Power grid resource planning method based on weak voltage region

Abstract

The invention relates to the technical field of power grid resource planning, and discloses a power grid resource planning method based on a voltage weak region. The method comprises the following steps: acquiring an initial voltage data set of a power grid, and generating a target voltage data set through integrity verification and format standardization processing; evaluating the voltage stability of the target voltage data set to generate a voltage weak identifier, if the voltage weak identifier indicates potential weakness, splitting the target voltage data set into a plurality of voltage analysis units, and filtering invalid units after independent evaluation; voltage index values are extracted from the filtered units, deviation is calculated from the voltage index values and reference voltage index values, the voltage index values and the reference voltage index values are grouped to form an index sequence, the index values in the sequence are arranged in an ascending order, continuous values are taken as candidate association groups, association is detected through an association detection algorithm, and the number of the association groups is summarized; and in combination with the number of the association groups and the voltage index values which are not included in association, evaluating a power grid resource demand level, and generating a power grid resource optimization scheme according to the level.

Description

Technical Field

[0001] This invention relates to the field of power grid resource planning technology, specifically a power grid resource planning method based on voltage-deficient areas. Background Technology

[0002] With the rapid development of the social economy, various types of electricity loads continue to grow, the power grid scale is constantly expanding, and the power grid operation status is becoming increasingly complex, placing higher demands on the scientific and precise nature of power grid resource planning. As a crucial task for ensuring the safe and stable operation of the power grid and meeting users' electricity needs, the effectiveness of power grid resource planning directly impacts the grid's operational efficiency and service quality. During power grid operation, voltage, as one of the core operational indicators, plays a key role in the overall operational status of the power grid. The existence of voltage-weak areas can easily lead to voltage fluctuations, decreased power supply quality, and even more serious power grid faults. Therefore, conducting power grid resource planning around voltage-weak areas has become a key focus within the industry.

[0003] However, current power grid resource planning methods still have many shortcomings in the analysis and application of voltage-deficient areas. After acquiring power grid voltage data, existing methods often lack a comprehensive data processing workflow and fail to rigorously verify the integrity of initial voltage data, easily leading to incomplete or abnormal data entering subsequent analysis stages. Furthermore, they fail to standardize data formats, making it difficult to integrate and analyze voltage data from different sources and in different formats, thus affecting the accuracy of subsequent voltage stability assessments and failing to provide a reliable data foundation for power grid resource planning.

[0004] Existing methods for assessing grid voltage stability often employ a holistic approach, failing to segment and independently analyze areas potentially prone to voltage vulnerabilities. When the overall grid voltage stability assessment results are critical or do not explicitly indicate vulnerabilities, it is difficult to accurately pinpoint localized voltage-vulnerable areas. This leads to subsequent resource planning failing to specifically cover these potentially vulnerable areas, resulting in uneven grid resource allocation, with some non-vulnerable areas having redundant resources while vulnerable areas suffer from insufficient resources, thus failing to effectively address potential voltage risks in grid operation.

[0005] Existing methods for analyzing voltage indicators lack in-depth detection of the correlations between indicators. They typically analyze individual voltage indicators independently, neglecting the inherent relationships between different indicators. This makes it difficult to identify correlated combinations of voltage indicators, and consequently, to determine the specific distribution and impact range of weak voltage areas in the power grid based on indicator correlation characteristics. Furthermore, when assessing the level of power grid resource demand, existing methods often rely on single-dimensional indicator data, failing to comprehensively consider indicator correlations and independent indicator information. This results in one-sided resource demand level assessments that cannot fully reflect the actual resource needs of the power grid. The resulting power grid resource optimization schemes often deviate from the actual operational needs of the power grid, failing to fully utilize power grid resources, effectively improve the stability and economy of power grid operation, and adapt to the resource planning requirements of complex power grid environments. Summary of the Invention

[0006] The purpose of this invention is to provide a power grid resource planning method based on voltage-deficient areas to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a power grid resource planning method based on voltage-deficient areas, the method comprising:

[0008] Acquire the initial voltage data set in the power grid operation status, and perform integrity verification and format standardization on the initial voltage data set to generate the target voltage data set;

[0009] The voltage stability of the target voltage dataset is evaluated, and a voltage weakness indicator is generated based on the evaluation results. When the voltage weakness indicator indicates a potential weakness, the target voltage dataset is split into multiple voltage analysis units, each voltage analysis unit is evaluated independently, and the voltage analysis units are filtered based on the independent evaluation results.

[0010] Voltage index values ​​are extracted from the filtered voltage analysis unit, and the deviation of each voltage index value is calculated with the corresponding reference voltage index value. Based on the deviation calculation results, the voltage index values ​​are grouped to form an index sequence. The voltage index values ​​in the index sequence are sorted in ascending order, and consecutive voltage index values ​​after sorting are taken as candidate association groups. The association detection algorithm is used to check the association within the candidate association groups, and the number of association groups is summarized.

[0011] Assess the level of power grid resource demand by combining the number of associated groups and the voltage index values ​​not included in the association;

[0012] Based on the power grid resource demand level, a power grid resource optimization scheme is generated.

[0013] Preferably, when acquiring an initial voltage data set in the power grid operating state, and performing integrity verification and format standardization on the initial voltage data set to generate a target voltage data set, the process includes:

[0014] Multi-resolution wavelet decomposition is performed on the initial voltage data set to extract high-frequency noise components and low-frequency signal components. After filtering out the high-frequency noise components, the voltage signal is reconstructed.

[0015] The reconstructed voltage signal is time-aligned by converting heterogeneous sampled data into a sequence with a uniform time step through linear interpolation.

[0016] Calculate the local smoothness index for each voltage sampling point in a sequence with a uniform time step, and evaluate the smoothness based on the voltage gradient change within the sliding window;

[0017] When the local smoothness index exceeds the preset smoothness threshold, the voltage sampling point is classified as a valid point; otherwise, it is considered an invalid point and excluded.

[0018] Aggregate all valid points to generate the target voltage data set.

[0019] Preferably, when evaluating the voltage stability of the target voltage data set and generating a voltage weakness indicator based on the evaluation result, the process includes:

[0020] The target voltage data set is segmented into time series, dividing the continuous voltage data into multiple time windows of fixed duration;

[0021] For each time window, the statistical characteristics of the voltage values ​​within the window are calculated, including the mean, variance, skewness, and kurtosis, forming a window feature vector;

[0022] The window feature vectors of all time windows are aggregated, and the dimensionality is reduced by principal component analysis to obtain the compressed overall feature vector.

[0023] Calculate the cosine similarity between the overall feature vector and the pre-stored standard voltage feature vector to obtain a similarity score;

[0024] A voltage stability indicator is generated when the similarity score is higher than or equal to the stability threshold.

[0025] When the similarity score is lower than or equal to the abnormal threshold, a voltage abnormality indicator is generated;

[0026] When the similarity score falls between the stable threshold and the abnormal threshold, a potential voltage weakness indicator is generated.

[0027] Preferably, when evaluating each voltage analysis unit independently and filtering voltage analysis units based on the independent evaluation results, the process includes:

[0028] For each voltage analysis unit, the voltage fluctuation variance is calculated. When the voltage fluctuation variance is lower than a preset variance threshold, the voltage analysis unit is determined to be a stable unit and removed.

[0029] For the remaining voltage analysis units, calculate the voltage deviation coefficient for each unit and sort them according to the size of the deviation coefficient;

[0030] Voltage analysis units with deviation coefficients higher than the preset coefficient are retained, and their voltage data are used as voltage index values.

[0031] Preferably, when extracting voltage index values ​​from the filtered voltage analysis unit, calculating the deviation between each voltage index value and the corresponding reference voltage index value, and grouping the voltage index values ​​based on the deviation calculation results to form an index sequence, the process includes:

[0032] Calculate the percentage of relative deviation between each voltage index value and the reference voltage index value to obtain the deviation value;

[0033] Based on the sign of the deviation values, the set of positive deviation values ​​is divided into the upper deviation group, and the set of negative deviation values ​​is divided into the lower deviation group.

[0034] In the upper deviation group, continuous deviation values ​​belonging to the same voltage analysis unit are clustered into upper deviation clusters, and isolated deviation values ​​are classified into upper deviation points.

[0035] In the lower deviation group, continuous deviation values ​​belonging to the same voltage analysis unit are clustered into lower deviation clusters, and isolated deviation values ​​are classified into lower deviation points.

[0036] Preferably, when the voltage index values ​​in the index sequence are sorted in ascending order, and consecutive voltage index values ​​after sorting are used as candidate association groups, the following steps are included:

[0037] The deviation values ​​in the upper deviation cluster are sorted from smallest to largest, and adjacent deviation values ​​after sorting are paired as upper deviation candidate pairs.

[0038] The deviation values ​​in the lower deviation cluster are sorted from smallest to largest, and adjacent deviation values ​​after sorting are paired as candidate pairs of lower deviations.

[0039] For the upper and lower deviation points, no pairing operation is performed; they are directly retained as independent values.

[0040] Preferably, when using an association detection algorithm to verify the association within a candidate association group, the following steps are included:

[0041] Collect historical voltage index values, reference voltage index values, and historical deviation values ​​during power grid operation, construct a training dataset, and randomly split it into a model training set and a model validation set;

[0042] A long short-term memory neural network model was selected, and the model was trained using the model training set. The model accuracy was evaluated using the model validation set.

[0043] If the loss function value of the current training cycle does not decrease during training, the model learning rate is adjusted and training continues until the maximum training cycle is reached.

[0044] When the loss function value decreases steadily, training is stopped, and the trained long short-term memory neural network model is used as the association detection algorithm.

[0045] Input the upper bias candidate pair and the lower bias candidate pair into the association detection algorithm, and output the association strength score;

[0046] When the association strength score is higher than the preset association threshold, the candidate is determined to be valid for association, and the number of association groups is recorded.

[0047] Preferably, when assessing the grid resource demand level by combining the number of associated groups and voltage index values ​​not included in the association, the following are included:

[0048] Calculate the deviation range width of the upper deviation cluster, that is, the difference between the maximum deviation value and the minimum deviation value;

[0049] Calculate the deviation range width of the lower deviation cluster, i.e., the difference between the maximum deviation value and the minimum deviation value;

[0050] The sum of the deviations at the upper deviation point and the sum of the deviations at the lower deviation point;

[0051] The number of associated groups, the deviation range width of the upper deviation cluster, the deviation range width of the lower deviation cluster, the sum of the deviation values ​​of the upper deviation points, and the sum of the deviation values ​​of the lower deviation points are input into the fuzzy inference system, and the power grid resource demand level is output. The fuzzy rules are dynamically configured according to the power grid operating load.

[0052] Preferably, when generating a power grid resource optimization scheme based on the power grid resource demand level, the process includes:

[0053] Set threshold values ​​for the power grid resource demand level and compare the power grid resource demand level with the threshold values;

[0054] When the power grid resource demand level exceeds the threshold, the resource optimization engine is activated to generate optimization solutions, including increasing capacitor compensation, adjusting transformer taps, and redistributing loads.

[0055] When the power grid resource demand level is lower than or equal to the threshold, the current resource allocation strategy is maintained, and only a monitoring report is generated.

[0056] Preferably, after generating the power grid resource optimization scheme, the method further includes: deploying the power grid resource optimization scheme to a distributed control system, monitoring voltage changes in real time, and dynamically adjusting the scheme parameters.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] In the data processing stage, by performing integrity verification and format standardization on the initial voltage dataset, incomplete and abnormal data in the initial data can be effectively eliminated, avoiding invalid data from interfering with subsequent analysis. At the same time, voltage data from different sources and in different formats are unified into a standard format, realizing efficient data integration and unified analysis. This ensures that the target voltage dataset entering subsequent stages has high reliability and consistency, providing high-quality data support for subsequent voltage stability assessment, voltage weak area identification, and other work, thus ensuring the accuracy of power grid resource planning and analysis from the data source.

[0059] In the identification of voltage-weak areas, this method first performs an overall voltage stability assessment on the target voltage dataset and generates voltage-weakness indicators. When a potential weakness is indicated, the target voltage dataset is further divided into multiple voltage analysis units for independent assessment and filtering. This "overall assessment + local segmentation" approach overcomes the limitation of existing methods that cannot accurately locate local weak areas through overall assessment. By analyzing each voltage analysis unit independently, specific areas in the power grid with voltage-weakness problems can be accurately identified, filtering out units without weakness risk. This allows subsequent analysis to focus on critical voltage-weak areas, avoiding ineffective attention to non-weak areas during resource planning, improving the targeting of power grid resource planning, and reducing resource waste.

[0060] In the voltage index analysis stage, this method extracts voltage index values ​​from the filtered voltage analysis units, calculates the deviation from the benchmark voltage index values, and groups them into index sequences. Then, through ascending sorting, identifying candidate correlation groups, and using correlation detection algorithms to verify correlations, it can deeply explore the inherent relationships between different voltage indices, identify correlated voltage index combinations, and summarize the number of correlation groups. This analysis of index correlations helps staff to more clearly understand the changing patterns and mutual influences of voltage indices in voltage-weak areas, and to identify key index combinations affecting voltage stability, rather than relying solely on a single index for judgment. This makes the analysis of voltage-weak areas more comprehensive and in-depth, providing richer and more valuable analytical basis for subsequent power grid resource demand level assessments.

[0061] In the power grid resource demand level assessment and scheme generation stage, this method combines the number of associated groups with the voltage index values ​​not included in the association for comprehensive evaluation. This approach considers both the correlation characteristics of the indicators and the information of independent indicators, avoiding the one-sided results caused by the single-dimensional assessment of existing methods. By comprehensively analyzing these two types of information, the actual demand for resources in different regions of the power grid can be more accurately determined, and a scientifically reasonable power grid resource demand level can be defined. The power grid resource optimization scheme generated based on this demand level can allocate resources according to the demand differences in different regions, prioritizing the allocation of necessary resources in voltage-deficient areas to specifically address voltage problems in these areas. Simultaneously, resource input in non-voltage-deficient areas is rationally controlled, achieving optimal allocation of power grid resources. This ensures that power grid resources can be fully utilized, helping to improve the stability and economy of power grid operation and better meet the electricity needs of society. Attached Figure Description

[0062] Figure 1 This is a schematic diagram illustrating the working principle of the power grid resource planning method based on voltage-weak areas described in this invention.

[0063] Figure 2 A flowchart for generating the target voltage dataset;

[0064] Figure 3 A flowchart for calculating deviations and grouping them to form an index sequence;

[0065] Figure 4 The graph shows the grouping and clustering analysis results of the grid voltage deviation.

[0066] Figure 5 This is a map for assessing the level of power grid resource demand. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Please see Figure 1This invention provides a power grid resource planning method based on voltage-weak areas. The method includes: acquiring an initial voltage data set from the power grid's operating state, performing integrity verification and format standardization on the initial voltage data set, and generating a target voltage data set. The voltage stability of the target voltage data set is evaluated, and voltage weakness indicators are generated based on the evaluation results. When a voltage weakness indicator indicates a potential weakness, the target voltage data set is split into multiple voltage analysis units, each of which is independently evaluated. Voltage analysis units are filtered based on the independent evaluation results. Voltage index values ​​are extracted from the filtered voltage analysis units. The deviation of each voltage index value from its corresponding reference voltage index value is calculated. Based on the deviation calculation results, the voltage index values ​​are grouped to form an index sequence. The voltage index values ​​in the index sequence are arranged in ascending order, and consecutive voltage index values ​​after arrangement are used as candidate association groups. An association detection algorithm is used to verify the association within the candidate association groups, and the number of association groups is summarized. The number of association groups and the voltage index values ​​not included in the association are combined to evaluate the power grid resource demand level. The power grid resource demand level is used to generate a power grid resource optimization scheme.

[0069] Example 1: See Figure 2 The initial voltage data set undergoes multi-resolution wavelet decomposition, which aims to separate the mixed original voltage signal based on its frequency components. This process is implemented using a discrete wavelet transform algorithm. The selected wavelet basis functions must possess tight support and regularity suitable for the power grid's power frequency signal. Through multi-level decomposition of the initial voltage data set, the signal is analyzed into high-frequency noise components and low-frequency signal components at different levels. High-frequency noise components typically correspond to random errors introduced by the measuring device, transient interference in the communication channel, or instantaneous spikes caused by switching operations. Low-frequency signal components carry the core voltage information of the power grid's steady-state operation and slow dynamic processes. Filtering out high-frequency noise components is accomplished by setting the wavelet coefficients above a specific scale after decomposition to zero. The retained low-frequency wavelet coefficients are used for signal reconstruction. The reconstructed voltage signal is based on the processed low-frequency wavelet coefficients. The inverse discrete wavelet transform algorithm is used to resynthesize the processed coefficients into a time-domain voltage signal. The reconstruction process eliminates most of the high-frequency noise, making the voltage waveform smoother and better reflecting the true operating status of the power grid.

[0070] The reconstructed voltage signal undergoes time alignment processing to address the inconsistency on the time axis caused by voltage data from different monitoring points or with different sampling rates. Heterogeneous sampling data means that the timestamps of the data points are not uniformly distributed. Linear interpolation is applied to solve this problem. Among the known non-uniform sampling points, a series of voltage estimates at uniform time points are calculated through linear relationships. Linear interpolation is then converted into a sequence with a uniform time step. The uniform time step is set according to the highest requirements or minimum common sampling rate of power grid monitoring. The voltage value of each newly generated data point is linearly calculated from the voltage values ​​of the two adjacent original data points, ultimately forming a voltage time series with a constant time interval.

[0071] The local smoothness index for each voltage sampling point in the unified time series is calculated. This index quantifies the drastic change in the voltage signal within a local time window. Smoothness is evaluated based on the voltage gradient change within the sliding window. The size of the sliding window must balance sensitivity to noise with the response speed to actual voltage changes. The voltage gradient within the window is obtained by calculating the absolute value or root mean square value of the voltage difference between adjacent sampling points. When the local smoothness index exceeds a preset smoothness threshold, the voltage sampling point is classified as a valid point. The preset smoothness threshold is determined based on the statistical distribution of voltage gradients in historical stable operating data; for example, the high percentile of the distribution is used as the threshold. When the calculated gradient value is below this threshold, it indicates that the voltage change near that point is gradual and falls within the normal fluctuation range. When the local smoothness index does not exceed the preset smoothness threshold, the voltage sampling point is considered an invalid point and is excluded. Invalid points may correspond to residual interference not completely removed by wavelet filtering, data acquisition interruptions, or the start point of abnormal events. These points are marked and removed from the dataset to prevent them from misleading subsequent stability analysis.

[0072] Aggregating all valid points generates the target voltage dataset. This aggregation operation gathers valid voltage sampling points scattered across the entire time axis, forming a clean and time-aligned voltage dataset. This target voltage dataset serves as the sole data source for subsequent voltage stability assessment and weak area identification; its quality directly determines the accuracy and reliability of the entire resource planning method. The specific implementation of multi-resolution wavelet decomposition involves selecting the number of decomposition layers. Too few layers may lead to incomplete noise filtering, while too many layers may cause excessive signal smoothing and loss of detail. The optimal number of decomposition layers is typically determined based on the relationship between the main power grid disturbance frequency and the sampling frequency. For wavelet basis functions, Db4 or Symlets series wavelets are considered. These functions have a good match with power signals in terms of time-frequency localization characteristics. The discrete wavelet transform is implemented using the Mallat algorithm, which iteratively decomposes the signal using high-pass and low-pass filter banks, offering high computational efficiency suitable for processing massive amounts of power grid data. Linear interpolation is the preferred method for time alignment due to its simplicity and computational speed. It provides sufficiently accurate estimates when most sampling points are dense. For more demanding applications, spline interpolation can be used to obtain smoother curves, but the computational complexity increases accordingly. The sliding window size needs to be set in conjunction with the actual data sampling rate. A window that is too small may not effectively filter out random fluctuations, while a window that is too large may mask real rapid voltage changes. An empirical approach is to make the window length cover several power frequency cycles. The calculation of local smoothness indices is not limited to gradients; it can also incorporate the first or second-order differential statistical characteristics of the signal. Voltage gradient change evaluation can be extended to the calculation of the standard deviation or coefficient of variation of voltage values ​​within the window, measuring local volatility from multiple perspectives. The determination of valid points is not absolute. A soft-decision mechanism can be introduced, assigning a smoothness confidence weight to each point. In subsequent analysis, a weighted processing method is used to ensure that points close to the threshold edge are not completely discarded, thus retaining more potential information. The generation of the target voltage dataset marks the completion of the data preprocessing stage. This dataset not only contains voltage numerical information but also implicitly includes timestamp information after cleaning and alignment.

[0073] Example 2: See Figure 3The target voltage dataset undergoes time-series segmentation, dividing the continuous voltage observation data stream into a series of non-overlapping, fixed-duration time windows. Each time window serves as an independent analysis unit. The window length must consider the periodicity of the power grid operation and the typical duration of disturbance events. A reasonable window length should capture several voltage fluctuation cycles without being too long to mask local dynamic characteristics. Fixed-duration time windows provide a consistent benchmark for statistical analysis. The window division begins at the start of the data sequence and progresses at fixed time intervals until the entire target voltage dataset is covered. Adjacent windows are closely linked in time to ensure no voltage data points are missed. For each time window, the statistical characteristics of the voltage values ​​within the window are comprehensively calculated. These characteristics include mean, variance, skewness, and kurtosis. The mean reflects the central tendency or average level of the voltage within the window period; variance quantifies the dispersion or fluctuation intensity of voltage values ​​around the mean; skewness describes the asymmetry of the voltage distribution, indicating a tendency for voltage values ​​to skew above or below the mean; and kurtosis measures the steepness of the voltage distribution and is associated with the likelihood of extreme voltage events. The calculated mean, variance, skewness, and kurtosis together constitute a multi-dimensional window feature vector. This window feature vector comprehensively represents the core statistical properties of voltage behavior within the time window. Each window feature vector corresponds to a snapshot of the power grid's operating status within a specific time period.

[0074] The feature vectors of all time windows are aggregated, resulting in a high-dimensional dataset whose dimension equals the number of windows multiplied by the number of features in each vector. Directly processing such a high-dimensional dataset presents challenges such as high computational complexity and potential multicollinearity among features. Principal component analysis (PCA) is used to reduce the dimensionality of the aggregated high-dimensional window feature vector set. PCA transforms the original correlated features into a new set of uncorrelated variables—principal components—through linear transformation. These principal components are arranged in descending order of variance contribution rate, retaining the top few principal components whose cumulative variance contribution rate reaches a preset threshold. This yields a compressed overall feature vector, which retains most of the variation information from the original window feature vector set at a lower dimension. This achieves data simplification while maximizing the preservation of key features reflecting the overall voltage status of the power grid.

[0075] The system calculates the cosine similarity between the overall feature vector and the pre-stored standard voltage feature vector. The pre-stored standard voltage feature vector is derived from historical data of the power grid under known stable operating conditions, obtained through the same time series segmentation, statistical feature extraction, and principal component analysis dimensionality reduction process. Cosine similarity measures the similarity by calculating the cosine of the angle between the two vectors in their directions. The cosine similarity score is output as a score, ranging from -1 to 1. A score closer to 1 indicates a more consistent direction between the overall feature vector and the standard voltage feature vector, meaning the current power grid voltage operating state is more similar to the historical stable state. When the similarity score is higher than or equal to a preset stability threshold (set to a value close to 1), the system generates a voltage stability indicator, indicating that the power grid voltage is operating within a safe and reliable stable range, and no immediate resource optimization measures are needed. When the similarity score is lower than or equal to a preset anomaly threshold (set to a relatively low value), the system generates a voltage anomaly indicator, indicating a significant deviation from the stable state and a potential significant voltage problem. When the similarity score is between the stable threshold and the abnormal threshold, the system generates a potential voltage weakness indicator. The potential voltage weakness indicator indicates that the grid voltage state is in the critical region. Although it has not reached the abnormal level, it has shown an unstable tendency and requires further analysis to identify potential weak links.

[0076] The generation of potential voltage weakness indicators triggers the subsequent action of splitting the target voltage dataset into multiple voltage analysis units. Each voltage analysis unit corresponds to a more refined data block, which may be divided based on the geographical area of ​​the power grid, electrical islands, or specific voltage levels. For each voltage analysis unit, the voltage fluctuation variance is calculated. Voltage fluctuation variance characterizes the degree of voltage value change over time within that unit, calculated as the average of the sum of squared deviations of all voltage sampling points within the unit relative to its average value. When the voltage fluctuation variance is below a preset variance threshold, the voltage analysis unit is determined as a stable unit and removed from the analysis set. The preset variance threshold is determined based on the allowable voltage fluctuation range under normal power grid operating conditions. Low variance indicates that voltage changes are gradual and within the normal operating state, requiring no further attention. For the remaining voltage analysis units after variance screening, the voltage deviation coefficient for each unit is calculated. The voltage deviation coefficient is typically defined as the percentage of the absolute difference between the unit's average voltage and the rated or expected voltage value, quantifying the average deviation of the voltage level.

[0077] The remaining voltage analysis units are sorted based on their voltage deviation coefficients. The sorting process places the unit with the highest voltage deviation coefficient at the top of the list and the unit with the lowest at the bottom, allowing the analysis to focus on the areas with the most significant deviations. Voltage analysis units with voltage deviation coefficients higher than a preset coefficient—a threshold value set according to voltage quality standards and grid safety margins—are retained. Units exceeding this threshold are considered voltage weak points, and their voltage data is extracted as key voltage indicators for subsequent in-depth deviation calculations and correlation analysis. The calculation of the voltage deviation coefficient can employ more complex formulas, such as introducing weighting coefficients to reflect the different impacts of voltage at different nodes on system stability. Voltage deviations at important hub nodes may be given higher weights, thus receiving more priority in the sorting. The preset coefficient is not fixed and can be dynamically adjusted based on the total system load level, seasonal changes, or operating modes. During peak load periods, when voltage stability margins decrease, the preset coefficient can be appropriately lowered to capture more potentially risky units. The retained voltage analysis units and their voltage index values ​​constitute the direct output for identifying voltage-weak areas, providing accurate input data for subsequent power grid resource demand assessment based on deviation analysis. The entire assessment process extends from overall stability judgment to the location of local weak points, achieving multi-level and refined perception of the power grid voltage status. There is an inherent relationship between the length of the time window and the granularity of the voltage analysis unit division. Fineer unit division helps to accurately locate weak points but increases computational load, requiring a balance between accuracy and efficiency. The calculation of window feature vectors can incorporate rolling averages or exponentially weighted moving averages to smooth instantaneous fluctuations, making statistical features better reflect trend changes. The number of principal components retained in principal component analysis is determined by the cumulative variance contribution rate. The standard voltage feature vector library can contain feature vectors from multiple typical operating scenarios. The similarity score can be the maximum similarity or weighted average of all standard vectors in the library to enhance the robustness of the status assessment. The calculation of voltage fluctuation variance can use the variance of the detrended residual sequence to eliminate the impact of long-term trend changes on fluctuation assessment. The ranking results of the voltage deviation coefficients can be visualized to help operators intuitively identify the weakest link in the system.

[0078] Example 3: The voltage index values ​​extracted from the filtered voltage analysis unit need to be compared with a benchmark to quantify their deviation. The deviation of each voltage index value from the corresponding benchmark voltage index value is calculated. The benchmark voltage index value is derived from the standard voltage value set under ideal steady-state operating conditions of the power grid or the expected voltage value obtained based on historical high-quality data. The calculation is expressed as a relative deviation percentage, and the formula is as follows:

[0079]

[0080] Where: symbol Representing the The percentage relative deviation of each voltage index value, symbol Representative from the first The actual measured voltage index value extracted from each voltage analysis unit, symbol... Representative and the For each voltage index value, a corresponding reference voltage index value is obtained. Through this calculation, a quantified deviation value is obtained for each voltage index value. A positive value indicates that the voltage is too high, and a negative value indicates that the voltage is too low.

[0081] Based on deviation value The positive and negative signs of the deviation values ​​are used to initially group the entire set of deviation values. All positive deviation values ​​are grouped together to form the upper deviation group, which includes all operating conditions where the voltage measurement value exceeds its corresponding reference value. All negative deviation values ​​are grouped together to form the lower deviation group, which includes all operating conditions where the voltage measurement value is below its corresponding reference value. Within the upper deviation group, further clustering analysis is conducted based on the temporal continuity or spatial correlation of the deviation values. Positive deviation values ​​belonging to the same voltage analysis unit and appearing consecutively in the time series are classified into an upper deviation cluster. An upper deviation cluster represents a persistent voltage spike in a specific region or node. The clustering process identifies consecutive positive subsequences in the deviation value sequence within the voltage analysis unit. Each such subsequence constitutes an independent upper deviation cluster, and the size of the cluster reflects the duration of the voltage spike. Isolated positive deviation values ​​that do not form a continuous sequence in the upper deviation group are individually marked as upper deviation points. Upper deviation points typically correspond to brief, transient voltage spikes, which may be caused by switching operations or sudden load shedding. Within the lower deviation group, clustering logic similar to that of the upper deviation group is performed. Negative deviation values ​​belonging to the same voltage analysis unit and occurring consecutively in the time series are grouped into lower deviation clusters. Lower deviation clusters indicate a persistent low voltage problem in a part of the power grid, which may be due to heavy load, insufficient reactive power, or excessive line impedance. Identification of consecutive negative deviation values ​​relies on scanning the data sequence of each voltage analysis unit. Once a consecutive segment with a negative sign is found, it is classified into a lower deviation cluster. The duration of the cluster reveals the severity and persistence of the voltage drop event. Individual negative deviation values ​​scattered throughout the sequence and not included in any lower deviation cluster are defined as lower deviation points. Lower deviation points are often associated with brief voltage dips, possibly caused by the starting of large motors or remote faults.

[0082] After grouping and clustering, the voltage index values ​​in the index sequence are sorted in ascending order. This sorting operation is performed on the deviation values ​​within each cluster. Instead of the original voltage measurements, the data is processed. All deviation values ​​in the upper deviation cluster are sorted in ascending order of magnitude, from smallest to largest. This sorted sequence more clearly shows the distribution range of voltage overshoot within the cluster, from slight over-limit to severe overvoltage. In the lower deviation cluster, the same ascending order is performed. Since the deviation values ​​are negative, ascending order means arranging from the least negative deviation value to the most negative one, revealing a gradient change in voltage undershoot. The sorted, consecutive voltage values ​​are used as candidate correlation groups. For the sorted upper deviation cluster, adjacent deviation values ​​are paired to form a series of upper deviation candidate pairs. Each upper deviation candidate pair contains two numerically adjacent deviation values, which may represent a continuous change in voltage overshoot within a small range. For the sorted lower deviation cluster, adjacent deviation values ​​are also paired to form lower deviation candidate pairs. These lower deviation candidate pairs help analyze the continuity of the voltage drop trend. The pairing operation correlates consecutive deviation values, preparing for subsequent testing of whether there is a statistical or physical intrinsic relationship between them. For the previously separated upper and lower deviation points, since they lack continuity and are isolated instances, no pairing operation is performed. These independent values ​​will participate separately in the subsequent power grid resource demand level assessment, representing sporadic and discontinuous voltage anomaly events in the system. The formation of candidate correlation groups is based on numerical proximity and sequence continuity, without pre-assuming their physical correlation. The true correlation needs to be verified by subsequent correlation detection algorithms. This method ensures a comprehensive and unbiased exploration of voltage data patterns. The calculation of deviation values ​​relies on accurate reference voltage index values. The setting of the reference value can be dynamically adjusted, for example, updated according to seasonal, daily load curve, or network topology changes, to improve the accuracy of deviation calculation. The clustering algorithm can introduce a maximum time interval threshold for judging continuity. Even if the indices are not absolutely continuous, deviation values ​​with the same sign whose timestamps differ within the threshold can be grouped into the same cluster, enhancing the ability to handle aligned continuous events. The sorting process can use efficient sorting algorithms such as quicksort to handle deviation clusters that may contain a large number of data points. The candidate pair generation strategy can be flexible, for example, overlapping pairs can be generated to capture more local correlation information. The processing of independent points can include analyzing their frequency and distribution patterns. High-frequency independent points may indicate some unidentified intermittent problem, and the absolute value of independent points is also an important indicator of their severity. The entire process, from deviation calculation to candidate correlation group formation, transforms the original voltage measurement value sequence into a series of structured deviation groups, clusters, points, and candidate pairs. This structured data representation greatly facilitates subsequent in-depth mining and correlation analysis of voltage weakness patterns.

[0083] See Figure 4The graph displays the deviation changes of multiple voltage analysis units, with different colored curves representing different analysis units. Red triangles mark significant positive deviation points, and blue inverted triangles mark significant negative deviation points. The graph clearly distinguishes between upper and lower deviation regions, and dashed threshold lines indicate the boundaries of normal operation. The graph visually illustrates the continuous patterns and transient anomalies of voltage deviation, including persistent high and low voltage clusters, as well as isolated transient deviation points. This visualization analysis provides important basis for identifying weak areas in the power grid and subsequent correlation analysis.

[0084] Example 4: The association detection algorithm relies on training with historical data. It collects voltage index values, reference voltage index values, and historical deviation values ​​from historical power grid operations. This data covers the power grid's state under different load levels, seasons, and operating conditions, forming a comprehensive dataset encompassing multiple scenarios. The constructed training dataset contains tens of thousands of records, each containing a voltage index value at a specific time point, the corresponding reference voltage index value, and the calculated historical deviation value. Data sources include SCADA systems, PMU measurement devices, and historical operation logs. A random partitioning operation divides the constructed training dataset into two independent parts: a model training set for directly adjusting the model's internal parameters, and a model validation set for independently evaluating the model's generalization ability during training to prevent overfitting. A common partitioning ratio is 70% for the model training set and 30% for the model validation set. A Long Short-Term Memory (LSTM) neural network model is chosen as the core architecture of the association detection algorithm. LSTM is a special type of recurrent neural network whose internal gating mechanism effectively captures long-term dependencies in time-series data, which is crucial for analyzing potential temporal correlation patterns in voltage deviation sequences. The model is trained using a training set, and the training process is an iterative optimization process. Historical bias sequences from the training set are used as input to the Long Short-Term Memory (LSTM) neural network model. The model's goal is to learn how to predict the strength of associations between candidate biases based on the input. The training process uses backpropagation and gradient descent algorithms to minimize the difference between the model's predicted output and the true associated labels. The model's accuracy is evaluated using a validation set. After each training cycle, the current LSM model is tested on unseen validation set data, and its prediction accuracy, precision, and recall are calculated to monitor the model's performance on unknown data.

[0085] During training, if the loss function value of the current training epoch does not decrease, the model learning rate is adjusted and training continues. The loss function value measures the degree of error in the model's predictions. A stagnant loss function value may indicate that the optimization process has fallen into a local optimum or that the learning rate is improperly set. Dynamically adjusting the model learning rate, such as reducing the learning rate, can help the model break out of the plateau. Training continues until the preset maximum number of training epochs is reached. This is a safe stopping mechanism to prevent infinite training loops. The maximum number of training epochs is set to a sufficiently large value based on the dataset size and computational resources. When the loss function value shows a stable decreasing trend on the model validation set and no longer shows significant improvement over several consecutive epochs, the training process is stopped. At this point, it is considered that the Long Short-Term Memory (LSTM) neural network model has acquired sufficient knowledge and is not overfitting the training data. The trained LSM model is saved and used as the final association detection algorithm. Upper-biased candidate pairs and lower-biased candidate pairs are input into the association detection algorithm. Each candidate pair is fed into the LSM model as a short sequence. After processing the sequence information, the LSM model generates an association strength score in its output layer. This score is a continuous value between 0 and 1. When the correlation strength score is higher than a preset correlation threshold, the candidate pair is deemed to be correlated. The preset correlation threshold is a threshold value that needs to be carefully selected, such as 0.7 or 0.8. A high threshold ensures that only strongly correlated deviation pairs are counted. Candidate pairs with valid correlation are marked and included in the correlation group count. The correlation group count is the total number of all upper and lower deviation candidate pairs that are determined to be correlated and valid. This value quantifies the number of voltage deviation patterns with inherent correlation in the system.

[0086] The deviation range width of the upper deviation cluster is calculated. This width is defined as the arithmetic difference between the maximum and minimum deviation values ​​within the upper deviation cluster. This indicator reflects the fluctuation range of the voltage spike event represented by the cluster; a large width value indicates unstable and drastic voltage spikes. Similarly, the deviation range width of the lower deviation cluster is calculated, defined as the difference between the maximum and minimum deviation values ​​within the lower deviation cluster. Since the deviation values ​​are negative, the absolute value is taken or the values ​​are directly subtracted. The result reflects the fluctuation range of the voltage dip event. The total deviation values ​​of the upper deviation points are calculated by algebraically adding all deviation values ​​classified as upper deviation points. This sum reflects the cumulative severity of those brief, isolated voltage spike events in the system. The total deviation values ​​of the lower deviation points are also calculated by algebraically adding all deviation values ​​classified as lower deviation points. Since the values ​​are negative, the sum is negative, and its absolute value represents the overall impact of isolated voltage dip events.

[0087] The number of correlation groups, the deviation range width of the upper deviation cluster, the deviation range width of the lower deviation cluster, the sum of the deviation values ​​of the upper deviation points, and the sum of the deviation values ​​of the lower deviation points are fed into a fuzzy inference system as input variables. Fuzzy inference systems are effective tools for handling imprecise and fuzzy information, particularly suitable for transforming verbal descriptions such as "strong correlation" and "large fluctuation range" into quantitative evaluation results. The fuzzy inference system outputs a power grid resource demand level, which is a comprehensive evaluation index. For example, it can be defined as "low," "medium," or "high," or represented by a numerical level of 1-5. The fuzzy inference system contains a rule base consisting of a series of "if...then..." fuzzy rules are dynamically configured according to the power grid operating load. This means that the rules are not static but adjust with changes in the total system load. For example, during high load periods, the rules may be more sensitive to negative voltage deviations, and even the same input value may trigger a higher resource demand level. The design of the fuzzy inference system comprises three steps: fuzzification, inference engine, and defuzzification. Input variables are transformed into fuzzy sets using membership functions. The inference engine calculates the output fuzzy set according to fuzzy rules. Finally, the defuzzification method transforms the fuzzy output into a precise power grid resource demand level. Table 1, a simplified fuzzy rule table, illustrates the logical relationship between the input variable states and the output power grid resource demand level.

[0088] Table 1: Fuzzy Rule Table

[0089]

[0090] The training data for Long Short-Term Memory (LSTM) neural network models requires high-quality annotation. The true labels for the strength of correlations may originate from the analysis of historical events; for example, simultaneous line overloads and voltage deviations can be considered a strong correlation. The annotation process requires domain knowledge. In a fuzzy inference system, each input variable needs a defined membership function. The shape and parameters of the membership function need to be designed based on variable characteristics and expert experience. The fuzzy set partitioning of the input variables must cover their possible value range. Dynamically configuring fuzzy rules can be achieved through an adjustment factor linked to real-time load. This adjustment factor modifies the membership function parameters of the output level in the rule conclusion, thereby achieving dynamic fine-tuning of the rule strength and making the evaluation results more closely reflect the current actual operational risks of the power grid. The correlation detection algorithm can not only handle candidate pairs but can also theoretically be extended to analyze longer deviation sequences and detect more complex correlation patterns. The output power grid resource demand level of the fuzzy inference system provides a direct and quantitative decision-making basis for subsequently formulating differentiated resource optimization schemes.

[0091] See Figure 5The graph displays the distribution of resource demand levels at different points in time, with colors transitioning from green to red, corresponding to changes in resource demand from low to high. The number in each cell represents a specific resource demand level, facilitating detailed analysis. The graph also includes assessment statistics, showing the frequency and proportion of each level. The black dashed line illustrates the trend of resource demand levels, reflecting the overall evolution of the power grid's operational status. This visualization effectively combines the output of the correlation detection algorithm with resource demand assessment, providing intuitive decision support for power grid operators and helping to adjust resource allocation strategies in a timely manner to ensure the safe and stable operation of the power grid.

[0092] Example 5: Setting the threshold value for the power grid resource demand level is a key decision point. The threshold value is determined based on a comprehensive analysis of historical power grid operating data, equipment capacity limits, and safety and stability guidelines. For example, in a regional power grid with many aging lines and heavy load growth, the threshold value might be set to level 3 (assuming the demand level range is 1 to 5, with 5 being the highest). The comparison operation between the power grid resource demand level and the threshold value is implemented through a numerical comparator logic module. The comparator receives the power grid resource demand level value output from the fuzzy inference system and the preset threshold value value, compares their magnitudes, and outputs a Boolean logic signal. When the power grid resource demand level exceeds the threshold value, the resource optimization engine is triggered. The resource optimization engine is a software module containing various optimization algorithms and strategy libraries, and its core task is to generate a specific and feasible power grid resource optimization scheme. The generated power grid resource optimization scheme is a collection of specific measures, including increasing capacitor compensation, adjusting transformer tap changes, and redistributing loads. These measures aim to synergistically improve the power grid voltage condition from three dimensions: reactive power compensation, voltage regulation, and active power balance. To address low voltage issues caused by excess inductive reactive power or insufficient capacitive reactive power in the system, the optimization algorithm calculates the required capacitor bank capacity on specific substation buses or lines based on the distribution and severity of lower deviation clusters and points. For example, if a persistent low voltage is identified at a load center in an industrial area, the solution might suggest adding a 6000 kvar capacitor bank to the 10 kV bus of the corresponding 110 kV substation. Adjusting transformer tap positions utilizes the transformation ratio adjustment capability of on-load tap-changing transformers to change voltage levels. The optimization engine calculates the optimal tap position combination based on the overall network voltage distribution. For example, to raise the voltage of a remote terminal node, it might be necessary to adjust the tap positions of multiple main transformers in substations connecting to that node step by step. The solution will explicitly specify the transformer number and target tap position for each transformer requiring adjustment. Load redistribution measures alleviate local overload and voltage problems by changing the distribution of network power flow. This may require coordination with the dispatch plan to transfer some loads from heavily loaded lines to lightly loaded lines, or to initiate demand-side response if conditions permit. The plan will provide specific load transfer suggestions, such as switching some non-critical loads in a certain area to another power supply line during the evening peak hours.

[0093] When the power grid resource demand level is lower than or equal to the threshold, the system maintains the current resource allocation strategy. This means that the switching status of capacitor banks is not changed, the transformer tap positions are not adjusted, and unplanned load transfer operations are not performed. Only a monitoring report is generated, which records in detail the current power grid resource demand level, the values ​​of each input variable, the comparison results with the threshold, and the basis for the decision to maintain the status quo. The report is generated in a standardized format for review and archiving by operators. The power grid resource optimization scheme is deployed to the distributed control system for execution. The distributed control system adopts a hierarchical distributed architecture, typically including a master station system located in the dispatch center and substation systems deployed in various substations. The master station system decomposes the optimization scheme into specific control commands, such as "activate capacitor bank #1 in substation XX" and "adjust the tap of main transformer #2 in substation YY to level 3." These commands are sent to the corresponding substation systems through a secure and encrypted communication network. After receiving the commands, the substation systems execute specific circuit breaker operations or tap adjustments using local intelligent electronic devices or programmable logic controllers, and feed back the execution results and new equipment status to the master station. Real-time monitoring of voltage changes is achieved through voltage transformers and data acquisition units distributed throughout the power grid. Monitoring data is continuously uploaded to the distributed control system at high frequency. A dynamic adjustment mechanism for scheme parameters is activated. The distributed control system continuously compares the actual voltage curve after the implementation of optimization measures with the expected target. If the deviation persists or increases, the system will initiate parameter adjustment logic. For example, if the initial scheme recommends adding 6000 kvar capacitors, but monitoring reveals that the voltage improvement effect is not as expected and the demand level is still above the threshold, the system may automatically calculate and update the new recommended capacity into the optimization scheme, forming a new control command loop. This closed-loop control method allows resource allocation to adapt to dynamic changes in the power grid state.

[0094] To illustrate with a concrete example, suppose a power grid in a certain region is assessed and its resource demand level is determined to be Level 4 by a fuzzy inference system, while the preset threshold value is Level 3. Level 4 exceeds the threshold, and the resource optimization engine immediately activates. The engine's data analysis reveals that voltage problems are mainly concentrated in a power supply area in the northwest driven by heavy industrial loads, exhibiting significant downward bias clusters and a large number of associated groups. The power grid resource optimization scheme generated by the resource optimization engine includes three core instructions: First, to install a parallel capacitor bank with a capacity of 8000 kvar on the 10 kV III busbar of the Chaoyang 220 kV substation in the northwest to compensate for the reactive power deficit in this area. Second, to adjust the tap of the #2 main transformer of the Longyuan 110 kV substation connecting this area from the current 5th tap to the 7th tap, aiming to increase the voltage level by approximately 2.5%. Third, during the peak load period from 18:00 to 21:00, load switching was performed through the dispatch automation system, transferring some non-productive loads from the Xishan Coal Mine from its original power supply lines to the less heavily loaded southeastern power grid. This complete power grid resource optimization plan was sent to the master station of the distributed control system. After verifying the rationality of the instructions, the master station of the distributed control system distributed them to the substations of Chaoyang Substation, Longyuan Substation, and load control terminals. The substation systems executed the operations, activating capacitors and adjusting transformer taps, and the load control terminals completed the load switching. After the operation was completed, the distributed control system began to monitor the voltage changes of key nodes in the northwest in real time. Monitoring revealed that the overall voltage had increased, but the voltage of one node was still near the lower limit of the target range. The dynamic adjustment mechanism was triggered, and the system calculated and generated a supplementary parameter adjustment suggestion: in the next optimization cycle, the target tap position of the #2 main transformer of Longyuan Substation should be slightly adjusted from tap 7 to tap 8. This new suggestion was added to the queue of instructions to be executed, thus realizing the dynamic and refined adjustment of the power grid resource optimization plan.

[0095] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0096] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A power grid resource planning method based on voltage-deficient areas, characterized in that, The method includes the following steps: Acquire the initial voltage data set in the power grid operation status, and perform integrity verification and format standardization on the initial voltage data set to generate the target voltage data set; The voltage stability of the target voltage dataset is evaluated, and a voltage weakness indicator is generated based on the evaluation results. When the voltage weakness indicator indicates a potential weakness, the target voltage dataset is split into multiple voltage analysis units, each voltage analysis unit is evaluated independently, and the voltage analysis units are filtered based on the independent evaluation results. Voltage index values ​​are extracted from the filtered voltage analysis unit, and the deviation of each voltage index value is calculated with the corresponding reference voltage index value. Based on the deviation calculation results, the voltage index values ​​are grouped to form an index sequence. The voltage index values ​​in the index sequence are sorted in ascending order, and consecutive voltage index values ​​after sorting are taken as candidate association groups. The association detection algorithm is used to check the association within the candidate association groups, and the number of association groups is summarized. Assess the level of power grid resource demand by combining the number of associated groups and the voltage index values ​​not included in the association; Based on the power grid resource demand level, a power grid resource optimization scheme is generated.

2. The power grid resource planning method based on voltage-weak areas according to claim 1, characterized in that, When acquiring the initial voltage data set in the power grid operating state, and performing integrity verification and format standardization on the initial voltage data set to generate the target voltage data set, the process includes: Multi-resolution wavelet decomposition is performed on the initial voltage data set to extract high-frequency noise components and low-frequency signal components. After filtering out the high-frequency noise components, the voltage signal is reconstructed. The reconstructed voltage signal is time-aligned by converting heterogeneous sampled data into a sequence with a uniform time step through linear interpolation. Calculate the local smoothness index for each voltage sampling point in a sequence with a uniform time step, and evaluate the smoothness based on the voltage gradient change within the sliding window; When the local smoothness index exceeds the preset smoothness threshold, the voltage sampling point is classified as a valid point; otherwise, it is considered an invalid point and excluded. Aggregate all valid points to generate the target voltage data set.

3. The power grid resource planning method based on voltage-weak areas according to claim 1, characterized in that, When evaluating the voltage stability of the target voltage dataset and generating a voltage weakness indicator based on the evaluation results, the process includes: The target voltage data set is segmented into time series, dividing the continuous voltage data into multiple time windows of fixed duration; For each time window, the statistical characteristics of the voltage values ​​within the window are calculated, including the mean, variance, skewness, and kurtosis, forming a window feature vector; The window feature vectors of all time windows are aggregated, and the dimensionality is reduced by principal component analysis to obtain the compressed overall feature vector. Calculate the cosine similarity between the overall feature vector and the pre-stored standard voltage feature vector to obtain a similarity score; A voltage stability indicator is generated when the similarity score is higher than or equal to the stability threshold. When the similarity score is lower than or equal to the abnormal threshold, a voltage abnormality indicator is generated; When the similarity score is between the stable threshold and the abnormal threshold, a potential weak voltage indicator is generated.

4. The power grid resource planning method based on voltage-weak areas according to claim 3, characterized in that, Each voltage analysis unit is evaluated independently, and the voltage analysis units are filtered based on the independent evaluation results, including: For each voltage analysis unit, the voltage fluctuation variance is calculated. When the voltage fluctuation variance is lower than a preset variance threshold, the voltage analysis unit is determined to be a stable unit and removed. For the remaining voltage analysis units, calculate the voltage deviation coefficient for each unit and sort them according to the size of the deviation coefficient; Voltage analysis units with deviation coefficients higher than the preset coefficient are retained, and their voltage data are used as voltage index values.

5. The power grid resource planning method based on voltage-weak areas according to claim 4, characterized in that, When extracting voltage index values ​​from the filtered voltage analysis unit, calculating the deviation between each voltage index value and its corresponding reference voltage index value, and grouping the voltage index values ​​based on the deviation calculation results to form an index sequence, the process includes: Calculate the percentage of relative deviation between each voltage index value and the reference voltage index value to obtain the deviation value; Based on the sign of the deviation values, the set of positive deviation values ​​is divided into the upper deviation group, and the set of negative deviation values ​​is divided into the lower deviation group. In the upper deviation group, continuous deviation values ​​belonging to the same voltage analysis unit are clustered into upper deviation clusters, and isolated deviation values ​​are classified into upper deviation points. In the lower deviation group, continuous deviation values ​​belonging to the same voltage analysis unit are clustered into lower deviation clusters, and isolated deviation values ​​are classified into lower deviation points.

6. The power grid resource planning method based on voltage-weak areas according to claim 5, characterized in that, When the voltage index values ​​in the index sequence are sorted in ascending order, and consecutive voltage index values ​​after the sorting are selected as candidate correlation groups, they include: The deviation values ​​in the upper deviation cluster are sorted from smallest to largest, and adjacent deviation values ​​after sorting are paired as upper deviation candidate pairs. The deviation values ​​in the lower deviation cluster are sorted from smallest to largest, and adjacent deviation values ​​after sorting are paired as candidate pairs of lower deviations. For the upper and lower deviation points, no pairing operation is performed; they are directly retained as independent values.

7. The power grid resource planning method based on voltage-weak areas according to claim 6, characterized in that, When using association detection algorithms to examine the association within candidate association groups, the following are included: Collect historical voltage index values, reference voltage index values, and historical deviation values ​​during power grid operation, construct a training dataset, and randomly split it into a model training set and a model validation set; A long short-term memory neural network model was selected, and the model was trained using the model training set. The model accuracy was evaluated using the model validation set. If the loss function value of the current training cycle does not decrease during training, the model learning rate is adjusted and training continues until the maximum training cycle is reached. When the loss function value decreases steadily, training is stopped, and the trained long short-term memory neural network model is used as the association detection algorithm. Input the upper bias candidate pair and the lower bias candidate pair into the association detection algorithm, and output the association strength score; When the association strength score is higher than the preset association threshold, the candidate is determined to be valid for association, and the number of association groups is recorded.

8. The power grid resource planning method based on voltage-weak areas according to claim 7, characterized in that, When assessing the grid resource demand level by combining the number of associated groups and voltage index values ​​not included in the association, the following should be included: Calculate the deviation range width of the upper deviation cluster, that is, the difference between the maximum deviation value and the minimum deviation value; Calculate the deviation range width of the lower deviation cluster, i.e., the difference between the maximum deviation value and the minimum deviation value; The sum of the deviations at the upper deviation point and the sum of the deviations at the lower deviation point; The number of associated groups, the deviation range width of the upper deviation cluster, the deviation range width of the lower deviation cluster, the sum of the deviation values ​​of the upper deviation points, and the sum of the deviation values ​​of the lower deviation points are input into the fuzzy inference system, and the power grid resource demand level is output. The fuzzy rules are dynamically configured according to the power grid operating load.

9. The power grid resource planning method based on voltage-weak areas according to claim 8, characterized in that, When generating a power grid resource optimization scheme based on the aforementioned power grid resource demand level, the following are included: Set threshold values ​​for the power grid resource demand level and compare the power grid resource demand level with the threshold values; When the power grid resource demand level exceeds the threshold, the resource optimization engine is activated to generate optimization solutions, including increasing capacitor compensation, adjusting transformer taps, and redistributing loads. When the power grid resource demand level is lower than or equal to the threshold, the current resource allocation strategy is maintained, and only a monitoring report is generated.

10. The power grid resource planning method based on voltage-weak areas according to claim 1, characterized in that, After generating the power grid resource optimization scheme, the process also includes: deploying the power grid resource optimization scheme to the distributed control system, monitoring voltage changes in real time, and dynamically adjusting the scheme parameters.

Images