Multi-time scale reactive power and voltage situation awareness method for new energy rich power grid

Abstract

This invention relates to the field of voltage situation awareness technology for renewable energy-rich power grids, and discloses a multi-timescale reactive power voltage situation awareness method for renewable energy-rich power grids. The method includes: collecting second-level real-time voltage data, minute-level renewable energy output data, and hourly load forecast data; extracting dynamic coupling features to determine a timeliness coordination benchmark; hierarchically integrating data to generate a preliminary estimation result without cumulative bias; using multi-scale Kalman filtering iterative optimization to generate a time-consistent voltage state trajectory; extracting short-term power balance and medium-to-long-term safety margin indices, adjusting the time coordination parameters of the time-series fusion algorithm accordingly, updating the filtering iteration process, and outputting the corrected reactive power voltage state estimation result. This invention can organically coordinate the timeliness differences of multi-scale data, suppress bias propagation and amplification, achieve a coherent description of the voltage state trajectory in the time dimension, and improve the accuracy and reliability of reactive power voltage state estimation.

Description

Technical Field

[0001] This invention relates to the field of voltage situation awareness technology for renewable energy-rich power grids, and in particular to a multi-timescale reactive voltage situation awareness method for renewable energy-rich power grids. Background Technology

[0002] In renewable energy-rich power grids, the large-scale integration of intermittent power sources such as wind and solar power has made the reactive power and voltage characteristics of the grid increasingly complex, leading to frequent voltage fluctuations and significantly increased difficulty in regulation. Reactive power and voltage state estimation, as a core component of power grid situational awareness, directly affects the effectiveness of subsequent voltage control decisions and has become a key technology for ensuring the safe and efficient operation of renewable energy-rich power grids.

[0003] Currently, reactive power voltage state estimation in renewable energy-rich power grids faces a major technical challenge: the inherent contradictions in the time scales of multi-source heterogeneous data. In actual power grid operation, data acquisition systems acquire real-time voltage data at a second-level frequency to reflect instantaneous fluctuations in the grid; renewable energy power forecasting systems update output data at a minute-level cycle to characterize short-term power change trends; and load forecasting systems provide load forecast data at an hour-level cycle to depict the medium- to long-term system state evolution. These data differ significantly in update frequency, timeliness, and physical meaning. Existing state estimation methods often simply fuse or process these asynchronous data independently, failing to effectively construct a coordination mechanism across time scales. This results in fragmented estimation results in the time dimension, failing to accurately depict the overall voltage dynamic behavior from instantaneous transients to medium- to long-term evolution, severely impacting the accurate judgment of voltage safety margins and the reliability of control decisions.

[0004] To address the aforementioned issues, some existing research has attempted to preprocess multi-source data through methods such as data interpolation, time alignment, or weighted fusion to mitigate the impact of data asynchrony. However, these methods typically only achieve data-level unification and fail to deeply consider the dynamic coupling characteristics of physical processes at different time scales. More critically, due to the lack of dynamic coordination regarding data timeliness, time-accumulated biases are prone to occur during state estimation. For example, short-term reactive power balance estimation based on minute-level data, if failing to incorporate second-level voltage transient information in a timely manner, will deviate from the actual rapidly fluctuating voltage trajectory. This bias accumulates over time, further interfering with the judgment of medium- and long-term voltage safety margins, creating a vicious cycle of bias propagation and amplification. Therefore, fundamentally constructing a mechanism that can organically coordinate data at multiple time scales, such as second-level, minute-level, and hour-level data, and ensure consistency and coherence of estimation results across different scales over time, is crucial for improving the accuracy of voltage state perception in renewable energy-rich power grids. Summary of the Invention

[0005] Therefore, the technical problem to be solved by this invention is to overcome the inherent contradictions in the existing technology due to the different time scales of multi-source heterogeneous data, such as second-level, minute-level, and hour-level, which leads to the fragmentation of state estimation results in the time dimension, the accumulation and propagation of deviations, and the inability to accurately characterize the overall voltage dynamic behavior. The invention provides a multi-time-scale reactive voltage state perception method for renewable energy-rich power grids. Through the synergistic mechanism of time-series fusion, hierarchical integration, multi-scale Kalman filtering, and feedback closed loop, it can organically coordinate the time-sensitivity differences of multi-scale data, suppress the propagation and amplification of deviations, and achieve a consistent and coherent description of the voltage state trajectory from instantaneous transient to medium- and long-term evolution in the time dimension, thereby improving the accuracy and reliability of reactive voltage state estimation.

[0006] To address the aforementioned technical problems, this invention provides a multi-timescale reactive voltage situational awareness method for renewable energy-rich power grids, comprising the following steps: The system collects real-time voltage data at the second level, renewable energy output data at the minute level, and load forecast data at the hour level from the power grid. It then uses a time-series fusion algorithm to align the time scales of the multi-source heterogeneous data and generate a unified time series. Based on a unified time series, dynamic coupling characteristics between instantaneous fluctuations and slow changes are extracted to determine the timeliness coordination benchmark for multi-scale data. Based on the timeliness coordination benchmark, the instantaneous information of second-level data, the short-term trend of minute-level data, and the medium- and long-term changes of hour-level data are integrated in a hierarchical manner to generate preliminary estimation results without cumulative bias. The preliminary estimation results are iteratively optimized using multi-scale Kalman filtering to suppress noise and bias propagation during dynamic coupling and generate a voltage state trajectory with time consistency. Short-term power balance indices and medium- to long-term safety margin indices are extracted from the voltage state trajectory to determine a coherent description of the overall voltage dynamic behavior. Based on a coherent description of the overall voltage dynamic behavior, the time coordination parameters of the time-series fusion algorithm are adjusted, and the iterative process of the multi-scale Kalman filter is updated using the adjusted parameters to output the corrected reactive voltage state estimation results.

[0007] In one embodiment of the present invention, a time-scale alignment of multi-source heterogeneous data is performed using a time-series fusion algorithm to generate a unified time series, including: The second-level real-time voltage data, minute-level renewable energy output data, and hourly load forecast data are stored with their respective original timestamps. The time point of the second-level data is used as the reference time axis. For the minute-level renewable energy output data, linear interpolation is used to calculate the output value corresponding to each second. For the hourly load forecast data, cubic spline interpolation is used to calculate the load forecast value corresponding to each second. The interpolated data is aligned with the second-level real-time voltage data according to the time point to form a unified time series with a time interval of seconds.

[0008] In one embodiment of the present invention, extracting the dynamic coupling characteristics between instantaneous fluctuations and slow changes to determine the timeliness coordination benchmark for multi-scale data includes: Calculate the root mean square value within a sliding window for second-level data in a unified time series to obtain the instantaneous voltage fluctuation amplitude; Low-pass filtering is applied to minute-level and hour-level data in a unified time series to obtain slowly changing trend components. Calculate the cross-correlation function between the instantaneous fluctuation amplitude and the slow trend component. Determine the impact delay time of the instantaneous fluctuation on the slow trend based on the peak position of the cross-correlation function. Use the impact delay time as the benchmark for timeliness coordination of multi-scale data.

[0009] In one embodiment of the present invention, based on a timeliness coordination benchmark, the instantaneous information of second-level data, the short-term trend of minute-level data, and the medium- and long-term changes of hour-level data are hierarchically integrated to generate a preliminary estimation result without cumulative bias, including: The time window length for tiered integration is determined based on the timeliness coordination benchmark; Within the time window, the instantaneous information of the second-level data is used as the first-level estimation input, and the voltage instantaneous change curve is fitted by the recursive least squares method. Using the short-term trend of minute-level data as the input for the second-level estimation, the trend features are extracted using the weighted moving average method. Using the medium- to long-term changes in hourly data as the third-level estimation input, the exponential smoothing method is used to extract the change patterns. The first-level estimation results are weighted and fused with the second-level estimation results, and then the fused results are superimposed with the third-level estimation results to obtain a preliminary estimation result without cumulative bias.

[0010] In one embodiment of the present invention, the preliminary estimation results are iteratively optimized using multi-scale Kalman filtering to suppress noise and bias propagation during dynamic coupling, generating a voltage state trajectory with time consistency, including: Using the preliminary estimation results as the observation input, a multi-scale state-space model containing second-level, minute-level, and hour-level state variables was established. The state transition matrix of the second-level state variables was set as a diagonal matrix according to the voltage fluctuation characteristics, the state transition matrix of the minute-level state variables was set according to the rate of change of new energy output, and the state transition matrix of the hour-level state variables was set according to the load change pattern. The process noise covariance matrix and observation noise covariance matrix for the corresponding state variables at the second, minute, and hour levels are set respectively. The state prediction and state update are performed sequentially using the Kalman filter recursive formula. In each iteration, the state prediction value for the next moment is corrected based on the state update result. The iteration continues until convergence, and the converged state estimate is used as the voltage state trajectory with time consistency.

[0011] In one embodiment of the present invention, in the multi-scale state space model, the state variables at the second, minute, and hour levels are associated through a coupling matrix. The coupling matrix is ​​determined based on the dynamic coupling characteristics, and the non-zero elements in the coupling matrix represent the influence weight of a state variable at one time scale on a state variable at another time scale.

[0012] In one embodiment of the present invention, short-term power balance indices and medium-to-long-term safety margin indices are extracted from the voltage state trajectory to determine a coherent description of the overall voltage dynamic behavior, including: Extract recent trajectory segments of a preset duration from the voltage state trajectory, calculate the average value and standard deviation of the voltage amplitude, and use the ratio of the average value to the preset voltage reference value as a short-term power balance index. Extract a long-term trajectory segment of a preset duration from the voltage state trajectory, calculate the minimum and maximum values ​​of the voltage amplitude, take the first difference between the minimum value and the preset low voltage threshold as the first safety margin index, and take the second difference between the maximum value and the preset high voltage threshold as the second safety margin index. By combining the short-term power balance index, the first safety margin index, and the second safety margin index in chronological order, a coherent description of the overall voltage dynamic behavior is formed.

[0013] In one embodiment of the present invention, adjusting the time coordination parameters of the timing fusion algorithm based on a coherent description of the overall voltage dynamic behavior includes: The coherent description of the overall voltage dynamic behavior is compared with the preset ideal voltage dynamic curve, and the deviation value at each time point is calculated. When the deviation value exceeds the preset deviation threshold, the time position of the deviation is determined. According to the data time scale corresponding to the time position, the interpolation weight or smoothing coefficient of the corresponding data type in the time series fusion algorithm is increased or decreased, and the adjusted interpolation weight or smoothing coefficient is used as the updated time coordination parameter.

[0014] In one embodiment of the present invention, the iterative process of the multi-scale Kalman filter is updated using the adjusted parameters, and the corrected reactive voltage state estimation result is output, including: The adjusted time coordination parameters are substituted into the time series fusion algorithm to regenerate a unified time series. The observation input in the multi-scale Kalman filter is updated based on the regenerated unified time series. The state prediction and state update process of the multi-scale Kalman filter is re-executed with the updated observation input. The converged state estimate after re-execution is output as the corrected reactive voltage state estimate result.

[0015] In one embodiment of the present invention, the timing fusion algorithm and the iterative process of the multi-scale Kalman filter form a closed-loop feedback. When the deviation between the coherent description of the overall voltage dynamic behavior and the preset ideal voltage dynamic curve continues to increase, the operation of adjusting the time coordination parameters of the timing fusion algorithm and updating the iterative process of the multi-scale Kalman filter is executed cyclically until the deviation is less than the preset convergence threshold.

[0016] The technical solution of the present invention has the following advantages compared with the prior art: This invention discloses a multi-timescale reactive voltage situational awareness method for a renewable energy-rich power grid. It collects second-level real-time voltage data, minute-level renewable energy output data, and hourly load forecast data from the power grid. A time-series fusion algorithm is used to align the time scales of the multi-source heterogeneous data, generating a unified time series. Based on the unified time series, dynamic coupling characteristics between instantaneous fluctuations and slow changes are extracted to determine the timeliness coordination benchmark for the multi-scale data. Based on the timeliness coordination benchmark, the instantaneous information of the second-level data, the short-term trend of the minute-level data, and the medium- and long-term changes of the hourly data are hierarchically integrated to generate preliminary estimation results without cumulative bias. The preliminary estimation results are iteratively optimized using multi-scale Kalman filtering to suppress noise and bias transmission during the dynamic coupling process. This method generates a voltage state trajectory with temporal consistency. Short-term power balance indices and medium-to-long-term safety margin indices are extracted from the voltage state trajectory to determine a coherent description of the overall voltage dynamic behavior. Based on this coherent description, the time coordination parameters of the time-series fusion algorithm are adjusted. The adjusted parameters are used to update the iterative process of the multi-scale Kalman filter, outputting the corrected reactive power voltage state estimation results. This method organically coordinates data from multiple time scales, achieving consistency and coherence of estimation results across different scales in terms of time evolution. It addresses the problem of inaccurate state estimation and accumulated bias caused by time scale contradictions in multi-source heterogeneous data in renewable energy-rich power grids. This method has the advantages of improving the accuracy and consistency of voltage state estimation, effectively avoiding time accumulation bias, and enhancing the safe operation level of the power grid. Attached Figure Description

[0017] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart of the multi-time-scale reactive voltage situational awareness method for the new energy-rich power grid of the present invention. Figure 2 This is a flowchart of the steps involved in generating a unified time series according to the present invention; Figure 3 This is a flowchart of the steps for determining the timeliness coordination benchmark in this invention; Figure 4 This is a flowchart of the steps involved in the layered integration of this invention; Figure 5 This is a flowchart of the iterative optimization steps of the present invention; Figure 6 This is a flowchart illustrating the steps of the present invention to determine the overall voltage dynamic behavior. Figure 7 This is a flowchart illustrating the steps of adjusting the time coordination parameters of the timing fusion algorithm in this invention. Figure 8 This is a flowchart of the steps for outputting the corrected reactive voltage state estimation result of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0019] Reference Figure 1 As shown, this invention proposes a multi-timescale reactive power and voltage situational awareness method for renewable energy-rich power grids. First, it collects second-level real-time voltage data, minute-level renewable energy output data, and hourly load forecast data. These data are typically generated independently by different monitoring or forecasting systems, each with its own sampling frequency and timestamp. For subsequent processing, a time-series fusion algorithm is used to align the time scales of these data to generate a unified time series. The core function of this step is to establish a unified time reference from the data entry point, laying the foundation for subsequent cross-scale collaborative processing and avoiding direct fusion chaos caused by inconsistent data formats and timestamps.

[0020] Based on this, dynamic coupling characteristics between instantaneous fluctuations and slow changes are extracted according to a unified time series, and a timeliness coordination benchmark for multi-scale data is determined. The essence of this step is to identify the dynamic behavior patterns of the power grid at different time scales, clarify the intrinsic relationship between second-level transient processes, minute-level power changes and hour-level load evolution, thereby providing a basis for hierarchical processing, rather than simply treating data at each scale as independent information sources.

[0021] Subsequently, based on the established timeliness coordination benchmark, the instantaneous information of second-level data, the short-term trends of minute-level data, and the medium- and long-term changes of hourly-level data are hierarchically integrated to generate preliminary estimation results without cumulative bias. This hierarchical integration step abandons the traditional method of simply fusing multi-source data, but instead processes them separately according to their respective timescale characteristics, and then organically fuses them under the guidance of the coordination benchmark, thus fundamentally preventing the introduction of initial bias caused by timescale mismatch.

[0022] To further improve estimation accuracy and suppress noise and bias propagation during dynamic coupling, multi-scale Kalman filtering is used to iteratively optimize the preliminary estimation results, generating a voltage state trajectory with time consistency. Multi-scale Kalman filtering can adaptively adjust filtering parameters according to the dynamic characteristics at different time scales, effectively filtering out various measurement noises while preventing estimation bias at one time scale from spreading to other scales, thus ensuring the smoothness and consistency of the voltage state trajectory over time.

[0023] Finally, short-term power balance indices and medium-to-long-term safety margin indices are extracted from the voltage state trajectory to determine a coherent description of the overall voltage dynamic behavior. Based on this coherent description, the time coordination parameters of the time-series fusion algorithm are adjusted. The iterative process of the multi-scale Kalman filter is updated using the adjusted parameters, and the corrected reactive power voltage state estimation result is output. This feedback closed-loop mechanism transforms the entire situational awareness process from a one-time open-loop calculation into a closed-loop system that dynamically optimizes the front-end fusion and back-end filtering parameters based on the estimation results. When a deviation amplification trend appears at a certain time scale, the system can promptly adjust the time coordination parameters through index feedback, correct the data fusion method from the source, and synchronously update the filtering process, thereby effectively suppressing the accumulated deviation.

[0024] Through the organic combination of the above technical solutions, the beneficial effects of this invention include: First, by leveraging the synergistic effect of time-series fusion algorithms and hierarchical integration, the inherent contradictions in the time scales of multi-source heterogeneous data are fundamentally resolved, enabling second-level, minute-level, and hourly-level data to perform their respective functions and work collaboratively within a unified coordination framework. Second, the introduction of multi-scale Kalman filtering and feedback closed-loop mechanisms enables the method to suppress deviation propagation and adaptively correct, avoiding the accumulation and amplification of estimation deviations caused by asynchronous data update rhythms in traditional methods. Third, by extracting short-term power balance and medium-to-long-term safety margin indicators from the voltage state trajectory and using these as feedback bases, the final output reactive power voltage state estimation results not only remain accurate during local transient processes but also exhibit consistency and coherence in the overall time evolution, thus providing a reliable state awareness basis for subsequent voltage control decisions. In summary, this invention significantly improves the accuracy and robustness of reactive power voltage state awareness in renewable energy-rich power grids through multi-scale coordination and closed-loop optimization at the algorithm level, without introducing additional hardware equipment.

[0025] When aligning multi-source heterogeneous data in a renewable energy-rich power grid across time scales, failure to employ refined interpolation and alignment strategies may result in the loss of dynamic information between data at different time scales, the introduction of unnecessary noise or bias, and consequently affect the accuracy and consistency of subsequent situational awareness.

[0026] In this regard, refer to Figure 2 As shown, this application further proposes a step to align the time scales of multi-source heterogeneous data using a time-series fusion algorithm to generate a unified time series. Specifically, this includes storing second-level real-time voltage data, minute-level renewable energy output data, and hourly load forecast data with their respective original timestamps, which is a fundamental step in data preprocessing. This approach aims to ensure the integrity and traceability of the original data. By retaining the original timestamp of each data point, its position on the timeline can be accurately identified, providing an accurate reference for subsequent time-scale alignment operations and effectively avoiding confusion or loss of time information due to differences in data sources.

[0027] Building upon this, using second-level data points as the baseline time axis is crucial for achieving multi-scale data alignment. Second-level data is chosen as the baseline because it typically has the highest sampling frequency, capturing the fastest dynamic changes in the power grid. Using it as the baseline ensures that high-frequency information is not lost during alignment and provides a fine-grained reference frame for other low-frequency data. This usually means creating a time-series index with second-level intervals, to which all other data will be mapped.

[0028] For minute-level renewable energy output data, linear interpolation is used to calculate the output value for each second. Linear interpolation is a simple and effective interpolation method, suitable for scenarios where data changes are relatively stable and the trend is linear. Since minute-level renewable energy output data usually does not change drastically in a short period of time, linear interpolation can better estimate the output value per second within a minute interval, maintaining the smoothness of the data and the continuity of the trend, while avoiding introducing excessive computational complexity. For example, let the timestamps of two adjacent minute-level data points be t1 and t2, and the corresponding output values ​​be P1 and P2, respectively. For any second-level time point t between t1 and t2, its corresponding output value P is calculated according to the following formula: P = P1 + (P2 - P1) × (t - t1) / (t2 - t1). When there are missing values ​​in the minute-level data, the missing data points are skipped, and interpolation is performed using the valid data points before and after them; when bidirectional interpolation cannot be performed at the data boundaries, the nearest neighbor data points are used for extrapolation and filling.

[0029] For hourly load forecast data, cubic spline interpolation is used to calculate the load forecast value for each second. Cubic spline interpolation is a more advanced interpolation method that generates smoother curves that better reflect the actual physical process. Cubic spline interpolation constructs a piecewise cubic polynomial function, ensuring the interpolation curve has continuous first and second derivatives at each data point. Natural boundary conditions are used, setting the second derivative to zero at the first and last data points. Let the timestamps of the hourly data points be T0, T1, …, Tn, and the corresponding load forecast values ​​be L0, L1, …, Ln. At each second between adjacent data points, the coefficients of the cubic polynomial within that interval are obtained by solving a tridiagonal system of equations, thus calculating the load forecast value per second. Hourly load forecast data typically represents the macroscopic trend of the power grid load, and its changes may contain nonlinear characteristics. Using cubic spline interpolation can better capture this nonlinear change, making the interpolated data have second-order continuity at the connection points, avoiding the sharp transitions that linear interpolation may bring, and thus more accurately reflecting the actual change pattern of the load.

[0030] Finally, the interpolated minute-level output data and interpolated hour-level load forecast data are merged with the original second-level real-time voltage data at second-level time points. Using the second-level timestamp as the primary key, the three are associated according to the same timestamp, forming a unified time series record containing three fields: voltage value, renewable energy output value, and load forecast value. For any second-level time point, if a certain type of data cannot generate a valid value due to interpolation boundary issues, the field is marked as empty, and data from the previous valid time point is used to fill it in subsequent processing to ensure the continuity of the unified time series.

[0031] In reactive power and voltage situational awareness methods for renewable energy-rich power grids, a key step is to extract the dynamic coupling characteristics between instantaneous fluctuations and slow changes based on a unified time series, and to determine the timeliness coordination benchmark for multi-scale data. However, the lack of clear and quantitative methods to identify and correlate voltage behaviors at different time scales may lead to insufficient understanding of the interaction between instantaneous events and long-term trends, thereby affecting the accuracy of the timeliness coordination benchmark. This could result in subsequent data integration and state estimation failing to fully reflect the true dynamics of the power grid.

[0032] In this regard, refer to Figure 3 As shown, this application further proposes specific steps for extracting the dynamic coupling characteristics between instantaneous fluctuations and slow changes, and determining the timeliness coordination benchmark for multi-scale data. These steps include: calculating the root mean square (RMS) value within a sliding window for second-level data in a unified time series, aiming to accurately quantify the rapid changes in grid voltage over extremely short time scales. Second-level data reflects the most real-time operating state of the grid, and the voltage fluctuations it contains are often caused by instantaneous events (e.g., sudden load changes, rapid changes in renewable energy output, etc.). The root mean square (RMS), as a commonly used statistical measure, can effectively characterize the effective value or average power of a signal. For fluctuating data, its RMS value can well reflect its fluctuation amplitude. In practice, a suitable sliding window length can be set, for example: setting the sliding window length to 5 seconds and the step size to 1 second, i.e., calculating the RMS value of the second-level voltage data within the previous 5 seconds by moving 1 second each time. The determination of the 5-second window length is based on the fact that the time scale of typical instantaneous voltage fluctuations in renewable energy-rich grids is usually between 2 and 5 seconds. This window length can effectively capture the complete instantaneous fluctuation process while suppressing high-frequency noise interference. The root mean square value is calculated using the following formula: ; Where N is the number of sampling points within the sliding window, Vi is the i-th second-level voltage sample value within the window, and RMS(t) represents the instantaneous voltage fluctuation amplitude at time t. The sliding window moves forward with a preset step size, continuously calculating and updating the root mean square value, thus obtaining a time-varying sequence of instantaneous voltage fluctuation amplitudes. The selection of the sliding window length should comprehensively consider both the sensitivity to instantaneous fluctuations and the ability to suppress noise.

[0033] Simultaneously, low-pass filtering was applied to minute-level and hourly-level data within the unified time series to separate the main, smooth trends and effectively remove potentially high-frequency noise or transient disturbances. Minute-level data typically reflects short-term trends in renewable energy output, while hourly-level data reflects medium- to long-term load forecasting patterns. These data exhibit relatively slow evolution characteristics during grid operation. Low-pass filtering is a signal processing technique that allows signals below a certain cutoff frequency to pass through while attenuating signals above that frequency. Specifically, for minute-level data, a second-order Butterworth low-pass filter was used, with a cutoff frequency set at 0.01Hz. This cutoff frequency was determined based on the fact that the variation period of minute-level renewable energy output data is approximately 1 to 5 minutes, corresponding to a frequency range of 0.0033Hz to 0.0167Hz. A cutoff frequency of 0.01Hz effectively preserves output trend information while filtering out second-level high-frequency disturbances. For hourly data, a second-order Butterworth low-pass filter is used with a cutoff frequency set at 0.001Hz. This cutoff frequency is determined based on the fact that the variation period of hourly load forecast data is approximately 1 to 24 hours, corresponding to a frequency range of 0.0000116Hz to 0.00028Hz. A cutoff frequency of 0.001Hz can fully preserve the daily variation pattern of the load while filtering out noise components at the minute and higher frequencies. By applying a low-pass filter to minute-level and hourly data, high-frequency components can be effectively filtered out, thereby extracting the slow-changing trend components that represent the core evolution pattern.

[0034] Based on this, the cross-correlation function between the instantaneous fluctuation amplitude and the slowly changing trend component is calculated. The peak position of the cross-correlation function determines the delay time of the instantaneous fluctuation's impact on the slowly changing trend, and this delay time is ultimately used as the benchmark for the timeliness coordination of multi-scale data. This step aims to accurately quantify the dynamic correlation between data at different time scales, especially the lag in the impact of instantaneous fluctuations on slowly changing trends. In renewable energy-rich power grids, instantaneous events (such as sudden changes in renewable energy output) may initially cause instantaneous voltage fluctuations, which then gradually affect voltage trends over longer time scales through the grid's dynamic response. The cross-correlation function is a statistical tool that measures the similarity of two signals at different time delays. By calculating the cross-correlation function between the voltage instantaneous fluctuation amplitude sequence and the slowly changing trend component sequence, the time delay of maximum correlation between them can be identified. The peak position of the cross-correlation function indicates the delay time required for the instantaneous fluctuation to have the maximum impact on the slowly changing trend. Specifically, let the instantaneous fluctuation amplitude sequence be R(t), and the slowly changing trend component sequence be S(t), both being time series with intervals of seconds and a sequence length of T. The cross-correlation function R... RS (τ) is calculated using the following formula: ; Where τ is the time delay, ranging from -60 seconds to 60 seconds. This range is determined based on the fact that the delay of instantaneous fluctuations on slowly changing trends in a renewable energy-rich power grid typically does not exceed 1 minute; μ R and μ S Let σ be the mean of R(t) and S(t), respectively. R and σ S Let R(t) and S(t) be the standard deviations, respectively. After calculating the cross-correlation function at each delay τ, find the value of R that makes R... RS (τ) The delay τmax at which the maximum value is obtained is used as the delay time for the effect of instantaneous fluctuation on the slowly changing trend. If there are multiple peaks within the range of τ, the delay time corresponding to the peak with the largest absolute value is taken as τmax.

[0035] Reference Figure 4 As shown, based on the timeliness coordination benchmark, this application further proposes a method for hierarchically integrating instantaneous information of second-level data, short-term trends of minute-level data, and medium- and long-term changes of hour-level data to generate preliminary estimation results without cumulative bias. The method specifically includes the following steps: First, the time window length for hierarchical integration is determined based on the timeliness coordination benchmark. This step aims to provide a reasonable time range for subsequent hierarchical integration operations. The timeliness coordination benchmark reflects the inherent dynamic coupling between data at different time scales, such as the time delay of the impact of instantaneous fluctuations on slowly changing trends. Determining the time window length based on this benchmark ensures that the information contained in each layer of data is temporally correlated and practically meaningful during the integration process, thereby avoiding information loss or redundancy due to improper time window settings. For example, let the timeliness coordination benchmark be τ. max This refers to the time delay between the instantaneous fluctuation and the slowly changing trend, measured in seconds. The time window length L for hierarchical integration is set to τ. max Integer multiples thereof, specifically determined as follows: L = k × τ max τmax is a preset scaling factor, ranging from 3 to 10. This range is determined based on the following: to ensure the time window fully covers the entire process of the impact of instantaneous fluctuations on short-term trends, the window length should include at least 3 times the impact delay time; simultaneously, to avoid introducing excessive historical redundancy information due to an excessively long window, the window length should not exceed 10 times the impact delay time. Based on the actual calculated value of τmax, the smallest k value that makes L an integer number of seconds is selected, and the value of L is limited to between 10 seconds and 300 seconds.

[0036] Secondly, within a defined time window, instantaneous information from second-level data is used as the first-level estimation input, and the voltage instantaneous change curve is fitted using the recursive least squares method. Second-level data carries the fastest and most direct dynamic response information of the power grid voltage. Using it as the first-level estimation input aims to capture the instantaneous fluctuation characteristics of the voltage. The recursive least squares method is used to fit the voltage instantaneous change curve because this method can update model parameters in real time, adapt to rapid voltage changes, and effectively suppress the influence of random noise. Specifically, the recursive least squares method continuously receives new second-level voltage data and iteratively updates the model parameters of the voltage change curve, thereby accurately depicting the voltage's rise, fall, or maintenance trend within a very short time, providing a high-precision instantaneous change estimate for subsequent fusion.

[0037] Next, the short-term trend of minute-level data is used as the input for the second-level estimation, and a weighted moving average method is employed to extract trend features. Minute-level data reflects the continuous changes in grid voltage over a short timescale, such as the gradual voltage fluctuations caused by changes in renewable energy output. Using this as the second-level estimation input aims to capture the short-term voltage trend. The weighted moving average method effectively smooths out the instantaneous noise contained in second-level data, highlighting the main direction and magnitude of voltage changes on the minute scale. By assigning higher weights to recent data, the weighted moving average method can more sensitively reflect the latest short-term trend while also considering the influence of historical data, thus providing a stable and responsive short-term voltage trend estimate. The window length of the weighted moving average is set to three times the sampling period of the minute-level data. Since the sampling period for minute-level data is one minute, the window length is three minutes, corresponding to three minute-level data points. The weight allocation uses a linear weighting method, meaning the most recent data point has the highest weight, and the earliest data point has the lowest weight, with the weight values ​​increasing linearly according to the time order of the data points. This calculation result serves as the short-term trend feature extraction result for the current moment.

[0038] Subsequently, the medium- to long-term changes in hourly data were used as the input for the third-level estimation, and exponential smoothing was employed to extract the patterns of change. Hourly data represents the basic operating patterns and variation modes of grid voltage over a longer time scale, such as the daily load curve reflected by load forecast data or seasonal variations. Using it as the third-level estimation input aims to capture the medium- to long-term patterns of voltage change. The exponential smoothing method effectively filters out short-term fluctuations and noise, revealing the potential trends and periodic characteristics of voltage. Exponential smoothing uses a weighted average of historical data, with the weights decaying exponentially over time, making the model more sensitive to recent data while preserving the influence of long-term historical information, thus providing a stable and predictive baseline for medium- to long-term voltage changes. A quadratic exponential smoothing method is used to simultaneously capture both horizontal and slope trends. The smoothing constant is set between 0.1 and 0.3, with the following criteria: a smaller smoothing constant results in a stronger smoothing effect and more stable extraction of medium- to long-term trends; a larger smoothing constant results in a more sensitive response to recent changes. The initial values ​​were set as follows: the average value of the data points from the previous four hours was taken as the initial horizontal component, and the linear regression slope of the data points from the previous four hours was taken as the initial trend component. By iteratively updating point by point, the medium- to long-term variation patterns at the current moment were extracted.

[0039] Finally, the first-level and second-level estimation results are weighted and fused, and then the fused result is superimposed with the third-level estimation result to obtain a preliminary estimation result without cumulative bias. This step is crucial for achieving hierarchical integration of multi-timescale data. First, the second-level estimation results reflecting instantaneous fluctuations are weighted and fused with the minute-level estimation results reflecting short-term trends. This aims to effectively combine rapid dynamics with slower trends, forming a comprehensive estimate that includes both instantaneous response and short-term trends. The weights of the weighted fusion can be dynamically adjusted based on the confidence level of each level's estimation result or its importance to the overall voltage situation. Subsequently, this fused result is superimposed with the hour-level estimation results reflecting medium- and long-term variation patterns. This superposition method ensures that the final preliminary estimation result not only captures the instantaneous and short-term dynamics of the grid voltage but also reflects its long-term operating baseline and trends, thereby generating a comprehensive, accurate, and non-cumulative bias-free preliminary voltage estimate.

[0040] In the complex and dynamic environment of a power grid rich in new energy sources, the preliminary estimation results may still contain noise and biases caused by data acquisition, fusion and model simplification. If these noises and biases are not effectively suppressed, they will affect the accuracy and time consistency of the final voltage state estimation and make it difficult to fully reflect the real operating status of the power grid.

[0041] Reference Figure 5As shown, this application further proposes to iteratively optimize the preliminary estimation results using multi-scale Kalman filtering to suppress noise and bias propagation during the dynamic coupling process, thereby generating a voltage state trajectory with time consistency. Specifically, this method uses the aforementioned preliminary estimation results as observation input, providing initial, unprocessed observation data for the subsequent filtering process. Based on this, a multi-scale state-space model is established, including second-level, minute-level, and hour-level state variables. This model can hierarchically describe the dynamic behavior of the grid voltage at different time scales. The second-level state variables are used to capture instantaneous voltage fluctuations, the minute-level state variables are used to reflect short-term voltage trends caused by changes in new energy output, and the hour-level state variables are used to characterize medium- and long-term voltage changes brought about by load forecasting.

[0042] To accurately describe the evolution of state variables at different time scales, the state transition matrix for second-level state variables is set as a diagonal matrix based on voltage fluctuation characteristics. This diagonal matrix setting typically implies that within extremely short time scales, the mutual influence between different voltage components is small, or can be simplified into independent evolution processes, thus effectively capturing the instantaneous voltage fluctuation characteristics. The state transition matrix for minute-level state variables is set based on the rate of change of renewable energy output, because changes in renewable energy output are a significant factor affecting short-term voltage fluctuations in the power grid, and its rate of change directly reflects the dynamic response speed and amplitude of the voltage state. The state transition matrix for hourly-level state variables is set based on load change patterns to reflect the impact of periodic or trend changes in power grid load on the medium- and long-term stability of voltage.

[0043] Furthermore, to quantify the model's uncertainty and observation error, this method sets process noise covariance matrices and observation noise covariance matrices for state variables at the second, minute, and hour levels, respectively. The process noise covariance matrix describes the random perturbations of state variables during evolution, while the observation noise covariance matrix describes the error between the observed data and the true state. The appropriate setting of these covariance matrices is crucial to the performance of the Kalman filter, ensuring that the filtering process effectively models uncertainties at different time scales.

[0044] Based on the aforementioned model and parameter settings, a Kalman filter recursive formula is used for state prediction and state update sequentially. In the state prediction phase, the state estimate from the previous time step and the state transition matrix are used to predict the current state. In the state update phase, the state estimate is corrected by combining the observed input and predicted state at the current time step to reduce prediction error. In each iteration, the state prediction for the next time step is corrected based on the state update result. This iterative correction mechanism allows the filtering process to continuously approximate the true state. This iterative process continues until convergence, meaning the change in the state estimate in consecutive iterations is less than a preset threshold, indicating that the filtering result has stabilized. Finally, the converged state estimate is output as a time-consistent voltage state trajectory. This trajectory not only contains voltage state information at each time scale but also ensures logical consistency and physical rationality between different time scales, providing a reliable basis for subsequent power grid analysis and control.

[0045] Specifically, in multi-scale state-space models, the relationships between second-level, minute-level, and hourly-level state variables are established through a coupling matrix. This means that when constructing the state transition equations for a multi-scale Kalman filter, a matrix is ​​introduced to describe the interactions between state variables at different time scales. For example, rapid fluctuations in second-level voltage may be affected by minute-level changes in renewable energy output, while minute-level output changes may be correlated with hourly load forecast trends. By introducing the coupling matrix, these cross-scale influences can be explicitly incorporated into the state transition equations, allowing state changes at one time scale to affect state forecasts at another time scale, thus more comprehensively reflecting the dynamic behavior of the power grid. This coupling matrix can be an off-diagonal matrix, whose off-diagonal elements are used to quantify the mutual influences between state variables at different scales.

[0046] The coupling matrix is ​​determined based on dynamic coupling characteristics. Dynamic coupling characteristics refer to the patterns and strengths of mutual influence between data at different time scales. For example, in the method described above, the instantaneous voltage fluctuation amplitude can be obtained in advance by calculating the root mean square value within a sliding window for second-level data in a unified time series. Then, low-pass filtering is applied to minute-level and hour-level data in the unified time series to obtain the slow-changing trend component. The cross-correlation function between the instantaneous fluctuation amplitude and the slow-changing trend component is then calculated. The peak position of the cross-correlation function determines the delay time of the instantaneous fluctuation's impact on the slow-changing trend, thus obtaining a timeliness coordination benchmark for multi-scale data. These analytical results can serve as the basis for determining the coupling matrix. The coupling matrix can be determined using various methods. For example, by analyzing historical data offline, statistical methods (such as cross-correlation analysis, Granger causality tests, and information theory methods) can be used to identify the causal relationships and influence strengths between variables at different time scales, and then these relationships can be quantified into elements of the coupling matrix. Alternatively, an online adaptive adjustment approach can be adopted, for example, by using system identification technology or model predictive control-based optimization methods to dynamically update the elements of the coupling matrix based on real-time data streams to adapt to changes in the power grid's operating mode.

[0047] The non-zero elements in the coupling matrix represent the weights of the influence of a state variable at one time scale on a state variable at another time scale. These non-zero elements are the core of the coupling matrix, quantifying the strength of cross-scale influences. For example, if second-level voltage fluctuations have a significant impact on minute-level renewable energy output, the corresponding element in the coupling matrix will be a large non-zero value, representing the weight of this influence. These influence weights can be positive (representing a positive correlation) or negative (representing a negative correlation), and their absolute value reflects the strength of the influence. These weights can be set through expert experience or learned from historical operating data using data-driven methods (such as regression analysis, neural network training, etc.).

[0048] In practical applications, a key challenge is how to efficiently and accurately extract crucial information reflecting the health status of the power grid from this complex voltage state trajectory, and how to form a clear and coherent description of the overall voltage dynamic behavior to support subsequent power grid operation decisions and control optimization. (Refer to...) Figure 6 As shown, this application further proposes a method for extracting short-term power balance indices and medium-to-long-term safety margin indices from voltage state trajectories, and determining a coherent description of the overall voltage dynamic behavior. Specifically, the method includes the following steps: First, a recent segment of the voltage state trajectory for a preset duration is extracted, and the average and standard deviation of the voltage amplitude are calculated. The ratio of the average value to a preset voltage reference value is then used as a short-term power balance indicator. The voltage state trajectory is a time-consistent sequence of voltage estimates optimized by multi-scale Kalman filtering. Extracting a recent segment of the trajectory for a preset duration, such as voltage data from the past few minutes or tens of minutes, reflects the instantaneous or short-term operating characteristics of the power grid near the current time point. Calculating the average voltage amplitude reflects the overall voltage level in the recent period, while the standard deviation characterizes the severity of voltage fluctuations. The ratio of the average value to a preset voltage reference value (e.g., rated voltage or expected voltage) quantifies the deviation of the current voltage level from the ideal state, thus serving as a short-term power balance indicator. This indicator intuitively reflects the supply and demand balance of reactive power in the power grid in the short term; for example, a high ratio may indicate reactive power surplus, while a low ratio may indicate reactive power shortage. The preset duration can be determined based on the actual needs and response speed of the power grid, for example, set to 5 minutes, 10 minutes, or 15 minutes. Voltage reference values ​​are typically the standard voltage values ​​specified in power grid operation regulations. Calculating the average and standard deviation is a routine operation in statistics and can be achieved using sliding window averaging and standard deviation algorithms.

[0049] Secondly, a long-term segment of the voltage state trajectory for a preset duration is extracted, and the minimum and maximum voltage amplitudes are calculated. The first difference between the minimum value and a preset low-voltage threshold is used as the first safety margin indicator, and the second difference between the maximum value and a preset high-voltage threshold is used as the second safety margin indicator. This step focuses on the voltage safety boundary of the power grid during medium- to long-term operation. The long-term trajectory segment typically covers a longer time range, such as voltage data from the past few hours, a day, or even longer, aiming to assess the stability and reliability of the power grid over a longer timescale. Calculating the minimum and maximum voltage amplitudes can identify extreme voltage conditions that may occur in the power grid during long-term operation. The difference between the minimum value and the preset low-voltage threshold (e.g., the minimum voltage limit allowed by the power grid) is used to calculate the first safety margin indicator, reflecting the power grid's ability to withstand low-voltage risks. Similarly, the difference between the maximum value and the preset high-voltage threshold (e.g., the maximum voltage limit allowed by the power grid) is used to calculate the second safety margin indicator, reflecting the power grid's ability to withstand high-voltage risks. These two indicators together constitute the medium- to long-term safety margin indicators, used to assess the safety of power grid voltage operation. The preset duration can be determined based on the periodic characteristics of power grid operation and safety assessment requirements, such as being set to 1 hour, 6 hours, or 24 hours. The low-voltage threshold and high-voltage threshold are safety limits explicitly defined in power grid operation specifications. The difference calculation directly reflects the distance between the current voltage and the safety boundary.

[0050] Finally, the short-term power balance index, the first safety margin index, and the second safety margin index are combined in chronological order to form a coherent description of the overall voltage dynamic behavior. This step aims to integrate voltage indices with different time scales and focuses to create a comprehensive, systematic, and time-continuous profile of the grid voltage operation. The short-term power balance index focuses on instantaneous and short-term reactive power supply and demand, while the medium- and long-term safety margin index focuses on long-term voltage stability. By combining these indices in the chronological order of their generation, a multi-dimensional voltage state description sequence can be constructed. This sequence not only includes the current operating state of the grid but also reflects its changing trends and safety margins over a past period, thus providing grid dispatchers and operators with a coherent and intuitive understanding of the grid voltage dynamic behavior. This coherent description forms the basis for subsequent grid control strategy adjustments and risk warnings. The combination method can be a simple time series concatenation or the construction of a multi-dimensional vector sequence. For example, at each time point, a vector containing [short-term power balance index, first safety margin index, second safety margin index] can be generated, and then these vectors can be arranged in chronological order.

[0051] Furthermore, based on a coherent description of the overall voltage dynamic behavior, referring to Figure 7 As shown, this application also proposes a method for adjusting the time coordination parameters of the time-series fusion algorithm, including: comparing a coherent description of the overall voltage dynamic behavior with a preset ideal voltage dynamic curve, aiming to assess the difference between the currently perceived grid voltage state and the expected or optimal state. The coherent description of the overall voltage dynamic behavior is a combination of short-term power balance indices and medium-to-long-term safety margin indices extracted from the voltage state trajectory in the aforementioned steps, comprehensively reflecting the voltage characteristics of the grid at different time scales. The preset ideal voltage dynamic curve can be set according to grid operation procedures, historical best operating data, simulation model results, or expert experience, serving as a benchmark for measuring the grid voltage health status. The comparison process can be implemented using various mathematical methods, such as calculating the absolute difference, relative difference, mean square error, or using more complex similarity measurement functions between the two curves at various time points.

[0052] The deviation values ​​at each time point are calculated to quantify the degree of difference between a coherent description of the overall voltage dynamic behavior and a preset ideal voltage dynamic curve. This deviation value can be an instantaneous deviation, or an average or maximum deviation over a certain time period. For example, the difference in voltage amplitude at each time point, the difference between the short-term power balance index and the ideal value, or the difference between the safety margin index and the ideal margin can be calculated. These deviation values ​​provide a quantitative basis for subsequent judgment on whether parameter adjustments are needed.

[0053] When the deviation value exceeds a preset deviation threshold, the system determines the time and location of the deviation. The preset deviation threshold is a pre-defined tolerance level used to determine whether the deviation reaches a point requiring intervention. This threshold can be determined based on the safety and economic requirements of power grid operation, as well as the need for sensing accuracy. Once the deviation value exceeds this threshold, the system records the time point or time period of the deviation, which helps to trace the source of the problem and provides temporal location information for subsequent parameter adjustments.

[0054] Based on the data timescale corresponding to the time location, the interpolation weights or smoothing coefficients of the corresponding data types in the time series fusion algorithm are increased or decreased. This step is crucial for achieving adaptive adjustment. When generating a unified time series, the time series fusion algorithm involves interpolation and / or smoothing of data at different time scales. For example, linear interpolation may be used for second-level real-time voltage data; weighted moving average may be used for minute-level renewable energy output data; and exponential smoothing may be used for hourly load forecast data. Interpolation weights or smoothing coefficients are key parameters in these algorithms, determining the influence or smoothing degree of data at different time scales during the fusion process. When the deviation is mainly reflected in instantaneous fluctuations, it may be necessary to increase the weight of second-level data in the interpolation process to make it more sensitive to rapid changes; when the deviation is mainly reflected in short-term trends, it may be necessary to adjust the smoothing coefficient of minute-level data to better capture trend changes; and when the deviation is mainly reflected in medium- to long-term changes, it may be necessary to adjust the smoothing coefficient of hourly data. The specific direction of increase or decrease depends on the nature of the deviation. For example, if the perceived voltage fluctuation is lower than the ideal value, it may be necessary to increase the weight of instantaneous data to enhance the fluctuation performance.

[0055] The adjusted interpolation weights or smoothing coefficients are used as updated time coordination parameters. These dynamically adjusted parameters replace the original or default parameters and are used in subsequent time-series fusion algorithms, thus forming a closed-loop adaptive optimization process.

[0056] After adjusting the time coordination parameters of the time-series fusion algorithm based on a coherent description of the overall voltage dynamic behavior, these adjustments need to be effectively fed back into the subsequent filtering process. The adjusted parameters are used to update the iterative process of the multi-scale Kalman filter, outputting the corrected reactive voltage state estimation result. Figure 8As shown, the process includes: substituting the adjusted time coordination parameters into the time series fusion algorithm to regenerate a unified time series. This step aims to ensure that the data fusion process can fully utilize the optimization information provided by the feedback mechanism. In the above method, when a deviation in voltage dynamic behavior is detected and the time coordination parameters (e.g., interpolation weights or smoothing coefficients) of the time series fusion algorithm are adjusted accordingly, these adjusted parameters are reapplied to the time series fusion algorithm. The algorithm will then process the original second-level real-time voltage data, minute-level renewable energy output data, and hourly load forecast data again, generating a more accurate unified time series that better matches the current grid state through methods such as linear interpolation and cubic spline interpolation. This process ensures that the observation data basis upon which subsequent state estimation relies is optimized and most accurate.

[0057] Subsequently, the observation inputs in the multi-scale Kalman filter are updated based on the regenerated unified time series. The accuracy of the multi-scale Kalman filter is highly dependent on the quality of its observation inputs. Once a more accurate unified time series is regenerated using adjusted time coordination parameters, this series is used to update the observation inputs for the multi-scale Kalman filter. Specifically, this regenerated unified time series serves as new observation data, replacing the previous results used for preliminary estimation, thus providing the Kalman filter with an observational basis that more closely reflects the actual system state. This allows the Kalman filter to perform state estimation based on the latest, calibrated data, avoiding estimation biases caused by lagging or inaccurate observation inputs.

[0058] Based on this, the state prediction and state update processes of the multi-scale Kalman filter are re-executed with the updated observation input. After updating the observation input, the multi-scale Kalman filter needs to restart its core iterative process. This means that the filter's state prediction step will be based on the best estimate from the previous time step, while the state update step will use the newly provided, updated observation input to correct the prediction. This re-execution process will continue until the filter's state estimate reaches the convergence condition. Through this iterative process, the Kalman filter can fully absorb new observation information, gradually eliminating noise and biases that may arise during dynamic coupling, thereby generating a more accurate and time-consistent voltage state trajectory.

[0059] Finally, the converged state estimate after re-execution is output as the corrected reactive voltage state estimate. When the multi-scale Kalman filter is re-executed under the updated observation input and reaches convergence, its final output state estimate is considered the corrected reactive voltage state estimate. This result represents the reactive voltage state of the power grid after multiple rounds of data fusion, feedback adjustment, and filtering optimization, exhibiting higher accuracy and reliability. This corrected estimate can be directly used for power grid operation control, security analysis, or further decision support, providing crucial information for the stable operation of renewable energy-rich power grids.

[0060] In this embodiment, the timing fusion algorithm and the iterative process of the multi-scale Kalman filter form a closed-loop feedback. When the deviation between the coherent description of the overall voltage dynamic behavior and the preset ideal voltage dynamic curve continues to increase, the operation of adjusting the time coordination parameters of the timing fusion algorithm and updating the iterative process of the multi-scale Kalman filter is executed cyclically until the deviation is less than the preset convergence threshold.

[0061] Specifically, the time-series fusion algorithm and the iterative process of the multi-scale Kalman filter form a closed-loop feedback, meaning that the system's output (i.e., the overall voltage dynamic behavior reflected in the corrected reactive power voltage state estimation) will, in turn, affect the system's input or internal parameters, thus forming a self-regulating loop. This feedback mechanism ensures that the system can continuously adapt to changes in the grid's operating state, improving the robustness of the estimation. When the deviation between the coherent description of the overall voltage dynamic behavior and the preset ideal voltage dynamic curve continues to increase, it indicates that the current sensing result may be deviating from the actual or expected grid state, and this deviation has a cumulative or worsening trend, requiring deeper intervention and adjustment by the system. The continuous increase in deviation can be judged by monitoring the trend of deviation changes over multiple consecutive time steps. For example, if the deviation value for N consecutive time steps is greater than that of the previous time step, the deviation is considered to be continuously increasing. At this time, the system will cyclically execute the operations of adjusting the time coordination parameters of the time-series fusion algorithm and updating the multi-scale Kalman filter iterative process. This emphasizes that this adjustment is not a one-time event but a continuous process until the system reaches a stable state. Adjusting the time coordination parameters of the time-series fusion algorithm, such as interpolation weights or smoothing coefficients, aims to optimize the alignment and integration of multi-source data, making it more accurately reflect the current power grid state. Updating the multi-scale Kalman filter iteration process means using these adjusted parameters to re-perform the Kalman filter state prediction and update, obtaining a more accurate and stable voltage state estimate. This iterative process continues until the deviation is less than a preset convergence threshold. The preset convergence threshold is a pre-defined, acceptable error range. When, after multiple iterations, the deviation between the coherent description of the overall voltage dynamic behavior and the ideal curve decreases below this threshold, it indicates that the system has converged to a stable and accurate estimation state. At this point, the current iterative adjustment can be stopped, or a monitoring mode can be entered, waiting for the next increase in deviation before triggering adjustment again. This ensures that the system does not perform unnecessary calculations after reaching satisfactory accuracy, improving efficiency.

[0062] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A multi-timescale reactive voltage situational awareness method for a renewable energy-rich power grid, characterized in that, Includes the following steps: The system collects real-time voltage data at the second level, renewable energy output data at the minute level, and load forecast data at the hour level from the power grid. It then uses a time-series fusion algorithm to align the time scales of the multi-source heterogeneous data and generate a unified time series. Based on a unified time series, dynamic coupling characteristics between instantaneous fluctuations and slow changes are extracted to determine the timeliness coordination benchmark for multi-scale data. Based on the timeliness coordination benchmark, the instantaneous information of second-level data, the short-term trend of minute-level data, and the medium- and long-term changes of hour-level data are integrated in a hierarchical manner to generate preliminary estimation results without cumulative bias. The preliminary estimation results are iteratively optimized using multi-scale Kalman filtering to suppress noise and bias propagation during dynamic coupling and generate a voltage state trajectory with time consistency. Short-term power balance indices and medium- to long-term safety margin indices are extracted from the voltage state trajectory to determine a coherent description of the overall voltage dynamic behavior. Based on a coherent description of the overall voltage dynamic behavior, the time coordination parameters of the time-series fusion algorithm are adjusted, and the iterative process of the multi-scale Kalman filter is updated using the adjusted parameters to output the corrected reactive voltage state estimation results.

2. The multi-timescale reactive power and voltage situational awareness method for renewable energy-rich power grids according to claim 1, characterized in that: Time-scale alignment of multi-source heterogeneous data is performed using a time-series fusion algorithm to generate a unified time series, including: The second-level real-time voltage data, minute-level renewable energy output data, and hourly load forecast data are stored with their respective original timestamps. The time point of the second-level data is used as the reference time axis. For the minute-level renewable energy output data, linear interpolation is used to calculate the output value corresponding to each second. For the hourly load forecast data, cubic spline interpolation is used to calculate the load forecast value corresponding to each second. The interpolated data is aligned with the second-level real-time voltage data according to the time point to form a unified time series with a time interval of seconds.

3. The multi-timescale reactive power and voltage situational awareness method for renewable energy-rich power grids according to claim 1, characterized in that: Extracting the dynamic coupling characteristics between instantaneous fluctuations and slow changes, and determining the timeliness coordination benchmark for multi-scale data, including: Calculate the root mean square value within a sliding window for second-level data in a unified time series to obtain the instantaneous voltage fluctuation amplitude; Low-pass filtering is applied to minute-level and hour-level data in a unified time series to obtain slowly changing trend components. Calculate the cross-correlation function between the instantaneous fluctuation amplitude and the slow trend component. Determine the impact delay time of the instantaneous fluctuation on the slow trend based on the peak position of the cross-correlation function. Use the impact delay time as the benchmark for timeliness coordination of multi-scale data.

4. The multi-time-scale reactive voltage situational awareness method for renewable energy-rich power grids according to claim 1, characterized in that: Based on a timeliness-coordinated benchmark, the instantaneous information of second-level data, the short-term trends of minute-level data, and the medium- and long-term changes of hour-level data are hierarchically integrated to generate preliminary estimation results without cumulative bias, including: The time window length for tiered integration is determined based on the timeliness coordination benchmark; Within the time window, the instantaneous information of the second-level data is used as the first-level estimation input, and the voltage instantaneous change curve is fitted by the recursive least squares method. Using the short-term trend of minute-level data as the input for the second-level estimation, the trend features are extracted using the weighted moving average method. Using the medium- to long-term changes in hourly data as the third-level estimation input, the exponential smoothing method is used to extract the change patterns. The first-level estimation results are weighted and fused with the second-level estimation results, and then the fused results are superimposed with the third-level estimation results to obtain a preliminary estimation result without cumulative bias.

5. The multi-timescale reactive power and voltage situational awareness method for a renewable energy-rich power grid according to claim 1, characterized in that: The preliminary estimation results are iteratively optimized using multi-scale Kalman filtering to suppress noise and bias propagation during dynamic coupling, generating a time-consistent voltage state trajectory, including: Using the preliminary estimation results as the observation input, a multi-scale state-space model containing second-level, minute-level, and hour-level state variables was established. The state transition matrix of the second-level state variables was set as a diagonal matrix according to the voltage fluctuation characteristics, the state transition matrix of the minute-level state variables was set according to the rate of change of new energy output, and the state transition matrix of the hour-level state variables was set according to the load change pattern. The process noise covariance matrix and observation noise covariance matrix for the corresponding state variables at the second, minute, and hour levels are set respectively. The state prediction and state update are performed sequentially using the Kalman filter recursive formula. In each iteration, the state prediction value for the next moment is corrected based on the state update result. The iteration continues until convergence, and the converged state estimate is used as the voltage state trajectory with time consistency.

6. The multi-timescale reactive voltage situational awareness method for renewable energy-rich power grids according to claim 5, characterized in that: In a multi-scale state-space model, state variables at the second, minute, and hour levels are linked through a coupling matrix. The coupling matrix is ​​determined based on dynamic coupling characteristics, and the non-zero elements in the coupling matrix represent the influence weight of a state variable at one time scale on a state variable at another time scale.

7. The multi-time-scale reactive voltage situational awareness method for renewable energy-rich power grids according to claim 1, characterized in that: Short-term power balance indices and medium- to long-term safety margin indices are extracted from the voltage state trajectory to determine a coherent description of the overall voltage dynamic behavior, including: Extract recent trajectory segments of a preset duration from the voltage state trajectory, calculate the average value and standard deviation of the voltage amplitude, and use the ratio of the average value to the preset voltage reference value as a short-term power balance index. Extract a long-term trajectory segment of a preset duration from the voltage state trajectory, calculate the minimum and maximum values ​​of the voltage amplitude, take the first difference between the minimum value and the preset low voltage threshold as the first safety margin index, and take the second difference between the maximum value and the preset high voltage threshold as the second safety margin index. By combining the short-term power balance index, the first safety margin index, and the second safety margin index in chronological order, a coherent description of the overall voltage dynamic behavior is formed.

8. The multi-timescale reactive power and voltage situational awareness method for a renewable energy-rich power grid according to claim 1, characterized in that: Based on a coherent description of the overall voltage dynamic behavior, the time coordination parameters of the timing fusion algorithm are adjusted, including: The coherent description of the overall voltage dynamic behavior is compared with the preset ideal voltage dynamic curve, and the deviation value at each time point is calculated. When the deviation value exceeds the preset deviation threshold, the time position of the deviation is determined. According to the data time scale corresponding to the time position, the interpolation weight or smoothing coefficient of the corresponding data type in the time series fusion algorithm is increased or decreased, and the adjusted interpolation weight or smoothing coefficient is used as the updated time coordination parameter.

9. The multi-time-scale reactive voltage situational awareness method for renewable energy-rich power grids according to claim 1, characterized in that: The iterative process of updating the multi-scale Kalman filter using the adjusted parameters outputs the corrected reactive voltage state estimation results, including: The adjusted time coordination parameters are substituted into the time series fusion algorithm to regenerate a unified time series. The observation input in the multi-scale Kalman filter is updated based on the regenerated unified time series. The state prediction and state update process of the multi-scale Kalman filter is re-executed with the updated observation input. The converged state estimate after re-execution is output as the corrected reactive voltage state estimate result.

10. The multi-timescale reactive voltage situational awareness method for a renewable energy-rich power grid according to claim 1, characterized in that: The timing fusion algorithm and the iterative process of the multi-scale Kalman filter form a closed-loop feedback. When the deviation between the coherent description of the overall voltage dynamic behavior and the preset ideal voltage dynamic curve continues to increase, the timing coordination parameters of the timing fusion algorithm and the iterative process of the multi-scale Kalman filter are cyclically adjusted until the deviation is less than the preset convergence threshold.

Images