Wind farm voltage imbalance treatment method based on working condition constraint and cause matching

Abstract

The application relates to the technical field of wind farm voltage treatment, and discloses a wind farm voltage imbalance treatment method based on working condition constraint and cause matching. The method comprises the following steps: acquiring wind farm multi-source operation data to construct a voltage imbalance feature vector and a time sequence; determining an operation working condition category and a working condition deviation degree based on working condition constraint features, and establishing a working condition parameter set; adaptively decomposing the voltage imbalance degree time sequence by using a CEEMDAN method, calculating an energy proportion, and determining a disturbance type; constructing a cause discrimination evidence set by using the working condition parameter set, the energy proportion and the disturbance type, calculating a fusion confidence of each cause type by using D-S evidence theory fusion, and discriminating a dominant cause; matching a treatment object according to the dominant cause, the fusion confidence and the working condition category, generating a treatment strategy, and dynamically updating the strategy according to a treatment improvement rate, a voltage recovery time and an action cost. The application improves the accuracy of voltage imbalance cause discrimination under different operation working conditions.

Description

Technical Field

[0001] This application relates to the field of wind farm voltage management technology, and in particular to a wind farm voltage imbalance management method based on operating condition constraints and cause matching. Background Technology

[0002] With the increasing penetration of wind power in the power system, power quality issues at wind farm grid connection points are receiving growing attention. Three-phase voltage imbalance is a significant factor affecting the safe and stable operation of wind farms, and in severe cases, it can lead to wind turbine disconnection from the grid, relay protection malfunctions, abnormal equipment overheating, and even damage. Unlike conventional synchronous generators, wind turbines often use power electronic converters for grid connection, making the causes of voltage imbalance more complex. These imbalances can originate from asymmetrical faults or load imbalances on the grid side, or be induced by factors within the wind farm such as differences in turbine parameters, asymmetrical line impedance, uneven spatial and temporal wind speed distribution, and inappropriate reactive power control strategies. Accurately identifying the dominant causes of imbalance and implementing targeted solutions is crucial for improving the operational reliability of wind farms.

[0003] Currently, wind farm voltage imbalance management mainly relies on set threshold judgments and empirical rules, which has the following shortcomings: It lacks adaptive identification capabilities for wind farm operating conditions; the response characteristics of voltage imbalance vary significantly under different wind speed ranges, active power output levels, and reactive power margins, making fixed-parameter analysis methods difficult to adapt to wide operating condition variations; the disturbance decomposition of the voltage imbalance time series is not refined enough, often mixing transient disturbances, periodic fluctuations, and persistent imbalances, leading to ambiguity in cause identification; cause identification is mostly based on single physical quantities such as negative sequence voltage amplitude or simple logical thresholds, failing to integrate multi-source evidence for uncertainty reasoning, which easily leads to misjudgment or omission in complex scenarios; the management strategy is decoupled from the cause type and operating conditions, and strategy generation lacks dynamic evaluation and self-updating mechanisms, making it difficult to balance management effectiveness and cost.

[0004] Therefore, there is an urgent need for a wind farm voltage imbalance management method that can integrate operating condition constraints, adaptively decompose disturbance components, perform cause matching based on multi-evidence fusion, and dynamically generate optimized management strategies to improve the accuracy, robustness, and economy of management. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching, which improves the accuracy of voltage imbalance cause identification under different operating conditions.

[0006] In a first aspect, this application provides a method for mitigating voltage imbalance in wind farms based on matching operating conditions with causes, the method comprising: Step S1: Obtain multi-source operation data of the wind farm at different sampling periods, obtain multiple samples to be identified, and preprocess them to obtain a standardized operation dataset. Based on the standardized operation dataset, construct a voltage imbalance feature vector and obtain a voltage imbalance time series. Step S2: Obtain the wind farm operating condition constraint features of each sample to be identified based on the voltage imbalance feature vector, determine the operating condition category and operating condition deviation of the sample to be identified, and establish the corresponding operating condition parameter set. Step S3: Based on the set of operating parameters, the CEEMDAN method is used to adaptively decompose the voltage imbalance time series to obtain multiple disturbance components, and the energy proportion of each disturbance component is calculated to determine the corresponding disturbance type. Step S4: Based on the set of operating parameters, energy ratio and disturbance type of the sample to be identified, construct a set of evidence for cause discrimination, calculate the fusion confidence of each cause type based on the fusion method of DS evidence theory, and determine the dominant cause type that leads to voltage imbalance in the sample to be identified. Step S5: Match the governance objects according to the dominant cause type, fusion confidence level and operating condition category, determine the evaluation weight of the governance strategy, generate the corresponding voltage imbalance governance strategy, and update the voltage imbalance governance strategy according to the governance improvement rate, voltage recovery time and governance action cost.

[0007] Compared with the prior art, the beneficial effects of the present invention are at least as follows: This application first constructs feature vectors from multi-source wind farm operation data and performs DBSCAN density clustering analysis to accurately extract multi-dimensional parameters such as active power, reactive power, and wind speed fluctuations as operating condition constraint features. By calculating the operating condition deviation and establishing a dedicated set of operating condition parameters, it effectively overcomes the shortcomings of traditional analysis methods that use a single fixed threshold, resulting in poor scenario adaptability. This provides a dynamic data benchmark that highly matches the actual operating state for voltage imbalance management under complex and variable wind conditions.

[0008] Secondly, by combining the operating condition parameter set with the CEEMDAN adaptive time-series decomposition technique and dynamically adjusting the introduced noise amplitude based on the operating condition deviation, the mode aliasing problem in non-stationary power grid signal processing is effectively solved, and transient, periodic, and persistent disturbance components are accurately separated and quantified. Based on this, the isolated forest algorithm is used to extract anomaly scores to construct a causal evidence set, and the DS evidence theory is introduced to calculate the conflict coefficient and cross-fusion degree of multi-source probabilities. This achieves high-confidence and accurate source tracing of single or compound dominant causes such as wind speed disturbances, reactive power limitation, and equipment anomalies, overcoming the limitation of conventional methods in diagnosing coupled faults.

[0009] Finally, this application constructs a comprehensive evaluation mechanism that integrates voltage suppression effectiveness, operational costs, and operational risks. It adaptively assigns differentiated evaluation weights based on operating conditions and dominant causes, achieving precise matching between governance strategies and dynamic causes. Simultaneously, it establishes a strategy effectiveness scoring and closed-loop feedback update architecture based on key indicators such as governance improvement rate and voltage recovery time, ensuring continuous optimization of the control strategy as it evolves with historical samples. This provides a system-level solution that balances high reliability and economy for the refined governance of complex power quality in wind farms. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart of the wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching in the embodiments of this application; Figure 2 This is an overall structural diagram of the wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching in the embodiments of this application. Figure 3 This is a schematic diagram of multi-source operational data preprocessing in an embodiment of this application; Figure 4 This is a schematic diagram illustrating the identification of the dominant causes of voltage imbalance and the matching of treatment targets in the embodiments of this application; Figure 5 This is a schematic diagram comparing the governance effects under different governance strategies in the embodiments of this application. Detailed Implementation

[0012] This application provides a method for mitigating voltage imbalance in wind farms based on operating condition constraints and cause matching. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0013] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 and Figure 2 The wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching in this application includes: Step S1: Obtain multi-source operation data of the wind farm at different sampling periods, obtain multiple samples to be identified, and preprocess them to obtain a standardized operation dataset. Construct a voltage imbalance feature vector based on the standardized operation dataset and obtain a voltage imbalance time series.

[0014] Specifically, the multi-source operation data acquisition and preprocessing of wind farms, such as... Figure 3 As shown, firstly, multi-source operational data from the wind farm at different sampling periods are acquired to obtain multiple samples to be identified. The multi-source operational data includes the active power output, reactive power output, three-phase voltage, three-phase current, grid connection point voltage, operating status of the dynamic reactive power compensation device, operating status of the box-type transformer, operating status of the collector line, and meteorological data. The meteorological data includes wind speed, wind direction, and ambient temperature; the operating status of the dynamic reactive power compensation device includes the reactive power output, switching status, and capacity margin of the static var generator (SVG) or static var compensator (SVC). All multi-source operational data collected in each sampling period are aligned by time and combined into a single original sample record; different sampling periods correspond to different samples, thus forming multiple samples to be identified for subsequent processing.

[0015] The collected raw data undergoes preprocessing, starting with data cleaning to remove or repair missing values, outliers, and noisy data. Let the original dataset be: ,in, The original dataset, where m represents the number of samples and n represents the number of features. This represents the j-th feature value of the i-th sample; for missing values ​​in the sample, the nearest-neighbor mean imputation method is used to repair them, and the repaired feature value is... For outliers in the sample, based on the mean and standard deviation of the feature data, using 3 The criteria were initially eliminated: ,in, and Let and represent the mean and standard deviation of the j-th feature, respectively. Since different types of multi-source operational data have different sampling periods during acquisition—for example, three-phase voltage and grid connection point voltage are sampled at the second level, while meteorological data are sampled at the minute level—to ensure joint analysis between different data, the sampling time of the grid connection point voltage is used as the reference time to align the wind turbine output data, dynamic reactive power compensation device operating status data, and meteorological data in time. Specifically, for data with a sampling period shorter than or equal to 1 second, a linear interpolation method is used to synchronize them to one data point every 0.2 seconds; for data with a sampling period longer than 1 second, a forward hold method is used, i.e., the previous time value is filled to the point before the next sampling time for alignment, and uniformly resampled to one data point per second, thus forming multi-source feature samples within the same time window.

[0016] After handling missing and outlier values, data filtering is performed according to the requirements of the voltage imbalance analysis task. Let the correlation coefficient between the j-th feature and the target variable be... The filtering rules are: ,in, To obtain cleaned and filtered sample data, features that meet the set correlation threshold are retained. The target variable is voltage imbalance, and the correlation coefficient is calculated using the Pearson correlation coefficient. (Setting the correlation threshold) =0.3, meaning features with an absolute value of correlation coefficient greater than or equal to 0.3 with respect to voltage imbalance are retained, while other features are discarded. The cleaned and filtered sample data are then uniformly mapped to the same numerical range for normalization to eliminate the influence of differences in feature dimensions on subsequent identification results. In this process, the minimum-maximum normalization method can be used. ,in, and Let represent the minimum and maximum values ​​of the j-th feature, respectively, or we can use the Z-score normalization method: The data is processed using the mean and standard deviation of the features. After missing value repair, outlier removal, time alignment, and normalization, it is possible to ensure that the wind turbine output, grid connection voltage, reactive power compensation device status, and meteorological disturbance data have a strict correspondence within the same time window, thereby obtaining a standardized operation dataset for characterizing the wind farm's operating status.

[0017] Step S1, obtaining the voltage imbalance time series, includes extracting the voltage imbalance based on a standardized operating dataset, including the ratio of negative-sequence voltage to positive-sequence voltage, and constructing a voltage imbalance feature vector by combining phase-to-phase voltage deviation, phase sequence offset, negative-sequence current ratio, active power change rate, reactive power change rate, reactive power regulation margin, grid connection point voltage change rate, and wind speed fluctuation rate. Based on the voltage imbalance in the voltage imbalance feature vector, the voltage imbalance time series is obtained according to the sampling time point order.

[0018] Specifically, based on a standardized operational dataset, characteristic parameters representing the voltage imbalance state of wind farms are extracted. These voltage imbalance parameters include voltage imbalance degree, negative-sequence voltage component, phase-to-phase voltage deviation, phase sequence offset, negative-sequence current ratio, active power change rate, reactive power change rate, reactive power regulation margin, grid connection point voltage change rate, and wind speed fluctuation rate. Among these, the voltage imbalance degree is represented by the ratio of negative-sequence voltage to positive-sequence voltage. ,in, Represents the positive sequence voltage component. Represents the negative sequence voltage component. This indicates the degree of voltage imbalance. The zero-sequence, positive-sequence, and negative-sequence components of the voltage can be determined using the symmetrical component method. Obtain, among which, , , They represent voltages, , , These represent the zero-sequence, positive-sequence, and negative-sequence voltage components, respectively. Where a is the rotation factor, j is the imaginary unit, and the phase-to-phase voltage deviation is expressed as... Reflecting the degree of difference between the three-phase line voltages, the phase sequence offset can be expressed as: ,in, , , These represent the fundamental phase angles of the three-phase voltages, reflecting the degree of deviation in the voltage phase relationship. The proportion of negative sequence current, used to reflect the degree of current asymmetry, is also extracted. : ,in, Represents the positive sequence current component. This represents the negative sequence current component.

[0019] Furthermore, to fully reflect the actual operating conditions and external disturbance characteristics of the wind farm, the characteristic parameters also include the active power change rate, reactive power change rate, reactive power regulation margin, grid connection point voltage change rate, and wind speed fluctuation rate. Based on the difference between the current data and the previous sampling data in the standardized operating dataset, and combined with the sampling time interval, the active power change rate, used to characterize the wind turbine output ramp-up, is calculated and expressed as: ,in, This indicates the active power of the wind turbine at the current moment. Indicates time The active power of the wind turbine unit The time interval represents the time difference between two power samples; the reactive power change rate, used to characterize the adjustment changes of wind turbine generators or dynamic reactive power compensation devices, is calculated and expressed as: , This indicates the reactive power of the wind turbine at the current moment. Indicates time The reactive power; calculate the reactive power margin used to characterize the residual reactive power support capacity of wind turbine generators and dynamic reactive power compensation devices, which is expressed as: , This represents the maximum reactive power regulation capacity that the dynamic reactive power compensation device or wind turbine can provide; the grid connection point voltage change rate, used to characterize the voltage fluctuation trend at the grid connection point, is calculated and expressed as: , This indicates the voltage at the grid connection point at the current moment. Indicates the previous moment The grid connection point voltage; in addition, the wind speed fluctuation rate, used to characterize the impact of external wind resource disturbances on unit output and voltage status, is calculated and expressed as: , Indicates the wind speed at the current moment. Indicates time The wind speed at that moment. The extracted feature parameters are combined to construct a voltage imbalance feature vector: ,in, This represents the voltage imbalance feature vector of the i-th sample, where the phase-to-phase voltage deviation... Phase sequence offset and the proportion of negative sequence current As an indicator characterizing the ontological features of electrical imbalance, it is used to assist in determining the dominant cause of equipment anomalies in subsequent steps. Furthermore, based on the constructed set of voltage imbalance feature vectors, the voltage imbalance degree is extracted and continuously arranged according to the sampling time point order to obtain a one-dimensional voltage imbalance degree time series for multi-scale feature decomposition. This provides a complete and accurate input foundation for subsequent operating condition identification, multi-time-scale feature decomposition, and determination of the dominant cause of voltage imbalance in wind farms.

[0020] This step involves preprocessing the multi-source operational data of the wind farm, including cleaning, time alignment, and normalization, to obtain a standardized operational dataset. Based on this, unbalance parameters of the electrical system and external operating condition disturbance characteristics are further extracted and fused to construct a voltage unbalance feature vector and time series. This lays a solid and reliable data and feature foundation for subsequent accurate identification of operating conditions, multi-scale disturbance component analysis, and fusion discrimination of dominant causes.

[0021] Step S2: Obtain the wind farm operating condition constraint features of each sample to be identified based on the voltage imbalance feature vector, determine the operating condition category and operating condition deviation of the sample to be identified, and establish the corresponding operating condition parameter set.

[0022] Step S2 further includes extracting the active power change rate, reactive power change rate, wind speed fluctuation rate, grid connection point voltage change rate, and reactive power adjustment margin from the voltage imbalance feature vector as operating condition constraint features. For each sample to be identified, the DBSCAN density clustering method is used to cluster the operating condition constraint features to determine the operating condition category, and the corresponding distance relative to the center point of the operating condition category is calculated as the operating condition deviation. Based on the operating condition category and the operating condition deviation, the corresponding operating condition parameter set is generated.

[0023] Specifically, after obtaining the voltage imbalance feature vector, the operating condition identification process is executed. First, the active power change rate, reactive power change rate, wind speed fluctuation rate, grid connection point voltage change rate, and reactive power regulation margin are extracted from the voltage imbalance feature vector as operating condition constraint features for each sample to be identified. These five feature parameters are then combined to form the operating condition constraint vector: ,in, Let represent the operating condition constraint vector of the i-th sample. For each sample to be identified, the DBSCAN density clustering method is used to perform cluster analysis on the operating condition constraint vector. First, based on the distribution characteristics of the measured data, the optimal neighborhood radius is determined using the k-distance curve method. and minimum sample size Specifically, based on the five characteristic dimensions of the operating condition constraint vector, it is assumed that... Set the value to a constant 2 × 5 = 10, and then calculate the entire dataset. Each sample point in the middle to its first The Euclidean distances of the nearest neighbor sample points are calculated, and the distance values ​​of all sample points are arranged in descending order to plot a k-distance curve. The distance value corresponding to the inflection point of the curve with the most dramatic change in slope is taken as the optimal neighborhood radius. If the clustering results cannot effectively distinguish between the four operating conditions—normal fluctuations, power output ramp-up, wind speed fluctuations, and reactive power limitation—then appropriate adjustments should be made. Repeat the above process until reasonable clustering parameters are obtained. After determining the parameters, define the sample points. of - Neighborhood: ,in, Represents sample points to sample point distance, To represent a point of - Neighborhood, For the entire dataset. When: At that time, the current specific sample point is determined as the core point.

[0024] Through the density clustering process described above, the operating states of wind farms are divided into normal fluctuation conditions, power output ramp-up conditions, wind speed fluctuation conditions, and reactive power limitation conditions. After obtaining several clusters through DBSCAN density clustering, the active power change rate is calculated for each cluster. Reactive power change rate Wind speed fluctuation rate Grid connection point voltage change rate and reactive power regulation margin The average of these five characteristics is used, and a threshold is set based on the physical laws of wind power operation: when the reactive power regulation margin within a cluster is significantly insufficient and the grid connection voltage fluctuates significantly, it is judged as a reactive power limited operating condition; if the wind speed changes significantly and causes violent fluctuations in active power output, i.e., the dispersion is large, it is judged as a wind speed fluctuation operating condition; when the active power output changes drastically, and the direction of change shows a highly consistent monotonicity within the cluster, it is judged as an output ramping operating condition, and further subdivided into upward or downward ramping based on the direction of output increase or decrease; if none of the above characteristics are significant, i.e., the rate of change of each state remains stable and the reactive power regulation margin is very sufficient, it is classified as a normal fluctuation operating condition. In addition, outlier noise points that are not classified into any cluster by the clustering algorithm are temporarily classified as other transient operating conditions. Through the above rule extraction, the massive samples of wind farms can be automatically classified into four typical operating states, providing a reliable basis for subsequent voltage imbalance early warning and control.

[0025] After identifying the operating condition category, the distance between each sample to be identified and the center point of its corresponding operating condition category is calculated and used as the operating condition deviation. First, extract all sample points assigned to the q-th cluster. Then, calculate the arithmetic mean of the operating condition constraint vector containing 5 feature parameters along each feature dimension to generate a multi-dimensional mean vector as the operating condition center of this cluster. Subsequently, the Euclidean distance calculation method is used to determine the current sample point to be identified. The constraint vector and the center of the working condition class The square root of the sum of the squared differences between the various feature dimensions is used as the deviation of the working condition for that sample. .in, Indicates the deviation from operating conditions. This represents the condition center for the q-th type of operating condition. The value is the arithmetic mean of the operating condition constraint vectors of all sample points within the q-th cluster. The operating condition deviation is used to characterize the degree of deviation between the current sample and its typical operating condition. When the operating condition deviation is small, it indicates that the characteristics of the current sample are relatively close to those of the typical operating condition; when the operating condition deviation is large, it indicates that the current sample may have abnormal fluctuations or complex disturbances superimposed on the basic operating condition.

[0026] Furthermore, based on the identified operating condition categories and deviations, a set of operating condition parameters corresponding to each operating condition category is established. In addition to the operating condition category and deviation, the set of operating condition parameters includes the clustering neighborhood radius, minimum sample size, disturbance decomposition layer number, noise amplitude, disturbance component discrimination threshold, cause discrimination threshold, and governance strategy evaluation weight for each corresponding operating condition category. Specifically, the clustering neighborhood radius and minimum sample size characterize the data distribution characteristics of the current operating condition category; the disturbance decomposition layer number and noise amplitude are used for subsequent decomposition of the voltage imbalance time series; the disturbance component discrimination threshold is used for subsequent operating condition perception classification of transient disturbances, periodic fluctuations, and persistent imbalance components; the cause discrimination threshold is used for subsequent cause discrimination of wind speed disturbance type, reactive power limitation type, equipment anomaly type, and composite anomaly type; and the governance strategy evaluation weight is used for subsequent comprehensive evaluation and priority update of governance strategies. The internal parameter configuration of the operating condition parameter set varies with the operating condition category and deviation. Adaptive adjustment directly calls the corresponding basic parameter template for different operating conditions and utilizes the deviation of the operating condition. Linear scaling is applied to various thresholds and weights. The specific mapping adjustment rules are as follows: ,in, This represents the final parameters after adaptive adjustment. This indicates the basic parameter template value invoked under the current operating condition. This is the adjustment coefficient.

[0027] For normal fluctuation conditions, the corresponding basic parameter templates are invoked to set lower anomaly sensitivity and stricter persistent imbalance discrimination conditions. For power output ramp-up conditions, the basic weight of active power output change rate and grid connection point voltage change rate in cause discrimination is increased. For wind speed fluctuation conditions, the sensitivity of transient disturbance component identification is increased, and the transient disturbance duration and energy proportion thresholds are adjusted accordingly. For reactive power constrained conditions, the basic weight of reactive power regulation margin and negative sequence voltage component in cause discrimination and strategy evaluation is increased. For other transient conditions, such as grid fault conditions and system oscillation conditions, when it is a grid fault condition, the discrimination thresholds for transient disturbance duration and energy proportion are reduced, while the basic weight of grid connection point voltage change rate and negative sequence voltage component is significantly increased in cause discrimination. When it is a system oscillation condition, the basic weight of periodic fluctuation energy proportion and duration in cause discrimination is increased.

[0028] This step does not directly apply a uniform risk assessment to all samples. Instead, based on a clear understanding of the operating condition categories, it further generates a set of operating condition parameters and transfers these parameters to subsequent disturbance component analysis, dominant cause identification, and governance strategy evaluation processes. This ensures that the various processing steps are not merely sequential in terms of workflow, but also form a parameter linkage driven by the operating condition categories.

[0029] Step S3: Based on the set of operating parameters, the CEEMDAN method is used to adaptively decompose the voltage imbalance time series to obtain multiple disturbance components, calculate the energy proportion of each disturbance component, and determine the corresponding disturbance type.

[0030] Step S3, which uses the CEEMDAN method to perform adaptive disturbance component decomposition on the voltage imbalance time series, includes: decomposing the voltage imbalance time series using the CEEMDAN time series decomposition method to obtain multiple intrinsic mode functions and residual components. ,in, This is a time series of voltage imbalance. Denotes the k-th eigenmode function. Represents the residual component. This represents the decomposition level under the q-th operating condition, which is determined by the set of operating condition parameters; the noise amplitude added during the decomposition process. According to the deviation of the working conditions The noise amplitude is adaptively adjusted using the following formula: ,in, This represents the adjusted noise amplitude under the q-th operating condition. Indicates the reference noise amplitude. This represents the noise amplitude adjustment coefficient.

[0031] Specifically, after determining the operating condition category to which the sample to be identified belongs, this step, based on the operating condition parameter set formed in step S2, employs an adaptive noise complete set empirical mode decomposition method to adaptively decompose the voltage imbalance time series into disturbance components. Furthermore, this step does not directly make a uniform judgment on all voltage imbalance samples, but rather distinguishes the imbalance disturbance characteristics at different time scales within the corresponding operating condition category. Let the voltage imbalance time series be... ,in The voltage imbalance at time t represents the time, and L represents the analysis window length.

[0032] First, obtain the decomposition level corresponding to the current operating condition category from the operating condition parameter set. The number of decomposition layers is not a fixed value, but is determined by the set of operating condition parameters formed in step S2, thus allowing the decomposition depth to adapt to the disturbance characteristics under different operating conditions. Subsequently, the CEEMDAN method and the adaptive noise method are used to further refine the decomposition. The decomposition process yields multiple intrinsic mode functions (IMFs) and residual components. The CEEMDAN method, an adaptive noise-complete set empirical mode decomposition method, is an adaptive time-frequency analysis method for nonlinear and non-stationary signals. It effectively eliminates mode aliasing by repeatedly adding white noise with adaptive amplitude to the original signal and performing empirical mode decomposition on the signal after each noise addition. Finally, all decomposition results are averaged, ensuring that the sum of all decomposed IMFs and residual components can accurately reconstruct the original signal. The obtained IMFs represent multiple types of perturbation components.

[0033] The decomposition process is as follows: First, let the original voltage imbalance time series be... Initialize residual components The decomposition order k=1. At the k-th decomposition, several sets of values ​​with zero mean and magnitude are generated. The white noise sequence is added to the current residual component. In the process, multiple noisy signals are obtained; empirical mode decomposition is performed once for each noisy signal to extract its first IMF number; then the arithmetic mean of the first IMFs obtained from all noisy signals is taken to obtain the k-th order IMF component. Then, subtract the IMF from the previous residual component to obtain the new residual component. Repeat the above steps until the residual amount is reached. If the function becomes monotonic or meets a preset stopping condition, such as having fewer than 2 extreme points, the decomposition stops, and the final result is obtained. One IMF component and one residual component ,satisfy ,in, Denotes the k-th eigenmode function. Represents the residual component. This represents the decomposition level under the q-th operating condition.

[0034] In the above decomposition process, the added noise amplitude is not a uniform value, but is adaptively adjusted according to the deviation of the current sample from the typical operating condition. Specifically, this step obtains the operating condition deviation from step S2. Then adjust the formula: Adjust the noise amplitude. As can be seen from this formula, when the current sample deviates significantly from typical operating conditions, the degree of deviation from operating conditions... Increase, noise amplitude The corresponding increase improves the ability of perturbation decomposition to distinguish abnormal fluctuations; conversely, when the sample is close to the typical operating condition, the noise amplitude remains at a low level to avoid over-decomposition.

[0035] In step S3, determining the corresponding disturbance type includes calculating the energy of each IMF component: And calculate the energy percentage of the k-th IMF component: ,in This represents the energy of the k-th IMF component. This represents the energy percentage of the k-th IMF component. Residual component energy is represented by T, where T is the total number of sampling points and K is the total number of intrinsic mode functions; wind speed fluctuation rate is extracted from the standardized operational dataset. Reactive power change rate and the rate of change of grid connection voltage Calculate the correlation between each IMF component and wind speed fluctuation rate, reactive power change rate, or grid connection point voltage change rate: ,in, The Pearson correlation calculation function is represented; based on the set of operating condition parameters, an operating condition-aware disturbance type mapping mechanism is established, and the center frequency of each IMF component is assigned. Energy percentage Duration and degree of correlation Input to mapping function In, among them, This indicates the type of disturbance corresponding to the k-th IMF component. This represents the disturbance type mapping function under the q-th type of operating condition; based on the disturbance component discrimination threshold in the operating condition parameter set, the disturbance types of each IMF component are divided into transient disturbance components, periodic fluctuation components, and persistent imbalance components.

[0036] Specifically, after completing the adaptive decomposition of the voltage imbalance time series, this step further determines the disturbance type corresponding to each intrinsic mode function. First, the center frequency of each extracted intrinsic mode function needs to be calculated. and duration For the center frequency For the k-th eigenmode function Perform a Fast Fourier Transform (FFT) to obtain its frequency domain amplitude spectrum, and take the frequency corresponding to the peak value of the largest amplitude in the spectrum or the centroid frequency of the power spectrum as the center frequency of that component. The calculation formula is: Where M is the total number of discrete frequency points. For the m-th frequency sequence value, This represents the corresponding power spectral density; for the duration Extracting this through Hilbert transform Instantaneous amplitude envelope of the component The calculation formula is: ,in, This represents the Hilbert transform operator. Subsequently, an effective magnitude threshold is set. ,in, This is a preset percentage coefficient. A time segment indicator function is constructed to calculate the effective duration. When the envelope amplitude season ;when season 0. Finally, the sum of the lengths of time segments with an indicator function of 1 in the statistical analysis window is taken as the effective duration of this IMF component. The calculation formula is: Where T is the total number of sampling points within the analysis window. The data sampling time interval, when the indicator function is 0, is the duration. Since the duration is invalid, the final valid duration is the duration itself.

[0037] Furthermore, the energy of each eigenmode function and the energy of the residual components are calculated. For the k-th eigenmode function... ,energy It equals the sum of the squares of the values ​​taken by the function at each sampling point within the analysis window, that is... , where T is the total number of sampling points. Residual components energy Calculate in the same way: Based on this, the energy proportion of the k-th eigenmode function is calculated. Where K is the total number of intrinsic mode functions. This represents the energy percentage of the k-th IMF component, reflecting the relative strength of this eigenmode function among all decomposed components. Wind speed volatility is extracted from the standardized operational dataset obtained in step S1. Reactive power change rate and the rate of change of grid connection voltage To measure the correlation between each intrinsic mode function and external disturbance factors, the correlation between the intrinsic mode function and wind speed fluctuation rate, reactive power change rate, and grid connection point voltage change rate were calculated, and the maximum value was taken as the degree of correlation. ,in, This represents the Pearson correlation calculation function. The degree of correlation is used to determine whether the intrinsic mode function is mainly caused by wind speed fluctuations, reactive power regulation, or changes in grid connection voltage.

[0038] Furthermore, the center frequencies of each IMF component are obtained. Energy percentage Duration and degree of correlation Subsequently, this step does not employ a fixed perturbation type classification rule. Instead, based on the operating condition parameter set formed in step S2, it establishes an operating condition-aware perturbation type mapping mechanism. For the q-th type of operating condition to which the current sample to be identified belongs, the perturbation type mapping function corresponding to the current operating condition category is obtained from the operating condition parameter set. And the center frequency range, energy percentage threshold, and duration threshold corresponding to the transient disturbance component, periodic fluctuation component, and persistent imbalance component, respectively. The k-th eigenmode function... , , , Input to mapping function Among them This indicates the type of disturbance corresponding to the k-th IMF component.

[0039] Under normal fluctuation conditions, if the effective duration of the k-th IMF component is... Less than the preset transient time lower limit threshold, and its energy percentage If the normal background noise energy threshold set for this operating condition is not exceeded, the component is determined to be a normal fluctuation component of the system. This type of component will be considered an invalid feature and will not be directly classified as a transient disturbance component to prevent minor natural fluctuations from causing false alarms in the system. Under power ramp-up conditions, the focus is on detecting the correlation between the active power change rate and the active power rate. If this IMF component correlates with the active power change rate... degree of correlation If the value exceeds a preset climbing correlation threshold, which is determined by the cumulative probability distribution of a large number of historical climbing event samples at a preset quantile, it distinguishes statistically strongly correlated events from general fluctuations, indicating a strong temporal synchronous change characteristic between the two. At this point, even if the center frequency of this component... If the frequency is lower than the upper limit of the fixed periodic fluctuation frequency, the component is also identified as a periodic fluctuation component caused by the output ramp-up through the mapping function. Under fluctuating wind speed conditions, the focus is on detecting the correlation between wind speed and wind speed. If the IMF component correlates with the wind speed volatility... degree of correlation Greater than the preset high relevance threshold, for example This indicates that there is a strong synchronous change characteristic between the two in the time domain. At this point, even the center frequency of this component... Transient disturbances below the conventional fixed high-frequency empirical lower limit are also identified as transient disturbance components caused by wind speed disturbances through the mapping function; Under reactive power limited operating conditions, if the center frequency of the IMF component... It falls within the typical periodic oscillation frequency range, such as the low-frequency oscillation band of 0.1Hz to 2Hz, and its correlation with the reactive power change rate is calculated. The correlation between the components is the largest among all external factors, indicating that the fluctuation is highly coupled with the operating state and response delay of reactive power compensation equipment such as SVG or SVC. Therefore, this component is preferentially identified as a periodic fluctuation component caused by reactive power regulation. In cases where the equipment exhibits a strong tendency to malfunction, if this component manifests as a low-frequency IMF component or a residual component, i.e. A value close to 0 corresponds to an effective duration. If the voltage significantly increases and exceeds the continuous degradation threshold set for the corresponding operating condition, and the negative sequence voltage component or voltage imbalance baseline at the grid connection point is monitored from the standardized operating data set and shows a continuous upward trend, then the component is determined to be a continuous imbalance component caused by defects in the equipment itself, such as single-phase grounding hazards or impedance imbalance. Under other transient conditions, such as grid fault conditions and system oscillation conditions, when it is a grid fault condition, if the effective duration of the IMF component is... Shorter than the preset upper limit of fault transient time and center frequency If a component is located in a high-frequency transient band and the correlation between the grid connection point voltage change rate and the negative-sequence voltage component both exceed the high correlation threshold, then this component is identified as a transient disturbance component caused by a grid-side fault, and the basic weight of evidence for negative-sequence voltage and grid connection point voltage change rate is prioritized in the cause determination. When the system is in an oscillating condition, if the center frequency of this IMF component... If a component falls into the low-frequency oscillation band and its correlation with the reactive power change rate or the grid connection point voltage change rate is significantly higher than its correlation with the wind speed fluctuation rate, and the duration of this component exceeds the preset oscillation duration threshold (which is usually determined based on the engineering definition of system oscillation and the statistical adjustment of actual waveform data), then this component is identified as a periodic fluctuation component caused by system oscillation.

[0040] After classifying the disturbance types of each intrinsic mode function, this step further constructs an energy vector to characterize the voltage imbalance disturbance: and duration vector: Furthermore, the energy proportions of eigenmode functions belonging to the same disturbance type can be added together to obtain the transient disturbance energy proportion. The proportion of periodic fluctuation energy and the proportion of persistent energy imbalance Simultaneously, the duration of each disturbance type within the analysis window is recorded. , and The original voltage imbalance time series was transformed into perturbation feature vectors at different time scales, providing a quantitative basis for subsequent determination of the dominant cause based on multi-source evidence fusion.

[0041] This step, through the aforementioned disturbance component analysis, transforms the voltage imbalance state from a single amplitude judgment into the identification of disturbance characteristics at different time scales. Furthermore, since both the disturbance decomposition parameters and the IMF disturbance type mapping rules are determined by the operating condition category, it avoids misjudgments caused by using a uniform disturbance classification rule under different operating conditions, providing a basis for subsequent fusion and discrimination of the dominant cause uncertainty.

[0042] Step S4: Based on the set of operating parameters, energy ratio and disturbance type of the sample to be identified, construct a set of evidence for cause discrimination, calculate the fusion confidence of each cause type based on the fusion method of DS evidence theory, and determine the dominant cause type that leads to voltage imbalance in the sample to be identified.

[0043] In step S4, constructing the causal discrimination evidence set based on the operating parameter set, energy ratio, and disturbance type of the sample to be identified includes: constructing a causal discrimination vector, including the operating parameter set, energy ratio, and disturbance type of the sample to be identified; and using the isolated forest algorithm to calculate the causal discrimination vector to obtain the anomaly score of the sample to be identified. ,in, Indicates the sample to be identified Abnormal scores, )) represents the sample to be identified Average path length in an isolated tree () is the sample to be identified Path length in a single isolated tree The normalization factor represents the number of samples to be identified as n; a set of evidence for causal discrimination is constructed based on the causal discrimination vector and the anomaly score.

[0044] Specifically, this step, based on the set of operating condition parameters of the sample to be identified, the energy proportion obtained from disturbance component analysis, and the disturbance type, forms multiple evidence sources for subsequent DS evidence theory fusion. First, a causal discrimination vector is constructed, which integrates multi-dimensional information reflecting the degree of deviation from operating conditions, disturbance characteristics, electrical severity, and external disturbance factors. The degree of deviation from operating conditions is extracted from the set of operating condition parameters. Obtain the transient disturbance energy percentage from the energy percentage. The proportion of periodic fluctuation energy and the proportion of persistent energy imbalance Extract the duration of transient disturbances from the duration corresponding to each disturbance type. Duration of periodic fluctuations Duration of persistent imbalance To construct a complete set of evidence for causal identification, the above parameters need to be combined with characteristic parameters representing electrical quantity anomalies and external disturbances. Therefore, the negative sequence voltage component is obtained from the voltage imbalance feature vector extracted in step S1. reactive power regulation margin and wind speed fluctuation rate The above parameters are combined to form a causal discriminant vector. , is used to characterize the state of the current sample in the multidimensional causal space.

[0045] To identify anomalous samples that significantly deviate from normal operating conditions, the Isolation Forest algorithm is further employed to calculate the causal discrimination vector, yielding anomaly scores for the samples to be identified. The Isolation Forest algorithm constructs multiple isolated trees, each randomly selecting features and split points to recursively partition the sample space. For the samples to be identified… Calculate its path length in each isolated tree. The average path length is calculated by taking the average of the path lengths of all isolated tree paths, which is the number of edges traversed from the root node to the leaf node containing the sample. Anomaly scoring Calculate using the following formula: ,in This represents the normalization factor when the number of samples to be identified is n, and it is used to normalize the path length to between 0 and 1. From this formula, it can be seen that the easier it is for samples to be isolated after a small number of segments (the shorter the average path length), the better. The smaller the value, the closer the index term is to 1, indicating anomaly scoring. The closer the score is to 1, the more likely the sample is to be an anomalous sample; conversely, normal samples usually require more segmentation to be isolated, have a longer average path length, and an anomalous score tends to be close to 0.5 or lower.

[0046] Obtaining the causation discriminant vector and abnormal scoring Subsequently, based on these two types of information, a causal discriminant evidence set was constructed for the fusion of DS evidence theory. Causal discriminant evidence set Among them, evidence of deviation from operating conditions Deviation from operating conditions Direct determination reflects the degree to which the current sample deviates from its typical operating condition; evidence of perturbation components. The energy proportions of transient disturbances, periodic fluctuations, and persistent imbalances, along with their corresponding durations, are jointly determined, reflecting the contribution of disturbances at different time scales to voltage imbalance; evidence of negative-sequence voltage. From negative sequence voltage components This confirms the severity of the voltage imbalance; evidence of reactive power margin. Reactive power adjustment margin This confirms the reactive power support capability of the wind turbine or dynamic reactive power compensation device; evidence of wind speed fluctuations. Wind speed fluctuation rate This confirms the impact of meteorological disturbances on voltage imbalance; anomaly scoring evidence. Anomaly scoring from isolated forests This evidence is used to assist in identifying abnormal states that significantly deviate from normal operating conditions. When the abnormality score is high, this evidence strengthens the support for equipment abnormalities or compound abnormality types. These six pieces of evidence together constitute a complete set of causal discrimination evidence, providing input for subsequent multi-source evidence fusion and dominant causal discrimination based on DS evidence theory.

[0047] In step S4, the fusion confidence of each cause type is calculated using the fusion method based on DS evidence theory. This includes: constructing a cause identification framework consisting of wind speed disturbance type, reactive power limitation type, equipment anomaly type, and composite anomaly type; and, based on the operating condition category, using a basic probability allocation function, calculating the fusion confidence of each type of evidence in the cause discrimination evidence set. Assign basic probabilities to different cause types in the cause identification framework: ,in, This represents the basic probability assignment of the j-th type of evidence to the l-th type of cause. Let represent the basic probability allocation function for the j-th type of evidence under the q-th type of operating condition. This represents the l-th type of cause; the evidence set for cause determination includes multiple sources of evidence. Calculate the conflict coefficient between any two sources of evidence in the evidence set for cause determination: ,in Let evidence a be the basic probability assigned to cause type B. Assigning a basic probability to cause type C for evidence b; performing probability fusion of the two evidence sources based on the combination rules of DS evidence theory: , Let A be the fusion confidence score of cause type A, where A is the causal type at the intersection of cause types B and C; the fusion confidence score corresponding to each cause type is obtained by successive fusion.

[0048] Specifically, a cause identification framework consisting of wind speed disturbance type, reactive power limitation type, equipment anomaly type, and composite anomaly type is first constructed. ,in This indicates a wind speed disturbance type. Indicates reactive power limited type. Indicates a device malfunction. This indicates a composite anomaly. This framework covers the main causes of voltage imbalance in wind farms, providing a unified proposition space for subsequent evidence allocation and fusion. Then, based on the operating condition category of the sample to be identified, each type of evidence in the causal discrimination evidence set is processed using a basic probability allocation function. For the j-th type of evidence in the causal discrimination evidence set... Obtain the preset basic probability allocation function corresponding to the current operating condition category from the set of operating condition parameters formed in step S2. , where q represents the qth type of operating condition.

[0049] Basic probability assignment function Using a pre-defined piecewise linear trapezoidal membership function, based on evidence... The specific value of the evidence is relative to the causal discrimination threshold under the corresponding working condition. The membership degree is calculated and normalized to a basic probability. First, a preset dynamic reliability coefficient is introduced to characterize the inherent confidence level of the j-th type of evidence under the current working condition. Then, the initial membership degree of the evidence to a specific causal proposition is divided by the sum of its initial membership degrees to all four causal propositions and normalized. The quotient is then multiplied by the dynamic reliability coefficient to obtain the effective basic probability allocation value of the evidence to the specific causal proposition. At the same time, to reasonably quantify the ambiguity and detection error of a single feature evidence, the sum of the basic probability allocation values ​​of the evidence to the above four specific causal propositions is subtracted from the number 1 to calculate its direction to the entire causal identification framework. The global uncertainty probability assignment value maps various underlying features with different dimensions into a standard evidence set containing both specific causal probabilities and unknown uncertain probabilities, laying a reliable data foundation for subsequent processing of conflicting evidence and implementation of multi-source fusion. Basic probabilities are assigned to different causal types within the causal identification framework, i.e. ,in This indicates that the j-th type of evidence relates to the l-th type of cause. The basic probability allocation, where l takes values ​​from 1 to 4, corresponds to wind speed disturbance type, reactive power limitation type, equipment anomaly type, and combined anomaly type, respectively. The basic probability allocation for each type of evidence satisfies the normalization condition, i.e. .

[0050] After obtaining the basic probability assignments of multiple evidence sources, the combination rules of the DS evidence theory are used for successive fusion. First, for any two evidence sources a and b in the causal discrimination evidence set, the conflict coefficient between them is calculated. The conflict coefficient is defined as the sum of the products of the basic probabilities of two sources of evidence assigned to disjoint causal types, i.e. ,in Let evidence a be the basic probability assigned to cause type B. Let B be the base probability assigned to causal type C, where B and C are both causal types in the identification frame. The conflict coefficient reflects the degree of inconsistency between the two pieces of evidence.

[0051] Then, based on the DS combination rule, the basic probability assigned to a certain causal type A after the fusion of the two evidence sources is calculated. For causal type A, the fusion confidence level is... Equal to all satisfying The sum of the basic probability products of the causal type combinations, divided by 1 and minus the conflict coefficient, is... . When the denominator is 1- When the coefficient is zero, it indicates that the evidence is completely conflicting and cannot be integrated. In this case, a correction strategy or separate processing is required. Specifically, when a conflict coefficient is detected... When the normalization denominator of the classic DS combination rule becomes invalid due to the value approaching or equaling 1, the system automatically triggers a conflict evidence downgrading and redistribution mechanism. First, an evidence distance algorithm is introduced to calculate the relative similarity between the evidence sources currently participating in the fusion, in order to identify anomalous evidence sources that significantly contradict other features. Based on this similarity, a penalty weight is dynamically generated to attenuate and discount the basic probability allocation value of anomalous evidence. Then, the division by 1 in the original formula is abandoned. Instead of the forced normalization step, a conflict inclusion rule is adopted, which determines the probability mass of failure to reach consensus due to complete conflict, i.e., the original conflict coefficient. The equivalent part is directly assigned to the global recognition framework, thereby completely avoiding the risk of collapse during division by zero operation, while truly preserving the serious discrepancies between the underlying data, providing a reasonable mathematical explanation and uncertainty representation for the subsequent introduction of historical time series fusion or triggering of manual review and early warning.

[0052] Using this combination rule, the basic probability allocation after fusion still satisfies the normalization condition. For cases with multiple sources of evidence, a successive fusion approach is adopted: the first piece of evidence is fused with the second piece of evidence according to the above rule, and this fusion result is then fused with the third piece of evidence, and so on, until all evidence participates in the fusion. Finally, the fusion confidence level corresponding to each dominant causal type is obtained. ,in, This represents the fusion confidence level of the dominant cause type l, i.e., the basic probability assigned to each cause type after fusion. For example, when the evidence of deviation from operating conditions indicates that the sample deviates significantly from its operating condition center, the transient disturbance energy accounts for a high proportion in the disturbance component evidence and is strongly correlated with wind speed fluctuations, and the wind speed fluctuation evidence also supports the existence of large wind speed disturbances, the basic probability assigned to the wind speed disturbance type after fusion will be significantly higher than other types, thus forming a higher fusion confidence level. Conversely, if multiple pieces of evidence point to different causes and their confidence levels are similar, the fusion confidence level of the composite anomaly type will increase accordingly. Through the above fusion process, this step can comprehensively utilize the uncertainty information of multi-source evidence to obtain the fusion confidence level of each cause type under the constraints of operating conditions, providing a quantitative basis for determining the dominant cause of three-phase voltage imbalance.

[0053] In step S4, the dominant cause type for the voltage imbalance of the sample to be identified includes: when the fusion confidence of a cause type among wind speed disturbance, reactive power limitation, or equipment malfunction exceeds a set confidence threshold, the corresponding cause type is determined as the dominant cause type; when the fusion confidence of two or more cause types exceeds the set confidence threshold, or the difference between the highest fusion confidence and the second highest fusion confidence is less than a set difference threshold, the current voltage imbalance is determined as a composite malfunction.

[0054] Specifically, after the fusion of multi-source evidence based on the DS evidence theory, the fusion confidence levels of four types of causes—wind speed disturbance, reactive power limitation, equipment anomaly, and combined anomaly—were obtained. Based on this, the dominant cause type leading to voltage imbalance in the sample to be identified was determined according to the relationship between the fusion confidence level and a preset threshold. First, a confidence threshold was set. and difference threshold The confidence threshold is used to determine whether a single causal type has a sufficiently high fusion confidence, and the difference threshold is used to determine whether there is a significant difference between the highest and second-highest fusion confidence. The formula is: And satisfy Among them, B ∗ This indicates the finally determined dominant cause type. When the fusion confidence level of any cause type among wind speed disturbance, reactive power limitation, or equipment malfunction exceeds a set confidence threshold... When this type of cause is identified, it is determined as the dominant cause type. For example, when the fusion confidence level of the wind speed disturbance type is... If the fusion confidence level is significantly higher than other types, then the dominant cause of the current three-phase voltage imbalance is determined to be the wind speed disturbance type.

[0055] When the fusion confidence of two or more cause types exceeds a set confidence threshold, or the difference between the highest and second-highest fusion confidence is less than a set difference threshold, it indicates that multiple causes are simultaneously significant or difficult to distinguish as primary or secondary. In this case, the current voltage imbalance is classified as a composite anomaly, as shown in the formula: ,in, This indicates the highest fusion confidence level. This indicates the second-highest fusion confidence level. For example, if the fusion confidence level for wind speed disturbance is 0.6 and the fusion confidence level for reactive power limitation is 0.55, both exceeding the confidence threshold of 0.5, and the difference of 0.05 is less than the difference threshold of 0.1, then it is judged as a composite anomaly. Similarly, if the fusion confidence level for equipment anomaly is 0.65 and the confidence level for wind speed disturbance is 0.60, although only the equipment anomaly exceeds the threshold, their confidence levels are close and the difference is less than the difference threshold of 0.1, it should also be judged as a composite anomaly.

[0056] This step, under operating condition constraints, comprehensively utilizes multi-source evidence such as operating condition deviations, disturbance components, negative sequence voltage, reactive power margin, wind speed fluctuations, and anomaly scores to form fused confidence scores for different cause types, and based on this, determines the dominant cause of voltage imbalance. This method can handle situations with contradictory multiple indicators, ambiguous boundary samples, and overlapping multiple causes, making the voltage imbalance cause identification results more reliable and directly providing a basis for matching subsequent remediation targets.

[0057] Step S5: Match the governance objects according to the dominant cause type, fusion confidence level and operating condition category, determine the evaluation weight of the governance strategy, generate the corresponding voltage imbalance governance strategy, and update the voltage imbalance governance strategy according to the governance improvement rate, voltage recovery time and governance action cost.

[0058] The step S5, which generates the corresponding voltage imbalance governance strategy, includes: constructing a comprehensive evaluation function for governance strategies that includes voltage imbalance suppression effect indicators, governance action cost indicators, and operational risk constraint indicators; calculating the comprehensive score value of each preset candidate governance strategy through the comprehensive evaluation function; adaptively assigning differentiated evaluation weights to the candidate governance strategies according to the operating condition category and the dominant cause type; using the fusion confidence level corresponding to the dominant cause type to correct the comprehensive score value of the comprehensive evaluation function; and selecting the candidate governance strategy whose corrected comprehensive evaluation value is higher than the preset threshold as the final governance strategy.

[0059] Specifically, based on the dominant cause type and its fusion confidence level determined in step S4, and the operating condition category obtained in step S3, a corresponding voltage imbalance mitigation strategy is further generated. For example... Figure 4 As shown in the figure, this diagram schematically illustrates the complete process of identifying the dominant cause of three-phase voltage imbalance and matching the target of treatment based on multi-source evidence fusion. Figure 4 The left side is the multi-source evidence fusion and discrimination stage, which includes operating condition deviation evidence, disturbance component evidence, negative sequence voltage evidence, reactive power margin evidence, and wind speed fluctuation and anomaly score evidence. After basic probability allocation, these evidences are fused through DS evidence to obtain the fusion confidence level Bel for each cause type, and based on this, the dominant cause type is determined to be wind speed disturbance type, reactive power limitation type, equipment anomaly type, or composite anomaly type. Figure 4 The right side represents the matching of governance targets. When the dominant cause is wind speed disturbance, priority is given to matching wind turbine parameter adjustment strategies. Specifically, this may include adjusting wind turbine reactive power control parameters, adjusting negative sequence voltage feedforward control parameters, and optimizing short-time control response parameters. The negative sequence current reference value can be expressed as: ,in, This represents the reference value for the negative sequence current. and These represent the proportional coefficient and the integral coefficient, respectively. This represents the negative sequence voltage component. Adjusting this reference value can suppress transient imbalances caused by sudden changes in wind speed and rapid changes in unit output. When the dominant cause is reactive power limitation, priority is given to matching SVG / SVC switching adjustments and unit operation mode optimization strategies, including SVG or SVC reactive power output command adjustments, reactive power compensation device switching sequence adjustments, and reactive power capacity allocation based on node voltage sensitivity.

[0060] To determine the switching priority of dynamic reactive power compensation devices, a node voltage sensitivity model is constructed: ,in The sensitivity of node voltage to reactive power changes is represented by U, where U represents node voltage and Q represents reactive power. Based on the node voltage sensitivity, reactive power regulation margin, and equipment operation constraints, the switching sequence of reactive power compensation devices is determined to improve the grid connection point's voltage support capability and imbalance suppression capability. When the dominant cause type is equipment anomaly, a priority equipment investigation strategy is matched. Specifically, this may include the generation of investigation priorities for box-type transformers, collector lines, step-up substation equipment, and measurement links. The investigation priority can be determined based on the duration of negative sequence voltage, the proportion of persistent imbalance components, the degree of deviation in equipment operating status, and historical anomaly records. When the dominant cause type is complex anomaly, a joint governance strategy for multiple governance objects is generated. This strategy may simultaneously include turbine parameter adjustment, SVG / SVC switching adjustment, unit operation mode optimization, and equipment investigation to address complex imbalance states caused by the combined effects of multiple factors.

[0061] Furthermore, the evaluation weights for the governance strategies in this invention are not fixed, but are jointly determined by the operating condition category obtained in step S2 and the dominant cause type and corresponding fusion confidence level obtained in step S4. For wind speed disturbance-type imbalances under wind speed fluctuation conditions, the evaluation weights focus more on voltage recovery time and control response speed; for reactive power-limited imbalances under reactive power-limited conditions, the evaluation weights focus more on the reactive power adjustment margin recovery effect and the operating cost of dynamic reactive power compensation devices; for equipment anomaly-type imbalances, the evaluation weights focus more on the long-term suppression effect of negative sequence voltage and the priority of equipment troubleshooting; for compound anomaly-type imbalances, the evaluation weights simultaneously consider the governance improvement rate, voltage recovery time, governance operation cost, and operational risk constraints. First, for the current operating condition category and dominant cause type, the corresponding governance strategy evaluation weight vector is obtained from the operating condition parameter set formed in step S2. ,in, This represents the strategy evaluation weight vector under the q-th operating condition and the dominant cause type B. This indicates the weight of the voltage imbalance suppression effect. Indicates the cost weight of governance actions. This indicates the operational risk weight. For example, under wind speed fluctuation conditions and when the dominant cause is wind speed disturbance, The medium-weighted voltage recovery time index is emphasized. The focus is on controlling response speed and cost; under reactive power-constrained operating conditions, and when the dominant cause is reactive power constraint, The focus is on the operating cost of dynamic reactive power compensation devices; in the event of equipment malfunction, The middle focuses on the effect of sustained imbalance inhibition. In the case of complex anomalies, the priority of equipment inspection is emphasized; in the case of complex anomalies, the weight of all three factors is considered simultaneously.

[0062] Then, a comprehensive evaluation function for governance strategies is constructed. For each pre-defined candidate governance strategy, the corresponding voltage imbalance suppression effect index is calculated. Cost indicators of governance actions and operational risk indicators Among them, the voltage imbalance suppression effect index The cost index of the treatment action is determined by the predicted reduction in voltage imbalance after the treatment. The operational risk indicators are determined by the number of control actions, the number of dynamic reactive power compensation device actions, and equipment troubleshooting costs. The risk level is determined by the current risk level, the duration of the negative sequence voltage, and equipment operating constraints. In practice, a preset extreme value normalization function is used to map the predicted decrease, the absolute value of the action cost, and the risk level to a unified dimensionless interval [0, 1], which serves as the basis for the determination. , and The final value of .

[0063] Subsequently, the three indicators mentioned above are linearly weighted with their corresponding evaluation weights to obtain the comprehensive evaluation value of the governance strategy. To demonstrate the impact of causal identification confidence on governance strategy selection, the fusion confidence levels corresponding to the dominant causal types obtained in step S4 are used. The comprehensive evaluation value is adjusted, and a confidence-corrected governance strategy evaluation function is constructed. When the confidence level of the dominant cause is high, the governance strategy matching that dominant cause is prioritized; when the confidence level of the dominant cause is low or the confidence levels of multiple causes are close, a joint governance strategy for multiple governance objects or a conservative governance strategy is prioritized. This is achieved by comparing the corresponding governance strategies of different candidates. The candidate governance strategy that meets the operational safety constraints and whose corrected comprehensive evaluation value is higher than the preset threshold is selected as the final governance strategy.

[0064] To further standardize the mapping logic between the dominant causes and the objects of governance, this embodiment establishes a mapping relationship: ,in, This indicates the generated governance strategy. Indicates the dominant cause type, This indicates the fusion confidence level corresponding to the dominant causal type. Represents the set of available governance objects. Indicates the weight of governance strategy evaluation. This represents the matching function from the cause type to the governance object. The specific mapping relationship of this matching function is as follows: That is, when the dominant cause is wind speed disturbance, Corresponding wind turbine parameter adjustment strategies; when the dominant cause is reactive power limitation. This corresponds to SVG / SVC switching adjustments and unit operation mode optimization strategies; when the dominant cause is equipment anomaly. This corresponds to the equipment troubleshooting strategy for box-type transformers, collector lines, step-up substation equipment, or measurement links; when the dominant cause type is a composite anomaly. This corresponds to a joint governance strategy. Based on the above mapping, a quantitative comparison of each candidate strategy is performed using a comprehensive evaluation function with corrected confidence levels. Finally, the governance strategy that meets operational safety constraints and has the highest corrected comprehensive evaluation value is selected as the final implementation plan, thereby achieving a closed-loop process from cause identification to governance object matching and strategy selection.

[0065] The step S5 of updating the voltage imbalance governance strategy includes: calculating the governance improvement rate based on the decrease in voltage imbalance before and after governance; determining the voltage recovery time based on the time it takes for the voltage imbalance to recover to within the target threshold; constructing a strategy effectiveness score in combination with the governance action cost; comparing the current governance scenario with historical governance samples; and dynamically updating the corresponding governance strategy based on the strategy effectiveness score if the two have the same operating condition category and dominant cause type.

[0066] Specifically, the effectiveness of the implemented governance strategies is evaluated, and the priority of governance strategies under the same operating condition category and the same dominant cause type is updated based on the evaluation results, thereby achieving closed-loop optimization of the governance strategies. First, the governance improvement rate is calculated based on the change in voltage imbalance before and after governance. Let the voltage imbalance before the treatment be . After treatment, the voltage imbalance is: The formula for calculating the governance improvement rate η is: This indicator reflects the degree to which the control strategy suppresses voltage imbalance; a larger η value indicates a better control effect. Secondly, the voltage recovery time is calculated. Let the time of implementation of the governance strategy be t. The moment when the voltage imbalance first recovers to within the preset target threshold is The voltage recovery time This reflects the responsiveness and timeliness of the governance strategy. A smaller value indicates faster governance. Furthermore, this is considered in conjunction with the cost of governance actions. Construct a strategy effectiveness score The costs of governance actions include the number of control actions, the number of times the dynamic reactive power compensation device operates, and equipment troubleshooting costs. Strategy effectiveness scoring. Calculate using the following formula: . , , These are the weighting coefficients corresponding to the governance improvement rate, voltage recovery time, and governance action cost, respectively. As can be seen from this formula, the higher the governance improvement rate, the shorter the voltage recovery time, and the lower the governance action cost, the higher the strategy effectiveness score.

[0067] After completing the above evaluation, the current governance scenario is compared with historical governance samples. If the current sample and a certain historical governance sample have the same operating condition category and the same dominant cause type, they are judged as similar governance scenarios. For similar governance scenarios, if the strategy effectiveness score... If the score exceeds a preset threshold, the priority of invoking this governance strategy will be increased; if the strategy effectiveness score is... If the threshold is lowered, the invocation priority of the governance strategy is reduced, and a new candidate governance strategy is matched. For similar governance scenarios, the effectiveness score of the current governance strategy is used. The invocation priority of this governance strategy is dynamically updated under the specific operating condition category and dominant cause type. Let the strategy priority before the update be... The threshold for strategy effectiveness scoring is The priority update coefficient is The updated policy priority Calculate using the following formula: .when Higher than hour, Greater than This indicates that the governance strategy should be invoked first in similar scenarios; when Below hour, Less than This indicates that the priority of this governance strategy should be reduced, and other candidate governance strategies can be re-matched.

[0068] Furthermore, the strategy effectiveness score is used not only to update the priority of governance strategy invocation, but also to adjust the evaluation weight of governance strategies under the same operating condition category and the same dominant cause type. Let the evaluation weight vector before the update be... The weight update coefficient is The weight adjustment direction vector is Then the updated evaluation weight : For example, when the governance improvement rate is high but the recovery time is long, in Increase the weighting of the voltage recovery time component; when the improvement rate is high and the action cost is low, Increase the priority of invoking governance strategies; when governance actions are costly but have limited improvement effects, The constraints on governance costs in subsequent strategy evaluations are increased. When the confidence levels of multiple causes in a composite anomaly sample are close, joint governance strategies involving multiple governance objects are prioritized, and the contribution weights of each governance object are adjusted based on the governance results. For example, under wind speed fluctuation conditions, if a negative-sequence voltage feedforward control parameter adjustment strategy can significantly reduce the voltage imbalance caused by wind speed disturbances and has a short recovery time, this strategy will be prioritized under subsequent conditions with the same cause type. Similarly, under reactive power constraint conditions, if a certain SVG switching sequence adjustment strategy has limited effectiveness in improving reactive power constraint imbalances, its priority will be reduced, and other reactive power allocation or unit control coordination strategies will be prioritized.

[0069] A comparative diagram of the effects of different governance strategies on wind farm voltage imbalance is shown below. Figure 5 As shown in the figure, the horizontal axis represents time, the vertical axis represents voltage imbalance, and the three curves represent the changes in voltage imbalance over time under the original operating state, the single-treatment state, and the optimized treatment state, respectively. Figure 5 It can be seen that the voltage imbalance decreases faster, recovers more quickly, and reaches a lower steady-state value under optimized governance, demonstrating the superiority of the method of this invention compared to a single governance strategy. Through the above-mentioned governance effect evaluation and strategy update, this embodiment transforms the execution effect of a single governance process into strategy invocation experience under similar operating conditions. The original operating state curve shows that the voltage imbalance remained at a high level before governance; the single governance state curve shows that the voltage imbalance decreased after adopting the conventional governance strategy, but the recovery time was long and the steady-state value was still high; the optimized governance state curve shows that after adopting the governance strategy updated by this invention, the voltage imbalance decreased rapidly, the recovery time was significantly shortened, and the steady-state value decreased to within the target threshold. This indicates that through the governance effect evaluation and strategy priority update in step S5, this embodiment can make the selection of governance strategies under subsequent similar operating conditions more accurate, achieving closed-loop optimization of the voltage imbalance governance strategy.

[0070] This step, through the aforementioned evaluation of governance effectiveness and strategy update, transforms the execution effect of a single governance process into experience for strategy invocation under similar operating conditions, achieving closed-loop optimization of the three-phase voltage imbalance governance strategy. Compared to methods that simply adjust based on the governance improvement rate, this invention simultaneously considers operating condition category, dominant cause type, cause confidence level, governance improvement rate, voltage recovery time, and governance action cost, making the strategy update results more consistent with the actual operating characteristics of wind farms.

[0071] In summary, the wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching proposed in this application constructs a voltage imbalance feature vector and time series through multi-source data preprocessing and feature extraction. Then, DBSCAN clustering is used to identify operating condition categories and generate operating condition parameter sets, enabling adaptive linkage of subsequent decomposition, discrimination, and mitigation processes. In disturbance analysis, the CEEMDAN method is used to adaptively decompose the voltage imbalance time series, combining the operating condition parameter set to achieve a fine classification of transient, periodic, and persistent disturbances. Based on DS evidence theory, multi-source evidence such as operating condition deviation, disturbance components, negative sequence voltage, reactive power margin, wind speed fluctuations, and anomaly scores are integrated to quantify the fusion confidence of each cause type, accurately identifying dominant causes such as wind speed disturbances, reactive power limitation, equipment anomalies, or combined anomalies. Mitigation targets are matched according to the dominant causes and operating condition categories, and a comprehensive evaluation function with confidence correction is used to optimize the strategy. Simultaneously, the strategy priority and evaluation weight are updated through feedback of mitigation improvement rate, recovery time, and action cost, forming a closed-loop optimization. This method effectively solves the problems of poor adaptability to operating conditions, ambiguity of multiple causes, and rigid strategies in traditional governance, and significantly improves the pertinence, adaptability, and continuous improvement capability of voltage imbalance governance.

[0072] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0073] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. 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. 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 this application.

Claims

1. A method for mitigating voltage imbalance in wind farms based on operating condition constraints and cause matching, characterized in that, The method includes: Step S1: Obtain multi-source operation data of the wind farm at different sampling periods, obtain multiple samples to be identified, and preprocess them to obtain a standardized operation dataset. Based on the standardized operation dataset, construct a voltage imbalance feature vector and obtain a voltage imbalance time series. Step S2: Obtain the wind farm operating condition constraint features of each sample to be identified based on the voltage imbalance feature vector, determine the operating condition category and operating condition deviation of the sample to be identified, and establish the corresponding operating condition parameter set; Step S3: Based on the set of operating parameters, the CEEMDAN method is used to adaptively decompose the voltage imbalance time series to obtain multiple disturbance components, and the energy proportion of each disturbance component is calculated to determine the corresponding disturbance type. Step S4: Based on the set of operating parameters, energy ratio and disturbance type of the sample to be identified, construct a set of evidence for cause discrimination, calculate the fusion confidence of each cause type based on the fusion method of DS evidence theory, and determine the dominant cause type that leads to voltage imbalance in the sample to be identified. Step S5: Match the governance objects according to the dominant cause type, fusion confidence level and operating condition category, determine the evaluation weight of the governance strategy, generate the corresponding voltage imbalance governance strategy, and update the voltage imbalance governance strategy according to the governance improvement rate, voltage recovery time and governance action cost.

2. The wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching according to claim 1, characterized in that, Step S1 to obtain the voltage imbalance time series includes: Based on the standardized operation dataset, voltage imbalance is extracted, including the ratio of negative sequence voltage to positive sequence voltage. Combined with phase-to-phase voltage deviation, phase sequence offset, negative sequence current ratio, active power change rate, reactive power change rate, reactive power adjustment margin, grid connection point voltage change rate, and wind speed fluctuation rate, a voltage imbalance feature vector is constructed. Based on the voltage imbalance degree in the voltage imbalance feature vector, the voltage imbalance degree time series is obtained according to the sampling time point order.

3. The wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching according to claim 1, characterized in that, Step S2 includes: The active power change rate, reactive power change rate, wind speed fluctuation rate, grid connection point voltage change rate, and reactive power regulation margin are extracted from the voltage imbalance feature vector as operating condition constraint features. For each sample to be identified, the DBSCAN density clustering method is used to cluster the operating condition constraint features to determine the operating condition category, and the distance relative to the center point of the operating condition category is calculated as the operating condition deviation. Based on the operating condition category and the operating condition deviation, the corresponding operating condition parameter set is generated.

4. The wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching according to claim 1, characterized in that, Step S3 involves using the CEEMDAN method to perform adaptive disturbance component decomposition on the voltage imbalance time series, including: The voltage imbalance time series was decomposed using the CEEMDAN time series decomposition method, yielding multiple intrinsic mode functions and residual components: ,in, This is a time series of voltage imbalance. Denotes the k-th eigenmode function. Represents the residual component. This represents the decomposition level under the qth type of operating condition, and the decomposition level is determined by the set of operating condition parameters. The amplitude of noise added during the decomposition process According to the deviation of the operating conditions The noise amplitude is adaptively adjusted using the following formula: ,in, This represents the adjusted noise amplitude under the q-th operating condition. Indicates the reference noise amplitude. This represents the noise amplitude adjustment coefficient.

5. The wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching according to claim 4, characterized in that, Step S3 involves determining the corresponding disturbance type, including: Calculate the energy of each IMF component: And calculate the energy percentage of the k-th IMF component: ,in This represents the energy of the k-th IMF component. This represents the energy percentage of the k-th IMF component. The residual component energy is represented by T, where T is the total number of sampling points and K is the total number of intrinsic mode functions. Extracting wind speed fluctuations from the standardized operational dataset Reactive power change rate and the rate of change of grid connection voltage Calculate the correlation between each IMF component and wind speed fluctuation rate, reactive power change rate, or grid connection point voltage change rate. Where corr(·) represents the correlation calculation function; Based on the operating condition parameter set, an operating condition-aware disturbance type mapping mechanism is established, which maps the center frequencies of each IMF component. Energy percentage Duration and degree of correlation Input to mapping function In, among them, This indicates the type of disturbance corresponding to the k-th IMF component. This represents the disturbance type mapping function under the q-th type of operating condition; Based on the threshold values ​​of the disturbance components in the aforementioned operating condition parameter set, the disturbance type of each IMF component is determined. It is divided into transient disturbance components, periodic fluctuation components, and persistent imbalance components.

6. The wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching according to claim 1, characterized in that, In step S4, based on the set of operating parameters, energy percentage, and disturbance type of the sample to be identified, the causal discrimination evidence set is constructed, including: A causal discrimination vector is constructed, including the set of operating parameters, energy proportion, and disturbance type of the sample to be identified. The isolated forest algorithm is used to calculate the causal discrimination vector to obtain the anomaly score of the sample to be identified. ,in, Indicates the sample to be identified Abnormal scores, )) represents the sample to be identified Average path length in an isolated tree () is the sample to be identified Path length in a single isolated tree The normalization factor represents the number of samples to be identified as n. A set of evidence for cause discrimination is constructed based on the cause discrimination vector and the anomaly score.

7. The wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching according to claim 1, characterized in that, Step S4, based on the DS evidence theory, calculates the fusion confidence of each causal type using the fusion method, including: A cause identification framework consisting of wind speed disturbance type, reactive power limitation type, equipment anomaly type, and composite anomaly type is constructed; Based on the operational condition category, and through a basic probability allocation function, based on each type of evidence in the causal discrimination evidence set... Assign basic probabilities to different cause types in the cause identification framework: ,in, This represents the basic probability assignment of the j-th type of evidence to the l-th type of cause. Let represent the basic probability allocation function for the j-th type of evidence under the q-th type of operating condition. This is the first type of cause; The causal identification evidence set includes multiple evidence sources. The conflict coefficient between any two evidence sources in the causal identification evidence set is calculated. ,in Let evidence a be the basic probability assigned to cause type B. Assign the basic probability of evidence b to causal type C; Probabilistic fusion of two evidence sources is performed based on the combination rule of DS evidence theory: , Let A be the fusion confidence of causal type A, where A is the causal type at the intersection of causal types B and C; The fusion confidence level corresponding to each cause type is obtained by using a successive fusion method.

8. The wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching according to claim 7, characterized in that, In step S4, the dominant cause type for the voltage imbalance of the sample to be identified includes: When the fusion confidence of a certain cause type among wind speed disturbance type, reactive power limitation type, or equipment abnormality type exceeds the set confidence threshold, the corresponding cause type will be determined as the dominant cause type. When the fusion confidence of two or more cause types exceeds the set confidence threshold, or the difference between the highest fusion confidence and the second highest fusion confidence is less than the set difference threshold, the current voltage imbalance is determined to be a composite anomaly.

9. The wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching according to claim 8, characterized in that, Step S5 generates the corresponding voltage imbalance mitigation strategy, including: A comprehensive evaluation function for governance strategies is constructed, which includes indicators of voltage imbalance suppression effectiveness, governance action cost, and operational risk constraints. The comprehensive evaluation function is used to calculate the comprehensive score of each preset candidate governance strategy. Based on the operating condition category and the dominant cause type, differentiated evaluation weights are adaptively assigned to candidate governance strategies. The comprehensive score value of the comprehensive evaluation function is corrected using the fusion confidence value corresponding to the dominant cause type. Candidate governance strategies with a corrected comprehensive evaluation value higher than a preset threshold are selected as the final governance strategies.

10. The wind farm voltage imbalance mitigation method based on operating condition constraints and cause matching according to claim 8, characterized in that, Step S5 updates the voltage imbalance mitigation strategy, including: The improvement rate is calculated based on the decrease in voltage imbalance before and after the treatment. The voltage recovery time is determined based on the time it takes for the voltage imbalance to recover to within the target threshold. Combined with the cost of the treatment action, a strategy effectiveness score is constructed. The current treatment scenario is compared with historical treatment samples. If the operating condition category and the dominant cause type are the same for both, the corresponding treatment strategy is dynamically updated according to the strategy effectiveness score.

Images