New energy grid-connected monitoring method and system based on multi-modal adaptive control

Abstract

The application discloses a new energy grid-connected monitoring method and system based on multi-modal adaptive control and belongs to the technical field of new energy grid connection. Through collecting historical power fluctuation rate data and setting a safety threshold by quantile statistics, the out-of-limit gap and fluctuation deviation of historical out-of-limit alarm events are recorded. A linear prediction function is established based on a feature point set, and a recursive least square algorithm is used to identify and update the model parameters online. Electrical operation data is collected by multiple monitoring terminals arranged at the grid connection point, the deviation of the preliminary prediction value of each terminal is corrected, and the prediction consistency score is calculated. The overall distribution characteristics of the score are constructed, the prediction dependence credibility and normalized weight of each terminal are calculated, the final prediction interval length is obtained through weighted fusion, and is output to the grid connection scheduling platform. The application can realize early warning of power out-of-limit events in the new energy grid connection area, provide sufficient advance response time for dispatchers, and reduce the risk of equipment disconnection caused by out-of-limit.

Description

Technical Field

[0001] This invention relates to the field of new energy grid connection technology, specifically to a new energy grid connection monitoring method and system based on multimodal adaptive control. Background Technology

[0002] With the continuous and rapid growth of installed capacity of new energy power generation such as wind power and photovoltaics, a high proportion of new energy connected to the grid has become the norm for the power system. New energy power generation exhibits significant intermittency, volatility, and uncertainty, and the frequent fluctuations in its grid-connected power pose a severe challenge to the safe and stable operation of the power grid. Studies show that photovoltaic output power is drastically affected by natural conditions such as light intensity and temperature; rapid cloud movement can cause sudden power changes in a short period, with a change rate exceeding 50% of the rated power. Wind power output is similarly affected by wind speed variations and exhibits strong randomness. When the grid-connected power volatility exceeds a certain threshold, it may trigger a chain reaction such as power quality deterioration, equipment disconnection, or even system instability. Therefore, real-time monitoring and early warning of grid-connected power volatility have become a key technical requirement for ensuring the safe operation of new energy grid-connected systems.

[0003] Currently, monitoring of new energy grid connection mainly relies on threshold-triggered monitoring using single-type data acquisition devices deployed at the grid connection point. However, existing technical solutions have the following prominent problems:

[0004] The sampling frequency, measurement accuracy and data format of different monitoring terminals (such as power quality monitoring devices, power measurement and control terminals and electrical data acquisition terminals) vary significantly. There is a lack of an effective multi-source data fusion mechanism, and the data of each terminal forms information silos, making it difficult to form a unified and comprehensive perception of the operating status of the grid-connected area. Measurement errors or communication anomalies of a single data source may lead to false alarms or missed alarms.

[0005] Existing forecasts mostly use fixed parameter models, but the output of new energy sources has significant time-varying characteristics as meteorological conditions (sunlight, wind speed, temperature, etc.) and operating conditions change. The prediction accuracy of fixed parameter models gradually decreases in long-term operation, and they cannot achieve continuous tracking and adaptive matching of power fluctuation characteristics.

[0006] Existing strategies are mostly based on simple threshold judgments using a single data source. Alarms are only triggered when the power fluctuation rate reaches the threshold. Warning signals often lag behind the actual occurrence of power exceeding the limit. Dispatchers cannot obtain sufficient advance response time to coordinate energy storage regulation or adjust the power generation output setpoint, resulting in the system being in a passive response state of "dealing with the limit only after it occurs" for a long time. Summary of the Invention

[0007] The purpose of this invention is to provide a new energy grid-connected monitoring method based on multimodal adaptive control to solve the problems mentioned in the background art.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0009] A new energy grid-connected monitoring system based on multimodal adaptive control, comprising:

[0010] The historical data analysis module is used to collect historical grid-connected power data in new energy grid-connected areas and record historical over-limit alarm events and their corresponding over-limit gaps and fluctuation deviations.

[0011] The adaptive prediction module establishes and identifies the linear prediction function between the fluctuation difference and the time difference based on the feature point set of historical over-limit alarm events, and outputs the predicted time difference value of the next over-limit alarm event.

[0012] The multi-terminal correction and evaluation module is used to collect electrical operation data from multiple monitoring terminals deployed in the new energy grid-connected area, correct the deviation of the initial prediction time difference of each monitoring terminal, and calculate the prediction consistency score of each monitoring terminal.

[0013] The multimodal fusion output module is used to construct the overall distribution characteristics of the prediction consistency scores of each monitoring terminal, calculate the prediction dependency credibility and normalized credibility weight of each monitoring terminal, obtain the final prediction interval duration through weighted fusion, and output it to the grid-connected scheduling platform.

[0014] Preferably, the historical data analysis module includes:

[0015] The power volatility safety threshold construction unit is used to collect grid-connected power time series data of new energy grid-connected areas during long-term historical operation periods, calculate the power volatility observation value in each window according to the preset time window, form a historical sample set of power volatility, sort the sample set in ascending order, and use the sample quantile corresponding to the preset confidence level as the power volatility safety threshold R.

[0016] The over-limit alarm event recording unit is used to feed back a power fluctuation over-limit alarm signal and record it as an over-limit alarm event when the power fluctuation rate touches or exceeds the power fluctuation rate safety threshold in the new energy grid-connected area. Based on the over-limit alarm events, an over-limit gap cluster is formed, and the fluctuation difference between the actual power fluctuation rate and the power fluctuation rate safety threshold is recorded when each over-limit alarm event occurs, forming a fluctuation deviation recording cluster.

[0017] Preferably, the adaptive prediction module includes:

[0018] The feature point construction unit is used to pair the time difference and fluctuation difference of the same over-limit alarm event to construct a set of feature points that characterize the disturbance characteristics of the grid-connected area.

[0019] The linear prediction function establishment unit is used to establish a linear prediction function describing the relationship between fluctuation difference and time difference using the set of feature points as training samples.

[0020] The recursive least squares online identification unit is used to recursively correct the model parameters using the predicted deviation of the time difference as a feedback signal. This includes model parameter initialization, confidence matrix initialization, recursive update, and prediction output, which outputs the predicted value of the time difference between the next time an over-limit alarm event occurs and the current time.

[0021] Preferably, the multi-terminal correction and evaluation module includes:

[0022] The diversified data acquisition unit, based on diversified monitoring terminals deployed in the new energy grid-connected area, collects electrical operation data of the grid-connected point according to the sampling frequency of each monitoring terminal. The monitoring terminals include power quality monitoring terminals, grid-connected power measurement and control terminals, and electrical data acquisition terminals deployed at the grid-connected point.

[0023] The monitoring data cluster creation unit is used to create an independent monitoring data cluster for each monitoring terminal in the same time period and the same grid-connected area, and a total of N monitoring data clusters are created. The monitoring data clusters are used to record the grid-connected point voltage data, grid-connected current data and grid-connected power data collected by the corresponding monitoring terminal.

[0024] The preliminary prediction unit is used to extract the grid-connected power time series from the nth monitoring data cluster, calculate the power fluctuation rate sequence of the grid-connected power time series according to a preset time window, substitute each power fluctuation rate data in the power fluctuation rate sequence as the measured power fluctuation rate value into the linear prediction function, and deduplicate and organize the predicted value of the time difference output by the linear prediction function to obtain the time difference prediction value sequence for preliminary prediction based on the nth monitoring data cluster.

[0025] The deviation correction unit calculates the deviation correction amount based on the deviation sequence between the actual and predicted time difference values ​​recorded by each monitoring terminal in the most recent M historical prediction weeks, and uses the deviation correction amount and the preliminary predicted time difference value sequence to correct the budgeted value of the time difference, thus obtaining the corrected time difference sequence.

[0026] The consistency score calculation unit is used to calculate the predicted consistency score of the nth monitoring terminal in the current time period based on the corrected time difference sequence.

[0027] Preferably, the multimodal fusion output module includes:

[0028] The prediction dependency credibility calculation unit is used to construct the overall distribution characteristics of the prediction consistency scores of each monitoring terminal, and to use the probability density of the temporal consistency scores of each monitoring terminal relative to the overall distribution characteristics as the prediction dependency credibility of the monitoring terminal in the current time period.

[0029] The normalized confidence weight calculation unit calculates the normalized confidence weight of each monitoring terminal based on the prediction dependency confidence calculation.

[0030] The weighted fusion output unit is used to calculate the weighted average prediction time difference of each monitoring terminal and output the weighted average prediction time difference as the final prediction interval duration to the grid-connected scheduling platform.

[0031] A new energy grid-connected monitoring method based on multimodal adaptive control, comprising the following steps:

[0032] Step S1: Collect historical grid-connected power data of the new energy grid-connected area, and record historical over-limit alarm events and their corresponding over-limit gaps and fluctuation deviations;

[0033] Step S2: Based on the feature point set of historical over-limit alarm events, establish and identify the linear prediction function between the fluctuation difference and the time difference online, and output the predicted time difference value of the next over-limit alarm event;

[0034] Step S3: Collect electrical operation data using multiple monitoring terminals deployed in the new energy grid-connected area, correct the deviation of the preliminary prediction time difference of each monitoring terminal, and calculate the prediction consistency score of each monitoring terminal.

[0035] Step S4: Construct the overall distribution characteristics of the prediction consistency scores of each monitoring terminal, calculate the prediction dependency credibility and normalized credibility weight of each monitoring terminal, obtain the final prediction interval duration through weighted fusion, and output it to the grid-connected scheduling platform.

[0036] Preferably, the specific implementation process of step S1 includes:

[0037] Time-series data of grid-connected power in renewable energy grid-connected areas are collected over long-term historical operating periods. Power volatility observations are calculated segment by segment according to preset time windows, forming a historical power volatility sample set. This sample set is then sorted in ascending order, and the sample quantile corresponding to a preset confidence level is used as the power volatility safety threshold, denoted as [missing information]. The power fluctuation rate is used to measure the rate of change of grid-connected power relative to the previous moment within a unit time window.

[0038] If the power fluctuation rate in the new energy grid-connected area has previously reached or exceeded the power fluctuation rate safety threshold, a power fluctuation over-limit alarm signal will be fed back and recorded as an over-limit alarm event. If the power fluctuation rate in the new energy grid-connected area has previously not reached or exceeded the power fluctuation rate safety threshold, no power fluctuation over-limit alarm signal will be generated.

[0039] Based on the over-limit alarm events, an over-limit gap cluster is formed. ,in, For the first The time of the first over-limit alarm and the first The time difference between each over-limit alarm. This represents the total number of historical out-of-limit alarm events.

[0040] Whenever an over-limit alarm signal is triggered, record the measured power fluctuation rate of the grid connection point at the time of the over-limit alarm relative to the safe power fluctuation rate threshold. The excess amplitude forms a cluster of fluctuation deviation records. ,in, Indicates the first When the over-limit alarm event occurs, the actual value of power fluctuation rate is different from the power fluctuation rate safety threshold. The fluctuation difference between them.

[0041] Preferably, the specific implementation process of step S2 includes:

[0042] The time difference belonging to the same over-limit alarm event Difference from volatility Pairing is performed to construct a set of feature points characterizing the disturbance characteristics of the grid-connected area. ;

[0043] Using the set of feature points as training samples, a linear prediction function is established to describe the correlation between fluctuation difference and time difference. In the formula, The input variable for the linear prediction function is the fluctuation difference. The output variable of the linear prediction function corresponds to the time difference. and These are the first model parameters and the second model parameters to be identified online, respectively.

[0044] The recursive least squares algorithm is used to identify and update the model parameters online, including:

[0045] Predicted deviation based on time difference As a feedback signal, the model parameters are recursively corrected:

[0046] Initialization of model parameters:

[0047] Let the initialized model parameter vector be ,in It is the transpose symbol. and Let these be the initialization parameters of the first and second models, respectively; let the initialization confidence matrix be... ,in, It is a second-order identity matrix. The preset positive number and The value is greater than the variance estimate of the historical power volatility data in the renewable energy grid-connected area;

[0048] Recursive update:

[0049] Let the total number of feature point samples collected for historical out-of-limit alarm events be . The currently established feature points The first model parameters and the second model parameters are combined to form a vector form. The current trust matrix is ​​denoted as ;

[0050] Get in the first The fluctuation difference and time difference at the time of the first over-limit alarm event constitute the first Measurement vector when the over-limit alarm event occurs , where the value 1 represents the reference constant of the time difference;

[0051] Using the currently established model parameter vector For the The time difference of each over-limit alarm event is predicted to obtain the predicted value of the time difference. Calculate the predicted deviation of the time difference ;

[0052] Update the trust matrix and model parameter vector. In the formula, For the first The two-dimensional adaptive correction gain vector under the second over-limit alarm event, wherein the first component of the two-dimensional adaptive correction gain vector corresponds to the first model parameter. The corrected weighting value is used to adjust the prediction bias. For the first model parameters The correction amount, the second component of the two-dimensional adaptive correction gain vector corresponds to the second model parameters. The corrected weighting value is used to adjust the prediction bias. For the second model parameters The correction amount; To utilize the first The confidence matrix updated from feature point samples after the first over-limit alarm event. This is the updated model parameter vector;

[0053] Predicted output:

[0054] Obtain the measured power fluctuation rate at the current grid connection point. Current fluctuation difference Substituting these values ​​into the linear prediction function under the updated model parameter vector, we obtain the predicted value of the time difference between the next over-limit alarm event and the current time. .

[0055] Preferably, the specific implementation process of step S3 includes:

[0056] Based on diversified monitoring terminals deployed in the new energy grid connection area, electrical operation data of the grid connection point are collected according to the sampling frequency of each monitoring terminal. The monitoring terminals include power quality monitoring terminals, grid-connected power measurement and control terminals and electrical data acquisition terminals deployed at the grid connection point.

[0057] Within the same time period and the same grid connection area, an independent monitoring data cluster was created for each monitoring terminal, resulting in a total of [number missing] monitoring data clusters. A monitoring data cluster, wherein the monitoring data cluster is used to record the grid connection point voltage data, grid connection current data and grid connection power data collected by the corresponding monitoring terminal;

[0058] From the Extract the grid-connected power time series from each monitoring data cluster, calculate the power volatility series of the grid-connected power time series according to a preset time window, and use each power volatility data in the power volatility series as the measured value of power volatility. Substituting these values ​​into the linear prediction function and then removing duplicates from the predicted time difference output by the linear prediction function, we obtain the result with the first... The time difference predicted value sequence for preliminary prediction of each monitoring data cluster is denoted as... ;

[0059] Based on the actual time difference values ​​recorded by each monitoring terminal within the most recent M historical prediction weeks. Compared with the predicted value Deviation sequence between ,in Calculation deviation correction amount ;

[0060] Using deviation correction amount Time difference predicted value sequence with preliminary prediction The budgeted time difference is adjusted to obtain the adjusted time difference series. ;

[0061] The corrected time difference series, accounting for the first The prediction consistency score of each monitoring terminal in the current time period ,in Time difference series variance Time difference series The average value and .

[0062] Preferably, the specific implementation process of step S4 includes:

[0063] Construct the overall distribution characteristics of the prediction consistency scores of each monitoring terminal, and use the probability density of the temporal consistency scores of each monitoring terminal relative to the overall distribution characteristics as the prediction dependency confidence of the monitoring terminal in the current time period. ,in, The mean of the prediction consistency scores for all monitoring terminals. The variance of the prediction consistency score for all monitoring terminals;

[0064] Based on the prediction dependency confidence level, the normalized confidence weight of each monitoring terminal is calculated. To assess the weighted average prediction time difference of each monitoring terminal. The weighted average prediction time difference is output as the final prediction interval duration to the grid-connected scheduling platform.

[0065] The beneficial effects achieved by this invention are:

[0066] By setting safety thresholds based on quantile statistics of historical power fluctuation data, the subjective defects of traditional manual threshold setting are overcome, making the threshold setting have clear statistical basis and scenario adaptability, and can flexibly adapt to the differentiated monitoring needs of different new energy types (wind power, photovoltaic), different grid connection voltage levels and different installed capacity areas.

[0067] The recursive least squares algorithm is introduced to identify and update the prediction model parameters online, so that the model parameters can be recursively corrected with new over-limit event data. This enables continuous tracking and adaptive matching of the time-varying characteristics of grid-connected power fluctuations. Compared with traditional batch least squares, the recursive least squares algorithm does not need to store all historical data, has low computational load and low storage overhead, and is particularly suitable for online real-time deployment scenarios. It overcomes the defect of decreased prediction accuracy after long-term operation of fixed parameter models.

[0068] By creating independent data clusters for the power quality monitoring terminals, grid-connected power measurement and control terminals, and electrical data acquisition terminals deployed at the grid connection point, and generating prediction sequences independently for each terminal, parallel acquisition and independent processing of multi-source data are realized. Then, the prediction sequences of each terminal are corrected by the historical residual mean of the sliding window, and the prediction consistency score of the corrected sequence of each terminal is calculated. This can effectively identify and suppress the negative impact of abnormal data sources, and improve the data quality and system robustness used for final decision-making.

[0069] Using the probability density of the prediction consistency scores of each terminal in the overall distribution as the basis for credibility assessment, the fusion weight of each terminal is dynamically allocated through normalization processing, so that terminals with high prediction consistency occupy a larger proportion in the fusion result. The entire process does not require manual preset of balance parameters or experience coefficients, and is completely determined by data-driven adaptive determination, giving the system the ability to adapt to different operating conditions (such as the morning and evening photovoltaic output ramp-up period, the nighttime wind power output peak period, etc.). Through weighted fusion, multi-source information is integrated into a single prediction result, effectively reducing the false alarm rate and missed alarm rate of over-limit warnings, providing dispatchers with sufficient advance response time to coordinate energy storage charging and discharging or adjust the active power output of new energy stations, and ensuring the safe and stable operation of the power grid. Attached Figure Description

[0070] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0071] Figure 1 This is a schematic diagram illustrating the steps of the new energy grid-connected monitoring method based on multimodal adaptive control according to the present invention. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] In this first embodiment: a new energy grid-connected monitoring system based on multimodal adaptive control is provided, the system comprising:

[0074] The historical data analysis module is used to collect historical grid-connected power data in new energy grid-connected areas and record historical over-limit alarm events and their corresponding over-limit gaps and fluctuation deviations.

[0075] Specifically, the historical data analysis module includes:

[0076] The power volatility safety threshold construction unit is used to collect grid-connected power time series data of new energy grid-connected areas during long-term historical operation periods, calculate the power volatility observation value in each window according to the preset time window, form a historical sample set of power volatility, sort the sample set in ascending order, and use the sample quantile corresponding to the preset confidence level as the power volatility safety threshold R.

[0077] The over-limit alarm event recording unit is used to feed back power fluctuation over-limit alarm signals and record them as an over-limit alarm event when the power fluctuation rate touches or exceeds the power fluctuation rate safety threshold in the new energy grid-connected area. Based on the over-limit alarm events, an over-limit gap cluster is formed, and the fluctuation difference between the actual power fluctuation rate and the power fluctuation rate safety threshold is recorded when each over-limit alarm event occurs, forming a fluctuation deviation recording cluster.

[0078] The adaptive prediction module establishes and identifies the linear prediction function between the fluctuation difference and the time difference based on the feature point set of historical over-limit alarm events, and outputs the predicted time difference value of the next over-limit alarm event.

[0079] Specifically, the adaptive prediction module includes:

[0080] The feature point construction unit is used to pair the time difference and fluctuation difference of the same over-limit alarm event to construct a set of feature points that characterize the disturbance characteristics of the grid-connected area.

[0081] The linear prediction function building unit is used to build a linear prediction function that describes the relationship between fluctuation difference and time difference, using the set of feature points as training samples.

[0082] The recursive least squares online identification unit is used to recursively correct the model parameters using the predicted deviation of the time difference as a feedback signal. This includes model parameter initialization, confidence matrix initialization, recursive update, and prediction output, which outputs the predicted value of the time difference between the next time an over-limit alarm event occurs and the current time.

[0083] The multi-terminal correction and evaluation module is used to collect electrical operation data from multiple monitoring terminals deployed in the new energy grid-connected area, correct the deviation of the initial prediction time difference of each monitoring terminal, and calculate the prediction consistency score of each monitoring terminal.

[0084] Specifically, the multi-terminal correction and evaluation module includes:

[0085] The diversified data acquisition unit, based on diversified monitoring terminals deployed in the new energy grid-connected area, collects electrical operation data of the grid-connected point according to the sampling frequency of each monitoring terminal. The monitoring terminals include power quality monitoring terminals, grid-connected power measurement and control terminals and electrical data acquisition terminals deployed at the grid-connected point.

[0086] The monitoring data cluster creation unit is used to create an independent monitoring data cluster for each monitoring terminal in the same time period and the same grid-connected area, creating a total of N monitoring data clusters. The monitoring data clusters are used to record the grid-connected point voltage data, grid-connected current data and grid-connected power data collected by the corresponding monitoring terminal.

[0087] The preliminary prediction unit is used to extract the grid-connected power time series from the nth monitoring data cluster, calculate the power fluctuation rate sequence of the grid-connected power time series according to the preset time window, substitute each power fluctuation rate data in the power fluctuation rate sequence as the measured power fluctuation rate value into the linear prediction function, and deduplicate and sort the predicted value of the time difference output by the linear prediction function to obtain the time difference prediction value sequence of the preliminary prediction based on the nth monitoring data cluster.

[0088] The deviation correction unit calculates the deviation correction amount based on the deviation sequence between the actual and predicted time difference values ​​recorded by each monitoring terminal in the most recent M historical prediction weeks, and uses the deviation correction amount and the preliminary predicted time difference value sequence to correct the budgeted value of the time difference, thus obtaining the corrected time difference sequence.

[0089] The consistency score calculation unit is used to calculate the predicted consistency score of the nth monitoring terminal in the current time period based on the corrected time difference sequence.

[0090] The multimodal fusion output module is used to construct the overall distribution characteristics of the prediction consistency score of each monitoring terminal, calculate the prediction dependency credibility and normalized credibility weight of each monitoring terminal, obtain the final prediction interval duration through weighted fusion, and output it to the grid-connected scheduling platform.

[0091] Specifically, the multimodal fusion output module includes:

[0092] The prediction dependency credibility calculation unit is used to construct the overall distribution characteristics of the prediction consistency scores of each monitoring terminal, and to use the probability density of the temporal consistency scores of each monitoring terminal relative to the overall distribution characteristics as the prediction dependency credibility of the monitoring terminal in the current time period.

[0093] The normalized confidence weight calculation unit calculates the normalized confidence weight of each monitoring terminal based on the prediction dependency confidence calculation.

[0094] The weighted fusion output unit is used to calculate the weighted average prediction time difference of each monitoring terminal and output the weighted average prediction time difference as the final prediction interval to the grid-connected scheduling platform.

[0095] Please see Figure 1 In this second embodiment: a new energy grid-connected monitoring method based on multimodal adaptive control is provided, which includes the following steps:

[0096] Step S1: Collect historical grid-connected power data of the new energy grid-connected area, and record historical over-limit alarm events and their corresponding over-limit gaps and fluctuation deviations;

[0097] For example, time-series data of grid-connected power in new energy grid-connected areas are collected during long-term historical operating periods. Power volatility observations are calculated segment by segment according to preset time windows, forming a historical power volatility sample set. This sample set is then sorted in ascending order, and the sample quantile corresponding to a preset confidence level is used as the power volatility safety threshold, denoted as... Power fluctuation rate is used to measure the rate of change of grid-connected power within a unit time window relative to the previous moment.

[0098] If the power fluctuation rate in the new energy grid-connected area has previously reached or exceeded the power fluctuation rate safety threshold, a power fluctuation over-limit alarm signal will be fed back and recorded as an over-limit alarm event. If the power fluctuation rate in the new energy grid-connected area has previously not reached or exceeded the power fluctuation rate safety threshold, no power fluctuation over-limit alarm signal will be generated.

[0099] Based on the over-limit alarm events, an over-limit gap cluster is formed. ,in, For the first The time of the first over-limit alarm and the first The time difference between each over-limit alarm. This represents the total number of historical out-of-limit alarm events.

[0100] Whenever an over-limit alarm signal is triggered, record the measured power fluctuation rate of the grid connection point at the time of the over-limit alarm relative to the safe power fluctuation rate threshold. The excess amplitude forms a cluster of fluctuation deviation records. ,in, Indicates the first When the over-limit alarm event occurs, the actual value of power fluctuation rate is different from the power fluctuation rate safety threshold. The fluctuation difference between them;

[0101] It should be noted that, considering that the window size affects the calculation results, the smaller the window, the more sensitive it is to capturing instantaneous fluctuations, while the larger the window, the more it reflects the changing trend over a longer time scale. Therefore, the observed power volatility can be obtained by the formula: observed power volatility = (maximum power value within the window, minimum power value within the window) / (2 × average power value within the window) × 100%, making the volatility comparable for grid-connected areas with different installed capacities or different operating power levels.

[0102] Step S2: Based on the feature point set of historical over-limit alarm events, establish and identify the linear prediction function between the fluctuation difference and the time difference online, and output the predicted time difference value of the next over-limit alarm event;

[0103] For example, the time difference belonging to the same over-limit alarm event. Difference from volatility Pairing is performed to construct a set of feature points characterizing the disturbance characteristics of the grid-connected area. ;

[0104] Using the set of feature points as training samples, a linear prediction function is established to describe the relationship between fluctuation difference and time difference. In the formula, The input variable for the linear prediction function is the fluctuation difference. The output variable of the linear prediction function corresponds to the time difference. and These are the first model parameters and the second model parameters to be identified online, respectively.

[0105] The recursive least squares algorithm is used to identify and update the model parameters online, including:

[0106] Predicted deviation based on time difference As a feedback signal, the model parameters are recursively corrected:

[0107] Initialization of model parameters:

[0108] Let the initialized model parameter vector be ,in It is the transpose symbol. and Let these be the initialization parameters of the first and second models, respectively; let the initialization confidence matrix be... ,in, It is a second-order identity matrix. The preset positive number and The value is greater than the variance estimate of the historical power volatility data in the renewable energy grid-connected area;

[0109] It should be noted that, when dealing with the confidence matrix, since it is essentially a measure of the covariance matrix of the parameter estimation error in recursive least squares, when... Taking a larger value means that the algorithm estimates the initial parameters during the startup phase. The confidence level is extremely low (i.e., the uncertainty is very high), so the first new sample data to arrive will have a significant impact on the parameter estimation. Therefore, the initial value of the confidence matrix directly affects the convergence performance of the algorithm in the early stage. The variance estimate of historical power fluctuation data reflects the inherent dispersion of power fluctuation in the grid-connected area. If the value is less than or close to this variance, then the numerical magnitude of the initial confidence matrix does not match the statistical dispersion of the data itself.

[0110] The second-order identity matrix is ​​used as the multiplicative basis of the initial confidence matrix because, in the absence of any prior information, it is assumed that the initial estimates of the two parameters have the same confidence (equal diagonal elements) and there is no prior coupling between the two parameters (zero off-diagonal elements).

[0111] Recursive update:

[0112] Let the total number of feature point samples collected for historical out-of-limit alarm events be . The currently established feature points The first model parameters and the second model parameters are combined to form a vector form. The current trust matrix is ​​denoted as ;

[0113] Get in the first The fluctuation difference and time difference at the time of the first over-limit alarm event constitute the first Measurement vector when the over-limit alarm event occurs , where the value 1 represents the reference constant of the time difference;

[0114] It should be noted that the second dimension component in the measurement vector is fixed at 1, and its function is to help identify the constant term (intercept) of the linear prediction function.

[0115] Using the currently established model parameter vector For the The time difference of each over-limit alarm event is predicted to obtain the predicted value of the time difference. Calculate the predicted deviation of the time difference ;

[0116] Update the trust matrix and model parameter vector. In the formula, For the first The two-dimensional adaptive correction gain vector under the sub-limit alarm event, the first component of the two-dimensional adaptive correction gain vector corresponds to the first model parameter. The corrected weighting value is used to adjust the prediction bias. For the first model parameters The correction amount, the second component of the two-dimensional adaptive correction gain vector corresponds to the second model parameters. The corrected weighting value is used to adjust the prediction bias. For the second model parameters The correction amount; To utilize the first The confidence matrix updated from feature point samples after the first over-limit alarm event. This is the updated model parameter vector;

[0117] Predicted output:

[0118] Obtain the measured power fluctuation rate at the current grid connection point. Current fluctuation difference Substituting these values ​​into the linear prediction function under the updated model parameter vector, we obtain the predicted value of the time difference between the next over-limit alarm event and the current time. .

[0119] Step S3: Collect electrical operation data using multiple monitoring terminals deployed in the new energy grid-connected area, correct the deviation of the preliminary prediction time difference of each monitoring terminal, and calculate the prediction consistency score of each monitoring terminal.

[0120] For example, based on the diversified monitoring terminals deployed in the new energy grid connection area, electrical operation data of the grid connection point is collected according to the sampling frequency of each monitoring terminal. The monitoring terminals include power quality monitoring terminals, grid connection power measurement and control terminals and electrical data acquisition terminals deployed at the grid connection point.

[0121] Within the same time period and the same grid connection area, an independent monitoring data cluster was created for each monitoring terminal, resulting in a total of [number missing] monitoring data clusters. Each monitoring data cluster is used to record the grid connection point voltage data, grid connection current data, and grid connection power data collected by the corresponding monitoring terminal.

[0122] From the Extract the grid-connected power time series from each monitoring data cluster, calculate the power volatility series of the grid-connected power time series according to a preset time window, and use each power volatility data point in the power volatility series as the measured value of the power volatility. Substituting these values ​​into the linear prediction function and then removing duplicates from the predicted time difference output by the linear prediction function, we obtain the result with the first... The time difference predicted value sequence for preliminary prediction of each monitoring data cluster is denoted as... ;

[0123] Based on the actual time difference values ​​recorded by each monitoring terminal within the most recent M historical prediction weeks. Compared with the predicted value Deviation sequence between ,in Calculation deviation correction amount ;

[0124] Using deviation correction amount Time difference predicted value sequence with preliminary prediction The budgeted time difference is adjusted to obtain the adjusted time difference series. ;

[0125] The corrected time difference series, accounting for the first The prediction consistency score of each monitoring terminal in the current time period ,in Time difference series variance Time difference series The average value and ;

[0126] It should be noted that the prediction consistency score is used to quantify the degree of consistency between the prediction results of the nth monitoring terminal within the same time period, and its core component is the coefficient of variation. The coefficient of variation (COP) is the ratio of the standard deviation to the mean. It is a dimensionless statistic used to measure the relative dispersion of data around the mean. The smaller the COP, the smaller the fluctuation range between the predicted values ​​of the terminal, and the more stable and consistent the prediction results are.

[0127] Step S4: Construct the overall distribution characteristics of the prediction consistency scores of each monitoring terminal, calculate the prediction dependency credibility and normalized credibility weight of each monitoring terminal, obtain the final prediction interval duration through weighted fusion, and output it to the grid-connected scheduling platform.

[0128] For example, the overall distribution characteristics of the prediction consistency scores of each monitoring terminal are constructed, and the probability density of the temporal consistency scores of each monitoring terminal relative to the overall distribution characteristics is used as the prediction dependency confidence of the monitoring terminal in the current time period. ,in, The mean of the prediction consistency scores for all monitoring terminals. The variance of the prediction consistency score for all monitoring terminals;

[0129] It should be noted that the reliability of prediction dependency is based on the relative position of each terminal's consistency score in the overall distribution, so as to reflect the consensus of most terminals; the probability density of each score value in the distribution is calculated by the normal probability density function: the closer the score is to the mean, the larger the probability density value, and the higher the "reliability of prediction dependency" of that terminal.

[0130] Based on the prediction dependency confidence level, the normalized confidence weight of each monitoring terminal is calculated. To assess the weighted average prediction time difference of each monitoring terminal. The weighted average prediction time difference is output as the final prediction interval to the grid-connected dispatch platform;

[0131] It should be noted that, based on the unbiased estimation, a weighted average of multiple independent estimates according to their accuracy (inverse of variance) can be obtained to obtain the unbiased estimate with the minimum variance. The normalized confidence weight is a relative measure of the prediction accuracy of each terminal. That is, the higher the consistency score of the terminal, the lower the dispersion of its prediction sequence and the higher the estimation accuracy, and should therefore receive a greater weight. The weighted average allows the information of high-precision terminals to occupy a larger proportion in the final result, so as to achieve optimal fusion of multi-source information. At the same time, the weighted fusion allows the prediction information of all terminals to participate in the final decision with different weights. Even if the weight of a certain terminal is low, its information is still partially retained rather than completely discarded. Since different monitoring terminals may exhibit different data quality at different times (such as the period of morning and evening irradiance variation and the period of nighttime wind power fluctuation), the weighted fusion can dynamically adapt to this change and avoid the risk of no prediction results output due to temporary anomalies of a single terminal. This strategy is particularly applicable in the monitoring of new energy grid connection.

[0132] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0133] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A new energy grid-connected monitoring method based on multimodal adaptive control, characterized in that, The method includes the following steps: Step S1: Collect historical grid-connected power data of the new energy grid-connected area, and record historical over-limit alarm events and their corresponding over-limit gaps and fluctuation deviations; Step S2: Based on the feature point set of historical over-limit alarm events, establish and identify the linear prediction function between the fluctuation difference and the time difference online, and output the predicted time difference value of the next over-limit alarm event; Step S3: Collect electrical operation data using multiple monitoring terminals deployed in the new energy grid-connected area, correct the deviation of the preliminary prediction time difference of each monitoring terminal, and calculate the prediction consistency score of each monitoring terminal. Step S4: Construct the overall distribution characteristics of the prediction consistency scores of each monitoring terminal, calculate the prediction dependency credibility and normalized credibility weight of each monitoring terminal, obtain the final prediction interval duration through weighted fusion, and output it to the grid-connected scheduling platform.

2. The new energy grid-connected monitoring method based on multimodal adaptive control according to claim 1, characterized in that, The specific implementation process of step S1 includes: Time-series data of grid-connected power in renewable energy grid-connected areas are collected over long-term historical operating periods. Power volatility observations are calculated segment by segment according to preset time windows, forming a historical power volatility sample set. This sample set is then sorted in ascending order, and the sample quantile corresponding to a preset confidence level is used as the power volatility safety threshold, denoted as [missing information]. The power fluctuation rate is used to measure the rate of change of grid-connected power within a unit time window relative to the previous moment. If the power fluctuation rate in the new energy grid-connected area has previously reached or exceeded the power fluctuation rate safety threshold, a power fluctuation over-limit alarm signal will be fed back and recorded as an over-limit alarm event. If the power fluctuation rate in the new energy grid-connected area has previously not reached or exceeded the power fluctuation rate safety threshold, no power fluctuation over-limit alarm signal will be generated. Based on the over-limit alarm events, an over-limit gap cluster is formed. ,in, For the first The time of the first over-limit alarm and the first The time difference between each over-limit alarm. This represents the total number of historical out-of-limit alarm events. Whenever an over-limit alarm signal is triggered, record the measured power fluctuation rate of the grid connection point at the time of the over-limit alarm relative to the safe power fluctuation rate threshold. The excess amplitude forms a cluster of fluctuation deviation records. ,in, Indicates the first When the over-limit alarm event occurs, the actual value of power fluctuation rate is different from the power fluctuation rate safety threshold. The fluctuation difference between them.

3. The new energy grid-connected monitoring method based on multimodal adaptive control according to claim 2, characterized in that, The specific implementation process of step S2 includes: The time difference belonging to the same over-limit alarm event Difference from volatility Pairing is performed to construct a set of feature points characterizing the disturbance characteristics of the grid-connected area. ; Using the set of feature points as training samples, a linear prediction function is established to describe the correlation between fluctuation difference and time difference. In the formula, The input variable for the linear prediction function is the fluctuation difference. The output variable of the linear prediction function corresponds to the time difference. and These are the first model parameters and the second model parameters to be identified online, respectively. The recursive least squares algorithm is used to identify and update the model parameters online, including: Predicted deviation based on time difference As a feedback signal, the model parameters are recursively corrected: Initialization of model parameters: Let the initialized model parameter vector be ,in It is the transpose symbol. and Let these be the initialization parameters of the first and second models, respectively; let the initialization confidence matrix be... ,in, It is a second-order identity matrix. The preset positive number and The value is greater than the variance estimate of the historical power volatility data in the renewable energy grid-connected area; Recursive update: Let the total number of feature point samples collected for historical out-of-limit alarm events be . The currently established feature points The first model parameters and the second model parameters are combined to form a vector form. The current trust matrix is ​​denoted as ; Get in the first The fluctuation difference and time difference at the time of the first over-limit alarm event constitute the first Measurement vector when the over-limit alarm event occurs , where the value 1 represents the reference constant of the time difference; Using the currently established model parameter vector For the first The time difference of each over-limit alarm event is predicted to obtain the predicted value of the time difference. Calculate the predicted deviation of the time difference ; Update the trust matrix and model parameter vector. In the formula, For the first The two-dimensional adaptive correction gain vector under the second over-limit alarm event, wherein the first component of the two-dimensional adaptive correction gain vector corresponds to the first model parameter. The corrected weighting value is used to adjust the prediction bias. For the first model parameters The correction amount, the second component of the two-dimensional adaptive correction gain vector corresponds to the second model parameters. The corrected weighting value is used to adjust the prediction bias. For the second model parameters The correction amount; To utilize the first The confidence matrix updated from feature point samples after the first over-limit alarm event. This is the updated model parameter vector; Predicted output: Obtain the measured power fluctuation rate at the current grid connection point. Current fluctuation difference Substituting these values ​​into the linear prediction function under the updated model parameter vector, we obtain the predicted time difference between the next over-limit alarm event and the current time. .

4. The new energy grid-connected monitoring method based on multimodal adaptive control according to claim 3, characterized in that, The specific implementation process of step S3 includes: Based on diversified monitoring terminals deployed in the new energy grid connection area, electrical operation data of the grid connection point are collected according to the sampling frequency of each monitoring terminal. The monitoring terminals include power quality monitoring terminals, grid-connected power measurement and control terminals and electrical data acquisition terminals deployed at the grid connection point. Within the same time period and the same grid connection area, an independent monitoring data cluster is created for each monitoring terminal, resulting in a total of [number missing] clusters. A monitoring data cluster, wherein the monitoring data cluster is used to record the grid connection point voltage data, grid connection current data and grid connection power data collected by the corresponding monitoring terminal; From the Extract the grid-connected power time series from each monitoring data cluster, calculate the power volatility series of the grid-connected power time series according to a preset time window, and use each power volatility data in the power volatility series as the measured value of power volatility. Substituting these values ​​into the linear prediction function and then removing duplicates from the predicted time difference output by the linear prediction function, we obtain the result with the first... The time difference predicted value sequence for preliminary prediction of each monitoring data cluster is denoted as... ; Based on the actual time difference values ​​recorded by each monitoring terminal within the most recent M historical prediction weeks. Compared with the predicted value Deviation sequence between ,in Calculation deviation correction amount ; Using deviation correction amount Time difference predicted value sequence with preliminary prediction The budgeted value of the time difference is adjusted to obtain the adjusted time difference series. ; The corrected time difference series, accounting for the first The prediction consistency score of each monitoring terminal in the current time period ,in Time difference series variance Time difference series The average value and .

5. The new energy grid-connected monitoring method based on multimodal adaptive control according to claim 4, characterized in that, The specific implementation process of step S4 includes: Construct the overall distribution characteristics of the prediction consistency scores of each monitoring terminal, and use the probability density of the temporal consistency scores of each monitoring terminal relative to the overall distribution characteristics as the prediction dependency confidence of the monitoring terminal in the current time period. ,in, The mean of the prediction consistency scores for all monitoring terminals. The variance of the prediction consistency score for all monitoring terminals; Based on the prediction dependency confidence level, the normalized confidence weight of each monitoring terminal is calculated. To assess the weighted average prediction time difference of each monitoring terminal. The weighted average prediction time difference is output as the final prediction interval duration to the grid-connected scheduling platform.

6. A new energy grid-connected monitoring system based on multimodal adaptive control, executing the new energy grid-connected monitoring method based on multimodal adaptive control as described in any one of claims 1-5, characterized in that, The system includes: The historical data analysis module is used to collect historical grid-connected power data in new energy grid-connected areas and record historical over-limit alarm events and their corresponding over-limit gaps and fluctuation deviations. The adaptive prediction module establishes and identifies the linear prediction function between the fluctuation difference and the time difference based on the feature point set of historical over-limit alarm events, and outputs the predicted time difference value of the next over-limit alarm event. The multi-terminal correction and evaluation module is used to collect electrical operation data from multiple monitoring terminals deployed in the new energy grid-connected area, correct the deviation of the initial prediction time difference of each monitoring terminal, and calculate the prediction consistency score of each monitoring terminal. The multimodal fusion output module is used to construct the overall distribution characteristics of the prediction consistency scores of each monitoring terminal, calculate the prediction dependency credibility and normalized credibility weight of each monitoring terminal, obtain the final prediction interval duration through weighted fusion, and output it to the grid-connected scheduling platform.

7. The new energy grid-connected monitoring system based on multimodal adaptive control according to claim 6, characterized in that, The historical data analysis module includes: The power volatility safety threshold construction unit is used to collect grid-connected power time series data of new energy grid-connected areas during long-term historical operation periods, calculate the power volatility observation value in each window according to the preset time window, form a historical sample set of power volatility, sort the sample set in ascending order, and use the sample quantile corresponding to the preset confidence level as the power volatility safety threshold R. The over-limit alarm event recording unit is used to feed back a power fluctuation over-limit alarm signal and record it as an over-limit alarm event when the power fluctuation rate touches or exceeds the power fluctuation rate safety threshold in the new energy grid-connected area. Based on the over-limit alarm events, an over-limit gap cluster is formed, and the fluctuation difference between the actual power fluctuation rate and the power fluctuation rate safety threshold is recorded when each over-limit alarm event occurs, forming a fluctuation deviation recording cluster.

8. The new energy grid-connected monitoring system based on multimodal adaptive control according to claim 6, characterized in that, The adaptive prediction module includes: The feature point construction unit is used to pair the time difference and fluctuation difference of the same over-limit alarm event to construct a set of feature points that characterize the disturbance characteristics of the grid-connected area. The linear prediction function establishment unit is used to establish a linear prediction function describing the relationship between fluctuation difference and time difference using the set of feature points as training samples. The recursive least squares online identification unit is used to recursively correct the model parameters using the predicted deviation of the time difference as a feedback signal. This includes model parameter initialization, confidence matrix initialization, recursive update, and prediction output, which outputs the predicted value of the time difference between the next time an over-limit alarm event occurs and the current time.

9. The new energy grid-connected monitoring system based on multimodal adaptive control according to claim 6, characterized in that, The multi-terminal correction and evaluation module includes: The diversified data acquisition unit, based on diversified monitoring terminals deployed in the new energy grid-connected area, collects electrical operation data of the grid-connected point according to the sampling frequency of each monitoring terminal. The monitoring terminals include power quality monitoring terminals, grid-connected power measurement and control terminals, and electrical data acquisition terminals deployed at the grid-connected point. The monitoring data cluster creation unit is used to create an independent monitoring data cluster for each monitoring terminal in the same time period and the same grid-connected area, and a total of N monitoring data clusters are created. The monitoring data clusters are used to record the grid-connected point voltage data, grid-connected current data and grid-connected power data collected by the corresponding monitoring terminal. The preliminary prediction unit is used to extract the grid-connected power time series from the nth monitoring data cluster, calculate the power fluctuation rate sequence of the grid-connected power time series according to a preset time window, substitute each power fluctuation rate data in the power fluctuation rate sequence as the measured power fluctuation rate value into the linear prediction function, and deduplicate and organize the predicted value of the time difference output by the linear prediction function to obtain the time difference prediction value sequence for preliminary prediction based on the nth monitoring data cluster. The deviation correction unit calculates the deviation correction amount based on the deviation sequence between the actual and predicted time difference values ​​recorded by each monitoring terminal in the most recent M historical prediction weeks, and uses the deviation correction amount and the preliminary predicted time difference value sequence to correct the budgeted value of the time difference, thus obtaining the corrected time difference sequence. The consistency score calculation unit is used to calculate the predicted consistency score of the nth monitoring terminal in the current time period based on the corrected time difference sequence.

10. The new energy grid-connected monitoring system based on multimodal adaptive control according to claim 6, characterized in that, The multimodal fusion output module includes: The prediction dependency credibility calculation unit is used to construct the overall distribution characteristics of the prediction consistency scores of each monitoring terminal, and to use the probability density of the temporal consistency scores of each monitoring terminal relative to the overall distribution characteristics as the prediction dependency credibility of the monitoring terminal in the current time period. The normalized confidence weight calculation unit calculates the normalized confidence weight of each monitoring terminal based on the prediction dependency confidence calculation. The weighted fusion output unit is used to calculate the weighted average prediction time difference of each monitoring terminal and output the weighted average prediction time difference as the final prediction interval duration to the grid-connected scheduling platform.

Images