An automatic voltage preventive control method based on voltage fluctuation rate prediction

Abstract

The application provides an automatic voltage preventive control method based on voltage fluctuation rate prediction, comprising the following steps: when a current control period arrives, acquiring voltage sequence data of a target area bus in a past period and correcting the voltage sequence data; based on the corrected voltage sequence, fitting an ARIMA model for predicting a voltage base value and a GARCH model for analyzing voltage fluctuation characteristics, respectively generating a voltage base value prediction sequence and a voltage fluctuation rate prediction sequence in a future period, finally generating a voltage prediction sequence in the future period, and then obtaining a dynamic voltage safety domain of the current control period; embedding the safety domain as a constraint condition into an automatic voltage control system to realize automatic voltage preventive control. Through the fusion of time series analysis and fluctuation rate modeling technology, the application can fuse voltage trend and fluctuation rate prediction, and generate a dynamic control threshold value accordingly, so that the accurate prediction and dynamic regulation of the power grid voltage are realized.

Description

Technical Field

[0001] This invention belongs to the field of automatic voltage control technology for power systems, and specifically relates to an automatic voltage prevention control method based on voltage fluctuation rate prediction. Background Technology

[0002] Voltage stability in power systems is one of the core elements for ensuring the safe and economical operation of the power grid. As a key indicator of power quality, voltage stability control not only relates to power supply reliability but also directly affects the lifespan of power equipment, user experience, and the overall operating efficiency of the power grid. With the integration of a high proportion of renewable energy sources, the power grid's power supply structure has shifted from being dominated by traditional synchronous machines to a hybrid form with widespread integration of power electronic equipment. This has significantly reduced system inertia and damping characteristics, while dramatically increasing voltage volatility and randomness. Traditional automatic voltage control (AVC) methods based on deterministic models and static thresholds face the challenge of insufficient adaptability.

[0003] Traditional automatic voltage control systems often employ a passive "over-limit response" mode, triggering control only when the actual voltage value exceeds a preset fixed threshold. This control strategy exhibits significant lag and often falls short in responding to rapid voltage changes caused by renewable energy fluctuations. Existing AVC systems typically have long control cycles (5-30 minutes), failing to capture minute-level voltage fluctuations, resulting in insufficient control frequency or inappropriate timing. Furthermore, traditional systems use fixed voltage limit settings, neglecting the dynamic characteristics of grid operation, especially in scenarios with alternating peak and off-peak loads and drastic fluctuations in renewable energy output, often leading to over-control or under-control. Research shows that in areas with high renewable energy penetration, the frequency of control actions by traditional AVC systems increases 3-5 times, while the voltage qualification rate decreases by 2-3 percentage points, and equipment lifespan is significantly shortened due to frequent operation.

[0004] In voltage prediction technology, existing short-term voltage prediction methods are mainly divided into two categories: physical model methods and data-driven methods. Physical model methods rely on precise grid topology and parameters, resulting in high computational complexity and difficulty in adapting to real-time grid changes. In data-driven methods, commonly used simple time series models such as AR and MA cannot effectively capture the nonlinear and time-varying characteristics of voltage fluctuations. Although some studies have introduced machine learning methods such as neural networks and support vector machines, these methods often require large amounts of training data and have limited ability to model volatility. In particular, existing prediction technologies generally lack quantitative analysis of voltage fluctuation uncertainty, failing to provide statistically meaningful prediction intervals, making it difficult to directly apply the prediction results to control decisions. In areas with high penetration of renewable energy, voltage fluctuations exhibit significant heteroscedasticity (volatility clustering effect), meaning that large fluctuations often follow large fluctuations, and small fluctuations often follow small fluctuations; this characteristic has not been fully considered by existing prediction models.

[0005] Looking at the current state of research both domestically and internationally, the IEEE PES working group pointed out in its 2020 report that modern power grid voltage control urgently needs to shift from a "reactive" to a "predictive" approach. The European Grid Operators Alliance (ENTSO-E) proposed a voltage control framework based on short-term load forecasting, but it did not consider the fluctuation characteristics of renewable energy sources. Some domestic power grid companies have attempted to introduce ultra-short-term load forecasting into AVC systems, but due to the non-simple linear relationship between voltage and load, and the influence of network structure, reactive power distribution, and other factors, the forecast accuracy is difficult to meet control requirements. While the preventative voltage control strategy developed by the US PJM grid considers future state prediction, its calculation cycle is as long as one hour, making it unsuitable for rapidly changing operating environments. Currently, there is a lack of practical technical solutions internationally that can simultaneously consider voltage trend forecasting and volatility quantification, especially a systematic approach to achieve accurate forecasting and dynamic control on a minute-level timescale.

[0006] Furthermore, existing research often focuses on improvements to single technical aspects, lacking a synergistic optimization across the entire chain of "prediction-decision-control-evaluation." Voltage control decisions are disconnected from prediction results; the output of prediction systems is often a point prediction rather than an interval prediction incorporating probabilistic characteristics, making it difficult to quantify control risks. Control strategy design fails to consider prediction uncertainty, resulting in overly rigid or conservative control boundaries. In areas with a high proportion of renewable energy, voltage fluctuation frequency increases by 2-3 times. Traditional control strategies lead to a surge in the number of actions of on-load tap-changing transformers, capacitor banks, and other equipment, increasing equipment failure rates by over 30% and significantly raising maintenance costs. More seriously, frequent control actions can trigger system oscillations; in 2021, a regional power grid experienced a localized voltage collapse due to uncoordinated actions of its automatic voltage control system.

[0007] From the perspective of technological development needs, the power system is rapidly evolving into a new type of power system characterized by coordinated interaction among power generation, grid, load, and storage. Voltage control technology urgently needs to break through the traditional framework and achieve an active management model of "prediction-prevention-prevention control." This requires new voltage control methods to possess minute-level accurate prediction capabilities, quantify the uncertainty of voltage fluctuations, and dynamically optimize control thresholds based on prediction results. Simultaneously, the new control methods should be compatible with existing AVC systems, supporting a smooth transition and lowering the technical implementation threshold. In the context of a "high-proportion renewable energy, high-proportion power electronic equipment" power system, developing forward-looking automatic voltage prevention and control technologies is not only an urgent need to enhance grid resilience but also a key support for achieving the safe and stable operation of the new power system. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of existing technologies and propose an automatic voltage prevention and control method based on voltage fluctuation rate prediction. This invention integrates time series analysis and fluctuation rate modeling techniques to combine voltage trends and fluctuation rate predictions, and generates dynamic control thresholds accordingly, thereby achieving accurate prediction and dynamic regulation of grid voltage.

[0009] This invention proposes an automatic voltage prevention control method based on voltage fluctuation rate prediction, comprising:

[0010] At the moment the current control cycle arrives, the voltage sequence data of the target area bus over the past period is acquired and corrected to obtain the corrected voltage sequence;

[0011] Based on the corrected voltage sequence, an ARIMA model for predicting the voltage baseline is fitted, and the fitted ARIMA model is used to generate a voltage baseline prediction sequence for future periods.

[0012] Based on the corrected voltage sequence and the fitted ARIMA model, a voltage data residual sequence for the past period is generated, thereby fitting a GARCH model for analyzing voltage fluctuation characteristics. Then, the fitted GARCH model is used to generate a voltage fluctuation prediction sequence for the future period.

[0013] Based on the voltage base prediction sequence and the voltage volatility prediction sequence, the final voltage prediction sequence for the future period is generated, thereby calculating the dynamic voltage safety domain of the current control cycle.

[0014] The dynamic voltage safety domain is embedded as a constraint in the automatic voltage control system to achieve automatic voltage prevention control of the target area.

[0015] In one specific embodiment of the present invention, obtaining the corrected voltage sequence includes:

[0016] 1) At the moment the current control cycle arrives Obtain the original voltage sequence of the target area bus over the previous 48 hours:

[0017]

[0018] in, This represents the magnitude of the bus voltage at time t, where t = -2880, -2879, ..., ;

[0019] 2) Correct the data in the original voltage sequence obtained in step 1) to obtain the corrected voltage sequence;

[0020] Among them, outliers in the original voltage sequence are identified by judging all the data in the original voltage sequence;

[0021] If the data in the original voltage sequence is not an outlier, then let the corrected voltage data at that moment be the original data, that is, let = , express The voltage value after real-time correction;

[0022] If the data in the original voltage sequence are outliers, cubic spline interpolation is used to correct them.

[0023]

[0024] in, Indicates the size of the time step for voltage correction; For the weighting coefficients, satisfying ;like Then let ;

[0025] After all data in the original voltage sequence has been evaluated and corrected, the corrected voltage sequence is obtained. .

[0026] In one specific embodiment of the present invention, the method for identifying outliers in the original voltage sequence is as follows:

[0027] like Then determine It is an outlier; otherwise, Not an outlier;

[0028] in, For Center time, window size Local average voltage over a window of minutes. For Center time, window size Voltage standard deviation for a window of minutes; This is the sensitivity adjustment coefficient.

[0029] In a specific embodiment of the present invention, generating a voltage baseline prediction sequence for future time periods using the fitted ARIMA model includes:

[0030] 1) Obtain the corrected voltage sequence The difference order required for quiescence is denoted as d;

[0031] 2) Based on the results of step 1), establish the ARIMA(p,d,q) model, with the following expression:

[0032]

[0033] Where B is the shift operator, ;

[0034] It is a p-order autoregressive polynomial. , These are p-order autoregressive coefficients, representing the voltage value at the current moment. The degree of dependence on the voltage value at the previous i time points; This means shifting the time series forward by i units, i.e. ;

[0035] It is a q-order moving average polynomial. , These are the q-order moving average coefficients, representing the error term at the current time step. The impact on the error term at the first j time steps;

[0036] It is a white noise sequence with zero mean. Homoscedasticity ; denoted as the variance of the white noise sequence;

[0037] 3) Determine the model parameters (p, d, q) by minimizing the AIC criterion:

[0038]

[0039] in, For based on Calculate the maximum likelihood estimate;

[0040] 4) Based on The data in the dataset were used to estimate the parameters using the conditional least squares method. and ;

[0041] 5) Based on the results of step 4), predict the future. to The voltage at a given time is recursively predicted to obtain the voltage baseline prediction for the next 30 minutes, thus generating a voltage baseline prediction sequence for the next 30 minutes. ;

[0042] in,

[0043]

[0044] in, express Prediction of the base voltage value at time;

[0045] This represents the historical or predicted voltage value at time t; when hour, Take the voltage correction value at time t ;when hour, Predicting the base voltage value at time t ;

[0046] Represents the residual observed or predicted value at time t; when hour, Take the actual residual value at time t. ;when hour, Take the predicted residual value at time t. .

[0047] In one specific embodiment of the present invention, the step of obtaining the corrected voltage sequence The difference order required for quiescence includes:

[0048] The ADF unit root test is used to determine the stationarity of a voltage sequence. The input parameter of the ADF test is the current voltage sequence, and the output result is the P value corresponding to that sequence.

[0049] Then, the P-value is determined: if the P-value is less than the preset fluctuation threshold, the current sequence is considered stable, and the current sequence is recorded relative to the corrected voltage sequence. The difference order d; if If the system passes the stationarity test, then the difference order d = 0.

[0050] If the P-value is greater than or equal to the preset fluctuation threshold, it indicates that the current sequence is non-stationary, and a first-order difference is performed on the current sequence. Then, the sequence obtained after the first-order difference is used as the new current sequence and a stationarity test is performed. If the new current sequence passes the test, the current sequence relative to the corrected voltage sequence is recorded. The difference order d is determined; if the new current sequence fails the test, the current sequence is further differencing by first order until the final current sequence passes the test, and the final difference order d is recorded.

[0051] In a specific embodiment of the present invention, generating the voltage volatility prediction sequence for the future time period using the fitted GARCH model includes:

[0052] 1) Using the fitted ARIMA model, generate the target region busbar from... 48 hours ago Predicting the base voltage value within a given time period;

[0053] 2) Based on the results of step 1), extract the residual sequence;

[0054]

[0055] in, This represents the residual between the voltage correction value at time t and the predicted voltage baseline value;

[0056] 3) Establish the GARCH(1,1) model:

[0057]

[0058] in, This represents the conditional variance at time t. Indicates the level of basic volatility. ; This represents the influence coefficient of the new disturbance. ; Indicates the historical volatility persistence coefficient. ; ;

[0059] based on The data in the dataset were used to determine the parameters of the GARCH model using the maximum likelihood estimation method. , , The estimated values ​​are denoted as follows: , , ;

[0060] 4) Based on the results of step 3), recursively predict the voltage fluctuation rate for the next 30 minutes;

[0061]

[0062] in, express The conditional variance of the time-based prediction is the squared volatility.

[0063] hour, Use actual residuals ; hour, Set to 0;

[0064] Finally, a voltage fluctuation prediction sequence for the next 30 minutes is obtained. .

[0065] In one specific embodiment of the present invention, generating the final voltage prediction sequence for the future time period includes:

[0066] 1) Based on the voltage fluctuation rate prediction sequence, generate random shock values ​​and form a random shock sequence;

[0067] Among them, the Monte Carlo simulation method is used to generate M sets of random shock sequences to simulate the random effects of external fluctuations on voltage:

[0068]

[0069] in, Indicates the m-th random shock sequence in The value at time; express Predict the voltage fluctuation rate at any time; To obtain from the standard normal distribution Random numbers drawn from the sample; Indicates the number of Monte Carlo simulations;

[0070] 2) Using the results of step 1), generate the predicted voltage value;

[0071]

[0072]

[0073] in, Indicates the m-th simulation in The predicted voltage value at that moment; express The final predicted voltage value at time;

[0074] Finally, the voltage prediction sequence for the next 30 minutes was obtained. .

[0075] In one specific embodiment of the present invention, calculating the dynamic voltage safety domain of the current control cycle includes:

[0076] 1) Calculate the maximum consecutive increase and the maximum consecutive decrease that may occur at each moment in the voltage prediction sequence within the prediction period;

[0077] 2) Based on the results of step 1), independently calculate the dynamic safety boundary for each time t:

[0078]

[0079]

[0080] in, This represents the dynamically calculated upper limit of voltage safety at time t; This represents the dynamically calculated lower voltage safety limit at time t; This indicates the pre-set planned voltage limit; This indicates the pre-set lower limit of the planned voltage; This represents the maximum continuous increase that may occur within the predicted time period, starting from time t. This represents the maximum consecutive decrease that may occur within the predicted time period, starting from time t. This represents the safety factor.

[0081] In a specific embodiment of the present invention, the calculation of the maximum continuous rise and maximum continuous fall that may occur at each moment corresponding to the voltage prediction sequence within the prediction period includes:

[0082] 1-1) Initially, let the current time be... Set a threshold;

[0083] 1-2) Order , This is a temporary variable used to record the cumulative voltage change during the current continuous change process;

[0084] 1-3) Judgment:

[0085] If satisfied If so, proceed to steps 1-4) to determine the continuous upward trend of voltage starting from time t;

[0086] If satisfied If so, proceed to steps 1-5) to determine the continuous downward trend of voltage starting from time t;

[0087] If none of these conditions are met, proceed to steps 1-6).

[0088] 5-1-4) Determining the continuous upward trend of voltage starting from time t; the specific steps are as follows:

[0089] 1-4-1) Judgment: If ,in, If the length of the voltage prediction sequence is equal to the length of the voltage prediction sequence, proceed to step 1-4-2; otherwise, proceed to step 1-4-3.

[0090] 1-4-2) If Then let Then let (Return to step 1-4-1); otherwise, record. Then proceed to steps 1-6). This represents the maximum continuous increase that may occur within the predicted time period, starting from time t.

[0091] 1-4-3) Record Then proceed to steps 1-6).

[0092] 1-5) Determining the continuous decreasing trend of voltage starting from time t; the specific steps are as follows:

[0093] 1-5-1) Judgment: If If yes, proceed to step 1-5-2); otherwise, proceed to step 1-5-3).

[0094] 1-5-2) If Then let Then let (Return to step 1-5-1); otherwise, record. Then proceed to steps 1-6). This represents the maximum consecutive decrease that may occur within the predicted time period, starting from time t.

[0095] 1-5-3) Record Then proceed to steps 1-6).

[0096] 1-6) Let t = t + 1, determine:

[0097] like If the result is positive, return to steps 1-2) and continue the judgment; otherwise, the judgment ends, and the result corresponding to each time t is output. and ; where, for each time t corresponding to the voltage prediction sequence, if the corresponding time is not generated after the determination is completed. or Then, the one that was not generated at that moment and All are recorded as 0.

[0098] The features and beneficial effects of this invention are as follows:

[0099] 1) High prediction accuracy: This invention adopts the ARIMA-GARCH combined model, which integrates the voltage base value trend and volatility characteristics, and combines Monte Carlo simulation to quantify random disturbances. The prediction error of the voltage every minute for the next 30 minutes can be controlled within ±1kV, which is significantly better than the traditional single model.

[0100] 2) Dynamic adaptation of the safety domain: This invention dynamically corrects the upper and lower limits of the safety domain based on the continuous change trend of voltage, avoiding the defect that fixed thresholds cannot cope with rapid fluctuations, and reserving adjustment space in advance, which can reduce the risk of voltage exceeding the limit by more than 40%.

[0101] 3) Strong forward-looking control: This invention uses a 5-minute control cycle and a 30-minute short-term prediction to achieve proactive "prediction-prevention" control, reducing the regulation lag time by 80% and effectively avoiding voltage anomalies caused by new energy fluctuations.

[0102] 4) Balancing economy and reliability: This invention reduces ineffective switching of reactive power equipment by more than 30% through volatility risk classification and control dead zone design, extends equipment life, and reduces grid losses, thus balancing voltage quality and operational economy. Attached Figure Description

[0103] Figure 1 This is an overall flowchart of an automatic voltage prevention control method based on voltage fluctuation rate prediction according to an embodiment of the present invention. Detailed Implementation

[0104] This invention proposes an automatic voltage prevention and control method based on voltage fluctuation rate prediction, which is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0105] This invention proposes an automatic voltage prevention control method based on voltage fluctuation rate prediction, comprising:

[0106] At the moment the current control cycle arrives, the voltage sequence data of the target area bus over the past period is acquired and corrected to obtain the corrected voltage sequence;

[0107] Based on the corrected voltage sequence, an ARIMA model for predicting the voltage baseline is fitted, and the fitted ARIMA model is used to generate a voltage baseline prediction sequence for future periods.

[0108] Based on the corrected voltage sequence and the fitted ARIMA model, a voltage data residual sequence for the past period is generated, thereby fitting a GARCH model for analyzing voltage fluctuation characteristics. Then, the fitted GARCH model is used to generate a voltage fluctuation prediction sequence for the future period.

[0109] Based on the voltage base prediction sequence and the voltage volatility prediction sequence, the final voltage prediction sequence for the future period is generated, thereby calculating the dynamic voltage safety domain of the current control cycle.

[0110] The dynamic voltage safety domain is embedded as a constraint in the automatic voltage control system to achieve automatic voltage prevention control of the target area.

[0111] In a specific embodiment of the present invention, the overall process of the automatic voltage prevention control method based on voltage fluctuation rate prediction is as follows: Figure 1 As shown, it includes the following steps:

[0112] 1) At the arrival time of the current control cycle, acquire and correct the voltage sequence data of the target area bus over the previous 48 hours to obtain the corrected voltage sequence; the specific steps are as follows:

[0113] 1-1) At the moment the current control cycle arrives, obtain the original voltage sequence of the target area bus over the previous 48 hours.

[0114] This embodiment uses a 5-minute control cycle as one complete control cycle. At the beginning of each control cycle... The system automatically collects data from the target area busbars via the EMS system. 48 hours ago The original voltage sequence is composed of minute-level historical voltage data within a given time period.

[0115]

[0116] in, This represents the bus voltage amplitude (pu) at time t, where 2880 is the number of minutes corresponding to 48 hours, and t = -2880, -2879, ..., .

[0117] 1-2) Correct the data in the original voltage sequence obtained in step 1-1) to obtain the corrected voltage sequence.

[0118] In this embodiment, outliers in the original voltage sequence are identified by evaluating all data in the original voltage sequence. A specific embodiment of the present invention employs an improved method. The criteria define outliers, where:

[0119] like Then determine It is an outlier; otherwise, It is not an outlier.

[0120] in, For Center time, window size Local average voltage over a window of minutes. For Center time, window size The standard deviation of voltage within a minute window period; in this embodiment This is the sensitivity adjustment coefficient.

[0121] In this embodiment, if the data in the original voltage sequence is not an outlier, then the corrected voltage data at that moment is set to the original data, i.e., ... = , express The voltage value after constant correction.

[0122] If the data in the original voltage sequence are outliers, cubic spline interpolation is used to correct them.

[0123]

[0124] in, Indicating the time step size for voltage correction, in this embodiment, for When making corrections, use The voltage data is centered at a time point and extended forward and backward by three time steps (a total of seven points). This means that the correction considers the voltage values ​​from t−3 to t+3, a total of seven adjacent time points, forming a local data window; where, if Then let .

[0125] These are the weighting coefficients, determined by a cubic spline function, where the coefficients satisfy... This ensures that the corrected voltage value is a weighted average of the original data and does not produce any offset.

[0126] After all data in the original voltage sequence has been evaluated and corrected, the corrected voltage sequence is obtained. .

[0127] 2) Use the results of step 1) to fit an ARIMA model for predicting the voltage base value, and use the fitted ARIMA model to generate a voltage base value prediction sequence for the next 30 minutes starting from the arrival time of the current control cycle; the specific steps are as follows:

[0128] 2-1) Obtain the difference order required to make the corrected voltage sequence stationary.

[0129] In this embodiment, for The sequence was subjected to a stationarity test. An enhanced Dickey-Fuller (ADF) unit root test was used to determine the stationarity of the sequence. The input parameters for the ADF test were the current voltage sequence (initially...). (Sequence), the output result is the P value corresponding to the sequence.

[0130] Then, the P value is determined: if the P value is less than a preset fluctuation threshold, in one specific embodiment of the present invention, if the P value is <0.05, the current sequence is considered to be stable, and then the current sequence is recorded relative to the corrected voltage sequence. The difference order d. If If the system passes the stationarity test, then the difference order d = 0.

[0131] If the P-value is greater than or equal to the preset fluctuation threshold, it indicates that the current sequence is non-stationary and requires first-order differencing. In a specific embodiment of the present invention, if... If the stationarity test is not passed, then... Perform first-order difference:

[0132]

[0133] in, This represents the change in voltage at time t relative to the previous time.

[0134] Next, the sequence obtained after first-order differencing is used as the new current sequence and a stationarity test is performed. If the new current sequence passes the test, the current sequence relative to the corrected voltage sequence is recorded. The difference order d is determined; if the new current sequence fails the test, the current sequence is further subdivided by the first order until the final current sequence passes the test, and the final difference order d is recorded.

[0135] 2-2) Based on the results of step 2-1), establish the ARIMA(p,d,q) model, with the following expression:

[0136]

[0137] Where B is the shift operator, .

[0138] It is a p-order autoregressive polynomial. , These are p-order autoregressive coefficients, representing the voltage value at the current moment. The degree of dependence on the voltage value at the previous i time points; This means shifting the time series forward by i units, i.e. .

[0139] It is a q-order moving average polynomial. , These are the q-order moving average coefficients, representing the error term at the current time step. The impact on the error term at the first j time steps.

[0140] d is to make the sequence The stationary difference order is obtained through step 2-1).

[0141] The sequence is white noise, representing unpredictable random disturbances or residuals, and has zero mean. Homoscedasticity ; The variance of the white noise sequence represents the degree of fluctuation of random disturbances or residuals.

[0142] 2-3) Determine the model parameters (p, d, q) by minimizing the AIC criterion:

[0143]

[0144] in, For based on Calculate the maximum likelihood estimate.

[0145] 2-4) Based on The data in the dataset were used to estimate the parameters using the conditional least squares method. and .

[0146] 2-5) Based on the results of steps 2-4), for the future... to The voltage at time (30 data points) is recursively predicted to obtain the base voltage prediction for the next 30 minutes:

[0147]

[0148] in, express Predicting the base voltage value at a given time.

[0149] This represents the historical or predicted voltage value at time t; when hour, Take the voltage correction value at time t ;when hour, Pick .

[0150] In this embodiment, historical data is first used to calculate... Then use the predicted value as Used for calculation And so on, until the calculation is obtained. .

[0151] This represents the observed or predicted residual value at time t, which also varies depending on the time point; when hour, Take the actual residual value at time t. ;when hour, Take the predicted residual value at time t. .

[0152] In this embodiment, the final 30-minute voltage base prediction sequence is obtained. (Unit: kV), providing a basis for subsequent volatility prediction.

[0153] 3) Using the corrected voltage sequence obtained in step 1) and the fitted ARIMA model in step 2), generate the voltage data residual sequence for the first 48 hours, thereby fitting the GARCH model used to analyze the voltage fluctuation characteristics. Then, use the fitted GARCH model to generate the voltage fluctuation rate prediction sequence for the next 30 minutes from the arrival time of the current control cycle.

[0154] In this embodiment, voltage fluctuation characteristics are analyzed, especially for non-stationary random disturbances caused by wind power fluctuations, to generate a voltage fluctuation rate prediction sequence, which can be used to accurately predict the intensity of voltage fluctuations within the next 30 minutes; the specific steps are as follows:

[0155] 3-1) Using the ARIMA model fitted in step 2), generate the target region busbar from... 48 hours ago Predicting the base voltage value within a given time period.

[0156] 3-2) Based on the results of steps 1) and 3-1), extract the residual sequence.

[0157]

[0158] in, This represents the residual between the voltage correction value at time t and the predicted voltage base value, reflecting the deviation between the actual voltage and the predicted base value; This represents the corrected voltage value at time t, derived from step 1). ;

[0159] This represents the voltage baseline prediction fitted by the ARIMA model at time t.

[0160] In this embodiment, voltage fluctuation information that the ARIMA model failed to capture in step 2) can be obtained by extracting the residual sequence.

[0161] 3-3) To address the volatility clustering characteristics of the residual sequence, a GARCH(1,1) model is established:

[0162]

[0163] in, This represents the conditional variance (squared volatility) at time t. Indicates the level of basic volatility. . This represents the influence coefficient of the new disturbance. This reflects the impact of new information, such as external mutations, on fluctuations. Indicates the historical volatility persistence coefficient. This reflects the fluctuation memory effect. This represents the residual at time t-1. This represents the conditional variance at time t-1.

[0164] There is a key constraint in this model: This ensures a smooth fluctuation process.

[0165] In this embodiment, based on The data in the figure were used to determine the parameters of the GARCH model through maximum likelihood estimation. , , The estimated values ​​obtained are denoted as follows: , , .

[0166] 3-4) Based on the results of step 3-3), recursively predict the voltage fluctuation rate for the next 30 minutes.

[0167] In this embodiment, the prediction process considers the current fluctuation state and historical patterns to provide forward-looking information for dynamic voltage control:

[0168]

[0169] in, express The conditional variance (squared volatility) of the time-based forecast represents a quantitative estimate of the strength of future volatility. , , This represents the estimated values ​​of the GARCH model parameters, which are the fitting results from step 3-3).

[0170] Residual usage strategy: (At historical moments) use actual residuals ; At the (future moment), the future residual is unknown and is set to 0.

[0171] Finally, a voltage fluctuation prediction sequence for the next 30 minutes is obtained. This provides the core input for the generation of voltage predictions and dynamic planning domain calculations in the next part.

[0172] 4) Based on the prediction results of steps 2) and 3), generate the final voltage prediction sequence for the next 30 minutes starting from the arrival time of the current control cycle.

[0173] This embodiment constructs a complete voltage prediction model by integrating deterministic trends (baseline prediction) and stochastic fluctuations (volatility prediction); the specific steps are as follows:

[0174] 4-1) Based on the voltage fluctuation rate prediction sequence in step 3), generate random shock values ​​to form a random shock sequence.

[0175] In this embodiment, the Monte Carlo simulation method is used to generate M=100 sets of random impact sequences to simulate the random effects of external fluctuations on voltage:

[0176]

[0177] in, Indicates the m-th random shock sequence in The value at time; express The predicted voltage fluctuation rate is derived from step 3). ; To obtain from the standard normal distribution Random numbers drawn from the sample; This indicates the number of Monte Carlo simulations; in this embodiment, the value is 100 to balance computational accuracy and efficiency.

[0178] 4-2) Using the results of step 4-1), generate the voltage prediction value.

[0179] In this embodiment, the random impact is superimposed on the voltage base value prediction obtained in step 2) to obtain multiple possible voltage prediction paths, and then the final prediction value is obtained by averaging.

[0180]

[0181]

[0182] in, Indicates the m-th simulation in The predicted voltage value at that moment; express The voltage base value prediction at time t is derived from step 2). ; express The final voltage prediction at time t is the average of all simulation paths.

[0183] Finally, the voltage prediction sequence for the next 30 minutes was obtained. This sequence contains both deterministic trend information and the stochastic characteristics of external fluctuations, which significantly improves the robustness and adaptability of the control system compared to single base value prediction.

[0184] 5) Based on the results of step 4), calculate the dynamic voltage safety domain for the current control cycle.

[0185] Traditional voltage control uses fixed safety boundaries, which are difficult to adapt to dynamic voltage changes caused by wind power fluctuations. This embodiment innovatively proposes a dynamic voltage limit calculation method to dynamically generate safety boundaries for each moment within the prediction period, achieving refined control; the specific steps are as follows:

[0186] 5-1) Using the voltage prediction sequence obtained in step 4), calculate the maximum continuous increase and the maximum continuous decrease that may occur at each moment in the prediction period.

[0187] In this embodiment, a 30-minute voltage prediction sequence is used. Calculate the continuous variation amplitude at each time point using the following steps:

[0188] 5-1-1) Initially, let the current time be... Set threshold ;

[0189] 5-1-2) Order , This is a temporary variable used to record the cumulative voltage change during the current continuous change process.

[0190] 5-1-3) Judgment:

[0191] If satisfied Then proceed to step 5-1-4) to determine the continuous upward trend of voltage from time t.

[0192] If satisfied Then proceed to step 5-1-5) to determine the continuous downward trend of voltage starting from time t;

[0193] If none of these conditions are met, proceed to step 5-1-6.

[0194] 5-1-4) Determining the continuous upward trend of voltage starting from time t; the specific steps are as follows:

[0195] 5-1-4-1) Judgment: If ,in, Equal to the length of the voltage prediction sequence, in this embodiment, =30, then proceed to step 5-1-4-2); otherwise, proceed to step 5-1-4-3).

[0196] 5-1-4-2) If Then let Then let (Return to step 5-1-4-1); otherwise, record. Then proceed to steps 5-1-6). This represents the maximum continuous increase that may occur within the predicted time period, starting from time t.

[0197] 5-1-4-3) Record Then proceed to steps 5-1-6).

[0198] 5-1-5) Determining the continuous decreasing trend of voltage starting from time t; the specific steps are as follows:

[0199] 5-1-5-1) Judgment: If If yes, proceed to step 5-1-5-2; otherwise, proceed to step 5-1-5-3.

[0200] 5-1-5-2) If Then let Then let (Return to step 5-1-5-1); otherwise, record. Then proceed to steps 1-6). This represents the maximum continuous decrease that may occur within the predicted time period, starting from time t.

[0201] 5-1-5-3) Record Then proceed to steps 5-1-6). This represents the maximum continuous decrease that may occur within the predicted time period, starting from time t.

[0202] 5-1-6) Let t = t + 1, determine:

[0203] like If the result is positive, return to step 5-1-2 and continue the judgment; otherwise, the judgment ends, and the result corresponding to each time t is output. and .

[0204] For each time t corresponding to the voltage prediction sequence, if the corresponding time is not generated after the determination is completed... or Then, the one that was not generated at that moment and All are recorded as 0. That is, if the voltage value at a certain time t does not satisfy the criteria for a continuous voltage increase trend or a continuous voltage decrease trend, then the voltage at that time is... and All are 0; if the voltage value at a certain time t is non-zero, it is determined that... or If the magnitude of the other trend at that moment is 0, then the magnitude of the other trend is 0.

[0205] 5-2) Based on the results of step 5-1), independently calculate the dynamic safety boundary for each time t:

[0206]

[0207]

[0208] in,

[0209] This represents the voltage safety limit dynamically calculated at time t, which is adjusted in real time according to the predicted trend;

[0210] This represents the voltage safety lower limit dynamically calculated at time t, which is adjusted in real time according to the predicted trend;

[0211] This indicates the planned voltage ceiling pre-set by the power grid dispatching department, provided by the operations department;

[0212] This indicates the planned lower voltage limit pre-set by the power grid dispatching department, provided by the operations department;

[0213] This represents the safety factor, with a value range of [0.7, 0.9] and a default value of 0.8. The larger the value, the greater the safety margin, but the less control flexibility.

[0214] 6) Embed the dynamic voltage safety domain obtained in step 5) as a core constraint into the automatic voltage control system. During the control strategy optimization process, solve for the constraint... The optimal control scheme ensures that the voltage always operates within a safe range.

[0215] The method described in this embodiment achieves preventative control before grid voltage fluctuations occur. By adjusting reactive power compensation equipment and transformer tap changes in advance, sufficient voltage safety margin is proactively reserved. Practical grid applications show that this strategy significantly reduces the risk of voltage exceeding limits, decreases frequent operation of control equipment, and improves the system's adaptability to load fluctuations and changes in renewable energy output, providing effective protection for the safe and economical operation of the power grid.

Claims

1. An automatic voltage prevention control method based on voltage fluctuation rate prediction, characterized in that, include: At the moment the current control cycle arrives, the voltage sequence data of the target area bus over the past period is acquired and corrected to obtain the corrected voltage sequence; Based on the corrected voltage sequence, an ARIMA model for predicting the voltage baseline is fitted, and the fitted ARIMA model is used to generate a voltage baseline prediction sequence for future periods. Based on the corrected voltage sequence and the fitted ARIMA model, a voltage data residual sequence for the past period is generated, thereby fitting a GARCH model for analyzing voltage fluctuation characteristics. Then, the fitted GARCH model is used to generate a voltage fluctuation prediction sequence for the future period. Based on the voltage base prediction sequence and the voltage volatility prediction sequence, the final voltage prediction sequence for the future period is generated, thereby calculating the dynamic voltage safety domain of the current control cycle. The dynamic voltage safety domain is embedded as a constraint in the automatic voltage control system to achieve automatic voltage prevention control of the target area.

2. The method according to claim 1, characterized in that, The obtained corrected voltage sequence includes: 1) At the moment the current control cycle arrives Obtain the original voltage sequence of the target area bus over the previous 48 hours: in, This represents the magnitude of the bus voltage at time t, where t = -2880, -2879, ..., ; 2) Correct the data in the original voltage sequence obtained in step 1) to obtain the corrected voltage sequence; Among them, outliers in the original voltage sequence are identified by judging all the data in the original voltage sequence; If the data in the original voltage sequence is not an outlier, then let the corrected voltage data at that moment be the original data, that is, let = , express The voltage value after real-time correction; If the data in the original voltage sequence are outliers, cubic spline interpolation is used to correct them. in, Indicates the size of the time step for voltage correction; Let be the weighting coefficient, satisfying ;like Then let ; After all data in the original voltage sequence has been evaluated and corrected, the corrected voltage sequence is obtained. .

3. The method according to claim 2, characterized in that, The method for identifying outliers in the original voltage sequence is as follows: like Then determine It is an outlier; otherwise, Not an outlier; in, For Center time, window size Local average voltage over a window of minutes. For Center time, window size Voltage standard deviation for a window of minutes; This is the sensitivity adjustment coefficient.

4. The method according to claim 2, characterized in that, The step of generating a voltage baseline prediction sequence for future time periods using the fitted ARIMA model includes: 1) Obtain the corrected voltage sequence The difference order required for quiescence is denoted as d; 2) Based on the results of step 1), establish the ARIMA(p,d,q) model, with the following expression: Where B is the shift operator, ; It is a p-order autoregressive polynomial. , These are p-order autoregressive coefficients, representing the voltage value at the current moment. The degree of dependence on the voltage value at the previous i time points; This means shifting the time series forward by i units, i.e. ; It is a q-order moving average polynomial. , These are the q-order moving average coefficients, representing the error term at the current time step. The impact on the error term at the first j time steps; It is a white noise sequence with zero mean. Homoscedasticity ; denoted as the variance of the white noise sequence; 3) Determine the model parameters (p, d, q) by minimizing the AIC criterion: in, For based on Calculate the maximum likelihood estimate; 4) Based on The data in the dataset were used to estimate the parameters using the conditional least squares method. and ; 5) Based on the results of step 4), predict the future. to The voltage at a given time is recursively predicted to obtain the voltage baseline prediction for the next 30 minutes, thus generating a voltage baseline prediction sequence for the next 30 minutes. ; in, in, express Prediction of the base voltage value at time; This represents the historical or predicted voltage value at time t; when hour, Take the voltage correction value at time t ;when hour, Predicting the base voltage value at time t ; Represents the residual observed or predicted value at time t; when hour, Take the actual residual value at time t. ;when hour, Take the predicted residual value at time t. .

5. The method according to claim 4, characterized in that, The obtained corrected voltage sequence The difference order required for quiescence includes: The ADF unit root test is used to determine the stationarity of a voltage sequence. The input parameter of the ADF test is the current voltage sequence, and the output result is the P value corresponding to that sequence. Then, the P-value is determined: if the P-value is less than the preset fluctuation threshold, the current sequence is considered stable, and the current sequence is recorded relative to the corrected voltage sequence. The difference order d; if If the system passes the stationarity test, then the difference order d = 0. If the P-value is greater than or equal to the preset fluctuation threshold, it indicates that the current sequence is non-stationary, and a first-order difference is performed on the current sequence. Then, the sequence obtained after the first-order difference is used as the new current sequence and a stationarity test is performed. If the new current sequence passes the test, the current sequence relative to the corrected voltage sequence is recorded. The difference order d is determined; if the new current sequence fails the test, the current sequence is further differencing by first order until the final current sequence passes the test, and the final difference order d is recorded.

6. The method according to claim 4, characterized in that, The step of generating the voltage volatility prediction sequence for the future period using the fitted GARCH model includes: 1) Using the fitted ARIMA model, generate the target region busbar from... 48 hours ago Predicting the base voltage value within a given time period; 2) Based on the results of step 1), extract the residual sequence; in, This represents the residual between the voltage correction value at time t and the predicted voltage baseline value; 3) Establish the GARCH(1,1) model: in, This represents the conditional variance at time t. Indicates the level of basic volatility. ; This represents the influence coefficient of the new disturbance. ; Indicates the historical volatility persistence coefficient. ; ; based on The data in the dataset were used to determine the parameters of the GARCH model using the maximum likelihood estimation method. , , The estimated values ​​are denoted as follows: , , ; 4) Based on the results of step 3), recursively predict the voltage fluctuation rate for the next 30 minutes; in, express The conditional variance of the time-based prediction is the squared volatility. hour, Use actual residuals ; hour, Set to 0; Finally, a voltage fluctuation prediction sequence for the next 30 minutes is obtained. .

7. The method according to claim 6, characterized in that, The generation of the final voltage prediction sequence for the future time period includes: 1) Based on the voltage fluctuation rate prediction sequence, generate random shock values ​​and form a random shock sequence; Among them, the Monte Carlo simulation method is used to generate M sets of random shock sequences to simulate the random effects of external fluctuations on voltage: in, Indicates the m-th random shock sequence in The value at time; express Predict the voltage fluctuation rate at any time; To obtain from the standard normal distribution A random number drawn from the sample; Indicates the number of Monte Carlo simulations; 2) Using the results of step 1), generate the predicted voltage value; in, Indicates the m-th simulation in The predicted voltage value at that moment; express The final predicted voltage value at time; Finally, the voltage prediction sequence for the next 30 minutes was obtained. .

8. The method according to claim 7, characterized in that, The calculation of the dynamic voltage safety domain for the current control cycle includes: 1) Calculate the maximum consecutive increase and the maximum consecutive decrease that may occur at each moment in the voltage prediction sequence within the prediction period; 2) Based on the results of step 1), independently calculate the dynamic safety boundary for each time t: in, This represents the dynamically calculated upper limit of voltage safety at time t; This represents the dynamically calculated lower voltage safety limit at time t; This indicates the pre-set planned voltage limit; This indicates the pre-set lower limit of the planned voltage; This represents the maximum continuous increase that may occur within the predicted time period, starting from time t. This represents the maximum consecutive decrease that may occur within the predicted time period, starting from time t. This represents the safety factor.

9. The method according to claim 8, characterized in that, The maximum consecutive increase and maximum consecutive decrease that may occur at each moment corresponding to the voltage prediction sequence within the prediction period include: 1-1) Initially, let the current time be... Set a threshold; 1-2) Order , This is a temporary variable used to record the cumulative voltage change during the current continuous change process; 1-3) Judgment: If satisfied If so, proceed to steps 1-4) to determine the continuous upward trend of voltage starting from time t; If satisfied If so, proceed to steps 1-5) to determine the continuous downward trend of voltage starting from time t; If none of these conditions are met, proceed to steps 1-6). 5-1-4) Determining the continuous upward trend of voltage starting from time t; the specific steps are as follows: 1-4-1) Judgment: If ,in, If the length of the voltage prediction sequence is equal to the length of the voltage prediction sequence, proceed to step 1-4-2; otherwise, proceed to step 1-4-3. 1-4-2) If Then let Then let (Return to step 1-4-1); otherwise, record. Then proceed to steps 1-6). This represents the maximum continuous increase that may occur within the predicted time period, starting from time t. 1-4-3) Record Then proceed to steps 1-6). 1-5) Determining the continuous decreasing trend of voltage starting from time t; the specific steps are as follows: 1-5-1) Judgment: If If yes, proceed to step 1-5-2); otherwise, proceed to step 1-5-3). 1-5-2) If Then let Then let (Return to step 1-5-1); otherwise, record. Then proceed to steps 1-6). This represents the maximum consecutive decrease that may occur within the predicted time period, starting from time t. 1-5-3) Record Then proceed to steps 1-6). 1-6) Let t = t + 1, determine: like If the result is positive, return to steps 1-2) and continue the judgment; otherwise, the judgment ends, and the result corresponding to each time t is output. and ; where, for each time t corresponding to the voltage prediction sequence, if the corresponding time is not generated after the determination is completed. or Then, the one that was not generated at that moment and All are recorded as 0.

Images