequent decisions such as lifetime prediction and SOC estimation. Existing detection methods mostly use fixed thresholds or single statistical rules, which are difficult to adapt to complex operating conditions and have obvious shortcomings. Specifically: First, static thresholds lack adaptability and are prone to false alarms and missed alarms. Existing fixed threshold methods are mostly based on experience to set uniform thresholds, which cannot cope with the phenomenon of data distribution drift. For example, during the daytime charging period, the battery temperature of energy storage power stations is generally higher and the voltage tends to rise, while during the nighttime discharging period, the temperature is lower and the voltage tends to fall. Static thresholds cannot adapt to the changes in data distribution caused by the difference between day and night operating conditions. When switching operating conditions, normal data is easily judged as abnormal (false alarm), or real abnormal data is missed (missed alarm).

[0003] Secondly, single detection methods have poor robustness and limited applicability. Commonly used single statistical methods each have their limitations: the IQR (interquartile range) method constructs thresholds based on the quartiles of the data, which has low sensitivity to time series data that follow a normal distribution and is difficult to identify minor anomalies; The criteria are only applicable to data that is approximately normally distributed, while most time-series data of energy storage power stations (such as current data at the moment of charge-discharge switching) are non-normally distributed, in which case the method is completely ineffective; in addition, a single method cannot meet the detection needs of different types of outliers (such as extreme outliers and abrupt change anomalies).

[0004] Third, the lack of contextual awareness makes it impossible to distinguish between reasonable inflection points and sudden anomalies. When energy storage power stations switch between charging and discharging, or experience sudden load changes, data will show natural inflection points (such as a slowdown in voltage growth at the end of charging or a sudden increase in discharge initiation current). These inflection points fall within the scope of reasonable operating condition changes. However, existing methods cannot combine the operating condition context with data change patterns, and are prone to misjudging these reasonable inflection points as sudden anomalies, further increasing the false alarm rate and increasing the ineffective workload of operation and maintenance personnel.

[0005] In summary, existing methods cannot meet the monitoring requirements of high precision, high adaptability, and low false alarm rate. There is an urgent need for an anomaly detection mechanism that adapts to data features, integrates multiple strategies, and takes into account working condition perception. Summary of the Invention

[0006] To address the aforementioned issues, this application provides a method for detecting outliers in time-series data of energy storage power stations, comprising: The raw time series data is preprocessed to obtain standard time series data; Perform normality tests on standard time series data to obtain data distribution characteristics; Based on the data distribution characteristics, the IQR interquartile range method was selected. The criteria and the inflection point capture + IQR combination method are used to select at least two detection methods, and the anomaly label candidate set of each detection method is output. An anomaly determination result is obtained by fusion of weighted voting mechanisms based on anomaly marker candidate sets; Output anomaly details based on the anomaly determination structure.

[0007] Furthermore, the raw time series data is preprocessed to obtain standard time series data, including: Remove obviously invalid data from the original time series data to obtain first-level time series data; Fill in the missing values ​​in the first-level time series data to obtain the second-level time series data; Z-score standardization was applied to the secondary time series data to obtain standard time series data.

[0008] Furthermore, a normality test is performed on the standard time series data to obtain the data distribution characteristics, including: Basic statistics are calculated based on standard time series data; Perform normality tests based on basic statistics to obtain data distribution characteristics.

[0009] Furthermore, based on the data distribution characteristics, the IQR interquartile range method was selected. The criteria and the inflection point capture + IQR combination method require at least two detection methods to output anomaly label candidate sets for each detection method, including: When the data distribution is normally distributed, the IQR interquartile range method is selected. The criteria and the inflection point capture + IQR combination method are used for detection, and the corresponding anomaly label candidate set is output; When the data distribution is non-normal, the IQR interquartile range method and the inflection point capture + IQR combination method are selected for detection, and the corresponding anomaly label candidate set is output.

[0010] Furthermore, the IQR interquartile range method includes: The upper limit of the voltage threshold is set to Q3 + 1.5 × (Q3 – Q1); The lower limit of the voltage threshold is set to Q1–1.5×(Q3–Q1); Q1 is the quartile, and Q3 is the 75th percentile.

[0011] Furthermore, The guidelines include: Set the temperature threshold to μ± ; Where μ is the mean, The standard deviation is denoted as .

[0012] Furthermore, the inflection point capture + IQR combination method includes: Calculate the second difference for standard time series data; When the second-order difference is greater than the threshold for battery model adaptation, it is marked as a potential inflection point; Identify potential inflection point windows and apply the IQR method within those windows; Calculate Q1, Q3, and IQR within the potential inflection point window.

[0013] Furthermore, the anomaly marker candidate set corresponding to each method is the Boolean marker vector output by each method.

[0014] Furthermore, a weighted voting mechanism is used to fuse the anomaly determination results based on the anomaly marker candidate set, including: Under a normal distribution, the 3σ criterion assigns a weight of 'a', the IQR interquartile range method assigns a weight of 'b', and the inflection point capture + IQR combination method assigns a weight of 'c'; where a + b + c = 1, and a > b > c; Under non-normal distribution, the IQR interquartile range method is assigned a weight of b1, and the inflection point capture + IQR combination method is assigned a weight of c1; where b1 + c1 = 1, and b1 > c1; The anomaly determination result is obtained by assigning weights to each method and using Boolean tag vectors.

[0015] Furthermore, based on the anomaly determination structure, anomaly details are output, including: Based on the anomaly determination results, the anomaly location, anomaly type, and detection method are output.

[0016] This application provides a time-series data outlier detection system for energy storage power stations, which adopts the above-mentioned time-series data outlier detection method for energy storage power stations.

[0017] Compared with the prior art, this application has the following advantages: 1. This application employs a weighted voting fusion strategy. Each detection method (IQR interquartile range, 3σ criterion, and inflection point capture + IQR combination method) outputs anomaly candidate sets (Boolean label vectors, marking whether each data point is anomaly). Weights are then dynamically allocated based on data distribution characteristics: the 3σ criterion has the highest weight under normal distribution (low false alarm rate), while the IQR method has the highest weight under non-normal distribution (strong robustness). Finally, anomalies are determined by comparing the weighted score (Σ weight × Boolean label value) with a threshold, effectively avoiding the impact of misjudgment caused by a single method and improving detection reliability.

[0018] 2. This application performs a normality test on standard time series data to accurately identify the data distribution characteristics, and dynamically selects the combination of detection methods accordingly: when the data is normal, the IQR quarter interval method, the 3σ criterion, and the inflection point capture + IQR combination method are enabled; when the data is non-normal, the 3σ criterion is automatically disabled, and the IQR quarter interval method and the inflection point capture + IQR combination method are selected to ensure that the detection method and data characteristics are always matched, and to avoid performance degradation caused by the abuse of methods.

[0019] 3. This application uses second-order difference to identify potential inflection points and applies IQR filtering within a local window to avoid misjudging normal operating condition changes as abnormalities. This method participates in weighted voting as an independent "expert," complementing the global method and achieving precise division of labor for different types of anomalies. It effectively distinguishes reasonable operating condition inflection points from real sudden anomalies, reducing the false alarm rate.

[0020] 4. This application ensures and improves data quality and reduces noise interference in the detection process by cleaning, filling missing values, and standardizing the original time-series data.

[0021] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

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

[0023] Figure 1 A schematic diagram of the steps of the method in an embodiment of the present invention is shown.

[0024] Figure 2 A flowchart illustrating the method in an embodiment of the present invention is shown.

[0025] Figure 3 A voltage distribution heatmap is shown in an embodiment of the present invention.

[0026] Figure 4 A line graph of the individual cell voltage in an embodiment of the present invention is shown. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] This application provides a method for detecting outliers in time-series data of an energy storage power station, referencing... Figure 1 ,include: S1: Preprocess the raw time series data to obtain standard time series data; S2: Perform a normality test on standard time series data to obtain data distribution characteristics; S3: Based on the data distribution characteristics, select the IQR interquartile range method. Two or three detection methods from the criteria and the inflection point capture + IQR (interquartile range) combination method are used to output the anomaly label candidate set for each detection method; S4: Based on the candidate set of anomaly markers, a weighted voting mechanism is used to fuse the results to obtain the anomaly determination result; S5: Outputs exception details based on the exception determination structure.

[0029] In step S2, the data distribution characteristics are calculated and a normality test is performed to adaptively select the detection strategy. Secondly, in step S3, an inflection point capture + IQR combination method is introduced. Second-order difference is used to identify natural inflection points caused by operating condition changes, and IQR filtering is applied within a local window to avoid misjudging reasonable operating condition changes as abnormal. Finally, in step S4, a weighted voting mechanism is adopted to dynamically allocate the weights of each method according to the data distribution, achieving multi-strategy fusion. These methods collectively improve the detection's adaptability to operating condition changes and reduce false alarms and false negatives.

[0030] This application avoids misjudging reasonable changes in operating conditions as abnormal by performing normality tests on standard time series data and introducing a combination of inflection point capture and IQR method. Furthermore, it incorporates a weighted voting mechanism to improve the adaptability of detection to changes in operating conditions and reduce false alarms and false negatives.

[0031] The raw time series data is preprocessed to obtain standard time series data, including: Remove obviously invalid data from the original time series data to obtain first-level time series data; Fill in the missing values ​​in the first-level time series data to obtain the second-level time series data; Z-score standardization was applied to the secondary time series data to obtain standard time series data.

[0032] The input raw time-series data undergoes cleaning, missing value imputation, and standardization to ensure data quality. Specifically, this involves first removing obviously invalid data (such as non-numerical values ​​or values ​​exceeding physical limits); second, for missing values, linear interpolation or mean-based imputation methods based on neighboring data are used to ensure sequence continuity; finally, standardization is performed, for example using Z-score standardization, to eliminate dimensional differences and ensure consistency in subsequent analyses. This step aims to improve data quality and reduce noise interference in the detection process.

[0033] This application ensures and improves data quality and reduces noise interference in the detection process by cleaning, filling missing values, and standardizing the original time-series data.

[0034] In one embodiment of the present invention, a normality test is performed based on standard time series data to obtain data distribution characteristics, including: Basic statistics are calculated based on standard time series data; Perform normality tests based on basic statistics to obtain data distribution characteristics.

[0035] Calculate the basic statistics (mean μ, standard deviation) The data is analyzed by identifying the quartiles (Q1 / Q3) and performing a normality test (such as the Shapiro-Wilk test) to determine whether the data follows an approximately normal distribution. The result of this test will guide the selection of subsequent strategies, ensuring that the detection method matches the data characteristics. The distribution characteristics (e.g., "normal" or "non-normal") are then determined based on whether the time series data follows an approximately normal distribution. These characteristics will guide the selection of the detection method in step S3 and the allocation of weights in step S4.

[0036] Based on the data distribution characteristics, the IQR interquartile range method was selected. The method combines two or three detection methods from the criteria and inflection point capture + IQR approach, outputting anomaly marker candidate sets for each detection method, including: When the data distribution is normally distributed, the IQR interquartile range method is selected. The criteria and the inflection point capture + IQR combination method are used for detection, and the corresponding anomaly label candidate set is output; When the data distribution is non-normal, the IQR interquartile range method and the inflection point capture + IQR combination method are selected for detection, and the corresponding anomaly label candidate set is output.

[0037] The IQR interquartile range (IQR) method outputs the candidate set of anomaly markers. Criterion Output The candidate set of anomaly labels for the criteria; the output of the candidate set of anomaly labels for the inflection point capture + IQR combination method.

[0038] In one embodiment of the present invention, the IQR interquartile range method includes: The upper limit of the voltage threshold is set to Q3 + 1.5 × (Q3 – Q1); The lower limit of the voltage threshold is set to Q1–1.5×(Q3–Q1); Q1 is the quartile, and Q3 is the 75th percentile.

[0039] The IQR (Interquartile Range) method accurately marks outliers in data such as voltage and temperature.

[0040] The guidelines include: Set the temperature threshold to μ± ; Where μ is the mean, The standard deviation is denoted as .

[0041] The criteria are applicable to anomaly detection in approximately normally distributed data such as temperature uniformity, and exhibit high accuracy when standard time-series data follows a normal distribution.

[0042] The inflection point capture + IQR combination method includes: Calculate the second difference for standard time series data; When the second-order difference is greater than the threshold for battery model adaptation, it is marked as a potential inflection point; Identify potential inflection point windows and apply the IQR method within those windows; Calculate Q1, Q3, and IQR within the potential inflection point window.

[0043] For scenarios involving switching between charging and discharging terminals, the first step is to analyze the voltage standard time-series data sequence. (i=1,2,...,n), first calculate the second-order difference:

[0044] when When the threshold is reached, it is marked as a potential inflection point, where G is a threshold adapted to the battery model. (Specifically, it is determined by multiplying the maximum second-order difference at the end of the historical charge and discharge cycles of the target battery model by a safety factor of 1.2 to 1.5, balancing the sensitivity of inflection point capture with the ability to resist disturbances.) Subsequently, the IQR method is applied to filter noise within a window near the inflection point (e.g., k data points before and after the inflection point, where k can be set to 5-10 depending on the sequence length): Q1, Q3, and IQR of the window data are calculated, and outliers are defined as values ​​exceeding [Q1-1.5×IQR, Q3+1.5×IQR], thereby reducing false alarms due to disturbances during operating condition switching (such as charge-discharge conversion).

[0045] The anomaly marker candidate set corresponding to each method is the Boolean marker vector output by each method.

[0046] The Boolean flag vector outputs either 1 or 0 for each method on each standard time series data.

[0047] In one embodiment of the present invention, reference is made to Figure 2 An anomaly determination result is obtained by fusion of a weighted voting mechanism based on anomaly marker candidate sets, including: Under a normal distribution, the 3σ criterion assigns a weight of 'a', the IQR interquartile range method assigns a weight of 'b', and the inflection point capture + IQR combination method assigns a weight of 'c'; where a + b + c = 1, and a > b > c; Under non-normal distribution, the IQR interquartile range method is assigned a weight of b1, and the inflection point capture + IQR combination method is assigned a weight of c1; where b1 + c1 = 1, and b1 > c1; The anomaly determination result is obtained by assigning weights to each method and using Boolean tag vectors.

[0048] A weighted voting mechanism is used to fuse the Boolean label vectors (i.e., the anomaly candidate set, with each method outputting 1 or 0 for each data point) output by Method A (IQR interquartile range), Method B (3σ criterion), and Method C (inflection point capture + IQR combination method) in step S3. Specifically, weights are assigned based on distribution characteristics: the weights are determined as follows: under a normal distribution, the 3σ criterion has the lowest theoretical false alarm rate (approximately 0.3%), so it is assigned the highest weight of 0.5; the IQR interquartile range method is slightly less efficient for normally distributed data, so it is assigned 0.3; the inflection point capture + IQR combination method is used as an auxiliary method, so it is assigned 0.2. Under non-normal distributions, the IQR method is more robust than 3σ, so the IQR interquartile range method has a weight of 0.6, the inflection point capture + IQR combination method has a weight of 0.4, and the 3σ criterion is disabled. The above weight values ​​can be fine-tuned based on actual application scenarios through historical data calibration or expert experience, and can be configured in the form of interval values ​​(e.g., the 3σ criterion weight can be set to 0.45~0.55). The specific values ​​given in this embodiment are typical reference values. For each data point, calculate its final anomaly score = Σ(method weight × whether it is anomaly (Boolean label vector)), where "whether it is anomaly" is 1 (anomaly) or 0 (normal). When the score > 0.5, it is judged as anomaly.

[0049] In one embodiment of the present invention, the output of anomaly details based on the anomaly determination structure includes: Based on the anomaly determination results, the anomaly location, anomaly type, and detection method are output.

[0050] Output the location of outliers, their type (outlier / mutation), the detection method used, and their confidence level. The output format is JSON structured data. The outlier type is determined as follows: if a data point is detected only by IQR interquartile range or... If a data point is marked as an anomaly by the criteria (a static method based on global thresholds) and is not marked within the inflection point window by the inflection point capture + IQR combination method, it is determined to be an "outlier". If the standard time series data point is marked as an anomaly by the inflection point capture + IQR combination method (i.e., triggered in a local IQR near the inflection point), or is marked by multiple methods simultaneously and is located in the condition switching zone, it is determined to be a "mutation point". The confidence level is based on the fusion score: 100% confidence level when the score = 1, 75% confidence level when the score = 0.75, and so on.

[0051] Specifically, for a 100MW / 200MWh energy storage power station, which includes 200 battery compartments and approximately 50,000 individual cells, the system workflow is as follows: Data Acquisition and Preprocessing (S1): Obtain raw time-series data such as voltage and temperature in real time from the big data monitoring platform, updating every second; perform cleaning, linear interpolation to fill missing values ​​and Z-score standardization to ensure data quality and obtain standard time-series data.

[0052] Data distribution characteristic analysis (S2): Perform the Shapiro-Wilk test on the temperature standard time series data (p=0.12>0.05). The p-value represents the probability that the null hypothesis (the standard time series data follows a normal distribution) is true. Generally, p>0.05 indicates that the data is approximately normally distributed. In this example, p=0.12>0.05, therefore the temperature standard time series data is determined to be normally distributed; enable... Criterion (μ=35℃, =2℃), identify the 42℃ anomaly; use IQR method (Q1=3.18V, Q3=3.22V) on voltage standard timing data to identify the 3.10V outlier; during charge / discharge switching, the second-order differential peak value 0.05V>G=0.03V triggers inflection point detection, and local IQR is applied within a ±10 second window to eliminate normal fluctuations.

[0053] Adaptive Multi-Strategy Anomaly Detection (S3): Dynamically Enable IQR Using the inflection point + IQR method, an anomaly candidate set is generated; a front-end heatmap displays the voltage distribution (see reference). Figure 3 (Battery compartment view), abnormal cells are highlighted in red; click to display a 24-hour line chart (reference). Figure 4 The anomaly is indicated by the message "Voltage outlier (IQR method, confidence level 0.75)".

[0054] Dynamic Fusion of Outlier Results (S4): Weighted Voting Fusion Output (Normal Data: Criterion 0.5, IQR interquartile range method 0.3, inflection point capture + IQR combination method 0.2; non-normal: IQR interquartile range method 0.6, inflection point capture + IQR combination method 0.4); score > 0.5 confirms anomaly; 300 abnormal individuals on the same day (accounting for 0.6%), mark all abnormal points, trigger a yellow warning.

[0055] This application provides an outlier detection system for time-series data of energy storage power stations, which adopts the above-mentioned outlier detection method for time-series data of energy storage power stations. The system cleans the original time-series data, performs normality tests on the standard time-series data, and introduces a combination of inflection point capture and IQR method to avoid misjudging reasonable changes in operating conditions as anomalies. Furthermore, it integrates a weighted voting mechanism to improve the adaptability of detection to changes in operating conditions and reduce false alarms and false negatives.

[0056] Although this application 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; and these 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 this application.

Claims

1. A method for detecting outliers in time-series data of an energy storage power station, characterized in that, include: The raw time series data is preprocessed to obtain standard time series data; Perform normality tests on standard time series data to obtain data distribution characteristics; Based on the data distribution characteristics, the IQR interquartile range method was selected. The criteria and the inflection point capture + IQR combination method are used to select at least two detection methods, and the anomaly label candidate set of each detection method is output. An anomaly determination result is obtained by fusion of weighted voting mechanisms based on anomaly marker candidate sets; Output anomaly details based on the anomaly determination structure.

2. The method for detecting outliers in time-series data of an energy storage power station according to claim 1, characterized in that, The raw time series data is preprocessed to obtain standard time series data, including: Remove obviously invalid data from the original time series data to obtain first-level time series data; Fill in the missing values ​​in the first-level time series data to obtain the second-level time series data; Z-score standardization was applied to the secondary time series data to obtain standard time series data.

3. The method for detecting outliers in time-series data of an energy storage power station according to claim 2, characterized in that, Perform normality tests on standard time series data to obtain data distribution characteristics, including: Basic statistics are calculated based on standard time series data; Perform normality tests based on basic statistics to obtain data distribution characteristics.

4. The method for detecting outliers in time-series data of an energy storage power station according to claim 3, characterized in that, Based on the data distribution characteristics, the IQR interquartile range method was selected. The criteria and the inflection point capture + IQR combination method require at least two detection methods to output anomaly label candidate sets for each detection method, including: When the data distribution is normally distributed, the IQR interquartile range method is selected. The criteria and the inflection point capture + IQR combination method are used for detection, and the corresponding anomaly label candidate set is output; When the data distribution is non-normal, the IQR interquartile range method and the inflection point capture + IQR combination method are selected for detection, and the corresponding anomaly label candidate set is output.

5. The method for detecting outliers in time-series data of an energy storage power station according to claim 3, characterized in that, IQR interquartile range method includes: The upper limit of the voltage threshold is set to Q3 + 1.5 × (Q3 – Q1); The lower limit of the voltage threshold is set to Q1–1.5×(Q3–Q1); Q1 is the quartile, and Q3 is the 75th percentile.

6. The method for detecting outliers in time-series data of an energy storage power station according to claim 3, characterized in that, The guidelines include: Set the temperature threshold to μ± ; Where μ is the mean, The standard deviation is denoted as .

7. The method for detecting outliers in time-series data of an energy storage power station according to claim 3, characterized in that, The inflection point capture + IQR combination method includes: Calculate the second difference for standard time series data; When the second-order difference is greater than the threshold for battery model adaptation, it is marked as a potential inflection point; Identify potential inflection point windows and apply the IQR method within those windows; Calculate Q1, Q3, and IQR within the potential inflection point window.

8. The method for detecting outliers in time-series data of an energy storage power station according to claim 6, characterized in that, The anomaly marker candidate set corresponding to each method is the Boolean marker vector output by each method.

9. The method for detecting outliers in time-series data of an energy storage power station according to claim 8, characterized in that, An anomaly determination result is obtained by fusion of a weighted voting mechanism based on anomaly marker candidate sets, including: Under a normal distribution, the 3σ criterion assigns a weight of 'a', the IQR interquartile range method assigns a weight of 'b', and the inflection point capture + IQR combination method assigns a weight of 'c'; where a + b + c = 1, and a > b > c; Under non-normal distribution, the IQR interquartile range method is assigned a weight of b1, and the inflection point capture + IQR combination method is assigned a weight of c1; where b1 + c1 = 1, and b1 > c1; The anomaly determination result is obtained by assigning weights to each method and using Boolean tag vectors.

10. The method for detecting outliers in time-series data of an energy storage power station according to claim 8, characterized in that, The anomaly detection structure outputs anomaly details, including: Based on the anomaly determination results, the anomaly location, anomaly type, and detection method are output.

11. A time-series data outlier detection system for an energy storage power station, characterized in that, The method for detecting outliers in time-series data of an energy storage power station is described in any one of claims 1-10.

Images