New energy station power prediction method and system based on weather forecast large model

Abstract

The invention relates to the technical field of power supply, in particular to a new energy station power prediction method and system based on a weather forecast large model, and the method comprises the following steps: detecting a weather forecast difference value and output fluctuation, marking an abnormal time period, calculating a voltage and frequency deviation to generate a correction threshold value, and matching phase deviation to generate a smooth reference. And analyzing inertia and response rate, applying dynamic attenuation, and collecting tie line power to adjust, predict and generate an output result. According to the method, abnormal time periods are screened through reverse correlation of meteorological difference and output fluctuation, error conduction is blocked, irradiance prediction is corrected through a voltage frequency integral quantity closed loop, power grid response is improved, actually measured wind speed rolling mean value is matched with AGC phase offset, output smoothness is optimized, inertia time delay and frequency modulation rate dynamic matching is performed, and attenuation constraints are adjusted. The frequency risk is reduced, the tie line fluctuation and deviation superimposed effect is quantified, the attenuation range is dynamically adjusted, the power transmission robustness is enhanced, and wind and light curtailment is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power supply, and in particular to a method and system for predicting power of new energy stations based on a large meteorological forecast model. Background Art

[0002] The field of power supply technology includes the grid-connected dispatching of new energy power generation and the power system supply and demand balance control technology system. The core of this technical field is to establish a dynamic power adjustment mechanism by combining meteorological environment monitoring with power load forecasting, focusing on solving the power fluctuation problem caused by intermittent renewable energy access to the power grid. Specifically, it involves the coupling analysis of numerical weather forecast models and generator output characteristics, constructing a power generation prediction model based on the spatiotemporal evolution of meteorological elements, and forming a collaborative optimization technical solution that includes wind speed and irradiance prediction error compensation and power grid frequency regulation demand response.

[0003] Among them, the power prediction method and system for new energy stations based on the large meteorological forecast model refers to a technical system that uses a deep learning framework to process multi-source meteorological observation data and establishes a prediction model through correlation analysis of historical power data and meteorological elements. The specific implementation steps of this method include: establishing a meteorological element prediction data set with a time resolution of minutes and a spatial resolution of kilometers, using the attention mechanism to extract the time series correlation between key meteorological features and power generation equipment status parameters, and constructing a power conversion model under meteorological fluctuation scenarios. The back propagation algorithm is used to optimize the model parameters in the training stage, and the temperature, air pressure, wind speed and irradiance parameters of the future period are input in the prediction stage, and the power prediction curve and confidence interval are output.

[0004] In the prior art, the correlation analysis between meteorological forecast error and power fluctuation only adopts a positive compensation mechanism, and no reverse relationship screening rules are established. When the meteorological elements suddenly change and the output fluctuation direction is opposite to the trend of the predicted difference, the traditional model cannot identify the abnormal period. For example, in the strong wind shear scenario, the wind speed forecast error is inversely correlated with the actual output fluctuation of the wind turbine, and the existing method will produce wrong correction. The coupling between the traditional numerical weather forecast model and the grid operation status lacks real-time feedback, and the correction parameters rely on offline calibration data. When the grid frequency continues to fluctuate, the irradiance prediction value cannot be dynamically adjusted, resulting in the correction threshold deviating from the actual demand. The existing power prediction model does not consider the impact of the AGC command phase offset on output tracking. When a fixed wind speed prediction sequence is used, when the command response delay exceeds the wind turbine ramp regulation capability, the traditional method will aggravate the phase mismatch between the power curve and the dispatching command. The frequency response mechanism built based on static inertia demand does not quantify the delay matching degree. When the inertia support capacity of the new energy station changes dynamically, the existing technology is difficult to accurately allocate the frequency regulation capacity. For example, in the scenario of time-varying inertia of the wind power cluster, it is easy to cause a secondary drop in frequency. The existing interconnection line power control strategy does not analyze the direction of the superposition effect of fluctuation and plan deviation. The use of a fixed attenuation range will lead to distortion in the adjustable margin assessment, which may trigger conservative power restrictions when the inter-regional power transmission fluctuates in both directions. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art and to propose a new energy station power prediction method and system based on a large meteorological forecast model.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a method for predicting power of new energy stations based on a large meteorological forecast model, comprising the following steps: S1: Detect the difference in temperature and irradiance forecast in the weather forecast of the area where the new energy station is located, synchronously collect the output fluctuation of photovoltaic modules and the output ramp rate data of wind turbines, screen out the abnormal period when the output fluctuation direction is opposite to the meteorological trend, and generate an output abnormality mark; S2: Based on the abnormal mark, extract the voltage fluctuation rate and grid frequency deviation of the station grid connection point, calculate the time domain integral accumulation, determine the deviation level by comparing with the preset reference value, and generate the output correction threshold; S3: calling the correction threshold, obtaining the AGC command power deviation and the actual output curve data, analyzing the matching of the phase offset direction and the preset angle, replacing the wind speed prediction value with the rolling mean sequence of the station measured wind speed, and generating an output smoothing benchmark; S4: extracting the grid inertia support demand and the frequency regulation response rate according to the smoothing benchmark, analyzing the delay matching, applying a dynamic attenuation constraint coefficient to the irradiance prediction value, and generating an output regulation boundary; S5: calling the regulation boundary, collecting the power fluctuation amount and output plan deviation data of the power grid interconnection line in real time, analyzing the direction of the superposition effect, dynamically adjusting the predicted value attenuation interval, and generating the station output prediction result.

[0007] As a further scheme of the present invention, the output abnormality mark is specifically a temperature prediction difference sequence, an irradiance difference fluctuation range, and an output fluctuation reverse time period identifier; the output correction threshold includes a voltage integral accumulation, a frequency deviation integral, and an irradiance dynamic correction value; the output smoothing benchmark is specifically an AGC instruction phase offset angle, a wind speed rolling mean sequence, and a phase matching benchmark angle; the output regulation boundary includes an inertia response delay value, an irradiance attenuation constraint coefficient, and a frequency modulation rate deviation; the output prediction result is specifically a superimposed fluctuation amplitude of the interconnection line, a planned deviation margin range, and a predicted attenuation range adjustment range.

[0008] As a further solution of the present invention, the specific steps of S1 are: S101: Detect the meteorological forecast temperature and the measured temperature, the irradiance prediction value and the measured irradiance in the area where the new energy station is located, match the data by time point, construct a hierarchical prediction difference sequence, calculate the mean square error of temperature and irradiance in the interval, and generate a temperature and irradiance prediction deviation sequence; S102: Acquire the output power of photovoltaic modules and wind turbines, calculate the power change values ​​at adjacent sampling moments, merge the photovoltaic output difference into the output fluctuation, merge the wind turbine output difference into the output ramp rate, and perform time alignment on the data to generate an output change correlation sequence; S103: Call the temperature and irradiance prediction deviation sequence and the output change association sequence, compare the signs of the output fluctuation direction and the meteorological forecast deviation trend, extract all data index sets with opposite signs in the same time interval, establish anomaly marking rules based on the set, and generate an output anomaly mark.

[0009] As a further solution of the present invention, the specific steps of S2 are: S201: Based on the output abnormality mark, extract the voltage measurement data and grid frequency data of the station grid connection point during the abnormal period, calculate the voltage change value between consecutive sampling points, and convert it into a percentage form to obtain a corresponding voltage fluctuation rate sequence, and at the same time perform a difference operation on the grid frequency value and the rated frequency to construct a corresponding frequency deviation sequence, and generate a voltage fluctuation rate and a frequency deviation; S202: calling the voltage fluctuation rate and frequency deviation, performing cumulative integration operations on the two sequences according to the sampling time, calculating the voltage fluctuation integral value and the frequency deviation integral value in the current period, and aligning the two types of integrals according to time to establish a joint sequence to obtain the time domain integral accumulation; S203: Based on the time domain integral accumulation, the voltage fluctuation integral value and the frequency deviation integral value are interval-compared with the set voltage fluctuation reference value and the frequency deviation reference value respectively, and the irradiance prediction value of each time period is linearly interpolated and corrected according to the comparison result, and a correction function is constructed to establish a prediction error suppression boundary under the time series and generate an output correction threshold.

[0010] As a further solution of the present invention, the calculation formula of the standardized voltage change value of the sampling point is specifically: ; in, Indicates The sampling point is relative to the The standardized voltage change value of the sampling points, Indicates The voltage measurement value of the sampling point, Indicates The voltage measurement value of the sampling point, Indicates The grid frequency value at each sampling point, Indicates the rated frequency of the power grid (usually 50Hz), It represents the absolute value sum of frequency deviations of all sampling points in the current abnormal period. Represents the total number of frequency sampling points, Indicates the 1st to 2nd abnormal period. The sum of the squares of the deviations between the voltage measurements and their average value, Represents the voltage mean of all sampling points during the abnormal period, is the voltage standard deviation as the normalization adjustment factor, Indicates the total number of voltage sampling points.

[0011] As a further solution of the present invention, the specific steps of S3 are: S301: calling the output correction threshold, obtaining the AGC command power data and the actual output power curve of the station, calculating the difference between the AGC command power value and the output curve value in the same time interval, extracting the continuous deviation sequence according to the sampling time and establishing the power deviation curve, and generating the AGC command power deviation; S302: Based on the AGC command power deviation, extract the deviation value change trend of each time period and the slope direction of the station output curve for sign comparison, determine whether the change directions of the two types of data are consistent, align the deviation trend direction and the output change direction sign difference by time, and calculate the offset angle, and obtain the phase offset direction matching value according to the sign matching result of the offset angle and the set angle tolerance range; S303: According to the phase offset direction matching value, the original wind speed data within the station sampling period is obtained, the rolling window length is set, the corrected wind speed mean is calculated, the rolling mean sequence of time points is established, and the predicted wind speed value is replaced by the rolling mean sequence according to the time point to generate an output smoothing benchmark.

[0012] As a further solution of the present invention, the corrected wind speed mean calculation formula is specifically: ; in, Representatives The corrected wind speed mean in the rolling window with the last time point as the end, Represents the length of the rolling window, Representative The original wind speed value obtained by monitoring at each time point is Indicates The absolute value of the deviation between the AGC command power and the actual output power at a time point, Indicates The phase shift angle at each time point is represents the mean of all phase offset angles in the rolling window, It represents the sum of the absolute values ​​of the deviations of all wind speed values ​​in the window from their corresponding mean.

[0013] As a further solution of the present invention, the specific steps of S4 are: S401: extracting the grid inertia support data and the frequency modulation response rate data within the matching time period according to the output smoothing benchmark, obtaining the time series of the inertia support time point and the corresponding power change rate, extracting the change interval data reflecting the system adjustment speed in the response rate, and generating the inertia support value and the response rate value; S402: calling the inertia support value and the response rate value, performing difference calculation on the delay difference between the inertia support demand time point and the actual frequency modulation response in the same time period, extracting the response delay length of each time period according to the sampling interval, comparing all interval delay values ​​with the system response delay tolerance value item by item, obtaining the response over-limit deviation of the interval, and obtaining the delay matching deviation degree; S403: Based on the time delay matching deviation, extract the irradiance prediction value sequence of the current time period, set the dynamic attenuation coefficient range, normalize the matching deviation value according to the weight and use it as the attenuation coefficient input for each time period to form a correction curve, establish the adjustment limit range at the time point, and generate the output adjustment boundary.

[0014] As a further solution of the present invention, the specific steps of S5 are: S501: calling the output regulation boundary, collecting the power measurement values ​​on both sides of the grid tie line in the current period and performing time point difference, obtaining the power fluctuation of the grid tie line, and extracting the difference sequence between the station output plan value and the actual output value, constructing the output plan deviation data, and generating the fluctuation and deviation value sequence; S502: Based on the fluctuation amount and deviation value sequence, the two sequences are added item by item in the same time interval, the positive and negative directions of the superposition results are determined, the same direction intervals are extracted according to the direction consistency, the values ​​of the corresponding time periods are calculated, and the matching vector is established in combination with the amplitude and the sign direction to obtain the direction and amplitude value of the superposition effect; S503: According to the direction and amplitude of the superposition effect, the coverage interval is calculated in combination with the upper and lower limits of the output regulation boundary in the current period, and the superposition deviation value and the regulation boundary coverage rate are cross-judged. The dynamic attenuation range of the current prediction value is adjusted according to the cross result, and a prediction curve under the time series is generated to generate the station output prediction result.

[0015] The new energy station power prediction system based on the large meteorological forecast model includes: The meteorological difference recognition module detects the temperature and irradiance forecast and measured values, calculates the predicted difference, synchronously obtains the PV output fluctuation and wind turbine ramp rate, screens the trend reverse period, and generates an output abnormality mark; The electrical disturbance correction module calls the output abnormality mark, extracts the voltage fluctuation rate and the frequency deviation, calculates the product of the two and compares them with the correction benchmark, interpolates and adjusts the irradiance prediction value, and generates an output correction threshold; The output benchmark construction module calls the output correction threshold, obtains the AGC command deviation and the station output curve, determines the phase offset direction of the two and matches the reference angle segment, replaces the wind speed prediction value of this segment with the measured wind speed sliding mean sequence, and generates an output smoothing benchmark; The dynamic boundary control module calls the output smoothing benchmark, extracts the inertia demand and the frequency modulation response rate, determines the matching delay and adjusts the irradiance prediction value according to the degree of deviation, and generates the output regulation boundary; The output forecast generation module calls the output regulation boundary, collects the power fluctuation of the interconnection line and the planned deviation, analyzes the superposition effect and compares the adjustable margin, adjusts the attenuation range of the predicted value, and generates the station output forecast result.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, abnormal time periods are screened through the reverse correlation between meteorological differences and output fluctuations, error conduction is blocked, irradiance prediction is corrected in a closed loop through the voltage-frequency integral, grid response is improved, the measured wind speed rolling mean matches the AGC phase offset, output smoothness is optimized, inertia delay and frequency modulation rate are dynamically matched to adjust attenuation constraints, frequency risks are reduced, the superposition effects of interconnection line fluctuations and deviations are quantified, the attenuation range is dynamically adjusted, transmission robustness is enhanced, and wind and solar power abandonment are reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 It is a schematic diagram of the steps of the present invention; Figure 2 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0019] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0020] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0021] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0022] In the embodiments of the present invention, sometimes the subscripts such as W 1 It may be written in non-subscript form such as W1. When the difference is not emphasized, the meaning is the same.

[0023] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0024] See also Figure 1 The power prediction method of new energy stations based on the meteorological forecast large model includes the following steps: S1: Detect the temperature prediction difference and irradiance prediction difference of the meteorological forecast in the area where the new energy station is located, and simultaneously obtain the output fluctuation of the photovoltaic module and the output ramp rate of the wind turbine. Perform time series correlation analysis on the prediction difference and the output fluctuation, select the abnormal period based on the inverse relationship between the output fluctuation direction and the meteorological difference trend, and generate an output abnormality mark; S2: Based on the abnormal output mark, the voltage fluctuation rate and grid frequency deviation of the station grid connection point are extracted, the time domain integral accumulation of the voltage fluctuation rate and the frequency deviation is calculated, and the irradiance prediction value is dynamically interpolated and corrected based on the comparison result between the accumulation and the preset reference value to generate the output correction threshold; S3: Call the output correction threshold, obtain the AGC command power deviation and the actual output curve of the station, analyze the phase offset direction between the command deviation and the output curve, and replace the wind speed prediction value with the rolling mean sequence of the actual wind speed measured at the station based on the matching of the phase difference and the preset angle to generate an output smoothing benchmark; S4: According to the output smoothing benchmark, the grid inertia support demand and frequency regulation response rate are extracted, the time delay matching degree between the inertia demand and the response rate is analyzed, and based on the deviation degree between the matching degree and the system requirements, dynamic attenuation constraints are imposed on the irradiance prediction value to generate the output regulation boundary; S5: Call the output regulation boundary, collect the power fluctuation of the grid interconnection line and the deviation from the station output plan in real time, analyze the direction and amplitude of the superposition effect of the fluctuation and the deviation, and dynamically adjust the attenuation range of the predicted value based on the coverage of the superposition effect and the adjustable margin to generate the station output prediction result.

[0025] The output abnormality marks specifically include the temperature prediction difference sequence, the irradiance difference fluctuation range, and the output fluctuation reverse period mark. The output correction threshold includes the voltage integral accumulation, the frequency deviation integral, and the irradiance dynamic correction value. The output smoothing benchmark specifically includes the AGC command phase offset angle, the wind speed rolling mean sequence, and the phase matching benchmark angle. The output regulation boundary includes the inertia response delay value, the irradiance attenuation constraint coefficient, and the frequency modulation rate deviation. The output prediction results specifically include the interconnection line superimposed fluctuation amplitude, the planned deviation margin range, and the predicted attenuation range adjustment range.

[0026] The specific steps of S1 are: S101: Detect the meteorological forecast temperature and the measured temperature, the irradiance prediction value and the measured irradiance in the area where the new energy station is located, match the data by time point, construct a hierarchical prediction difference sequence, calculate the mean square error of temperature and irradiance in the interval, and generate a temperature and irradiance prediction deviation sequence; First, access the regional weather forecast service to obtain the temperature and irradiance forecast values ​​every 15 minutes in the next 48 hours. For example, the predicted temperature at 08:00 in a certain area is 24℃ and the irradiance is 510W / m². At the same time, obtain the measured data at the corresponding time through the on-site weather station. Assuming that the measured temperature is 21.5℃ and the measured irradiance is 480W / m², the temperature difference at 08:00 is 2.5℃ and the irradiance difference is 30W / m². This process is continuously calculated according to the time point to form a forecast difference sequence in the whole time period. The temperature difference is set to a classification interval: the difference is less than Very low, to For lower, 1℃ to 1℃ is normal, 1℃ to 3℃ is high, and greater than 3℃ is extremely high. The irradiance difference is set to less than For low, to Light to low, To 20W / m² is normal, 20W / m² to 50W / m² is slightly high, and greater than 50W / m² is high. The difference at each time point is labeled according to the interval to which it belongs. Then the time period is divided into windows, and every two hours is set as an analysis cycle, that is, 08:00 to 10:00, 10:00 to 12:00, etc. The difference value range of all time points in each section is statistically analyzed, and the square value average of the temperature difference sequence and the square value average of the irradiance difference sequence in each section are calculated respectively. For example, there are 8 time points in a cycle, and their temperature differences are 2.5, 3.2, , 0.5, 1.0, , 3.8, 2.0, the irradiance difference is 30, , 50, , , 20, 60, , then the differences are squared and averaged in turn to obtain the error measurement value of the temperature and irradiance of the period. The larger the temperature prediction deviation, the inaccurate the temperature forecast for the area. All periodic error values ​​are sorted one by one in the entire sequence to form a prediction deviation sequence of temperature and irradiance.

[0027] S102: Acquire the output power of photovoltaic modules and wind turbines, calculate the power change values ​​at adjacent sampling moments, merge the photovoltaic output difference into the output fluctuation, merge the wind turbine output difference into the output ramp rate, and perform time alignment on the data to generate an output change correlation sequence; The historical operation records of two types of power data are extracted through the equipment acquisition interface. The photovoltaic output data is arranged at a sampling point of 15 minutes, such as 320kW at 08:00, 340kW at 08:15, 310kW at 08:30, and 335kW at 08:45. The output change between two adjacent moments is calculated, and the results are respectively , , , indicating that the output fluctuations are positive increase, negative fluctuation, and increase again. These three fluctuation values ​​are used to construct the photovoltaic output fluctuation sequence according to the time series. The wind turbine output is also taken at the same time granularity. For example, 08:00 is 850kW, 08:15 is 860kW, 08:30 is 870kW, and 08:45 is 860kW. Comparing the changes at adjacent moments, they are respectively , , However, considering the more stable output of wind turbines, in order to further observe their changing trends, the output change values ​​of wind turbines need to be uniformly expressed as the rate of change during the merging process, that is, the climbing rate is formed based on the change in unit time. For example, an increase of 10kW in 15 minutes is +0.67kW / min, and the wind turbine output climbing rate sequence is formed accordingly. Subsequently, the photovoltaic and wind turbine sequences need to be aligned at the same time. If there are inconsistent sampling times or missing points, they need to be supplemented by linear interpolation to ensure that the photovoltaic fluctuation amount and the wind turbine climbing rate have valid values ​​at each time point. They are sorted into output change association sequences through one-to-one matching, and the two output change values ​​at each moment are recorded completely, which is convenient for subsequent association analysis with meteorological deviation trends.

[0028] S103: calling the temperature and irradiance forecast deviation sequence and the output change association sequence, performing a sign comparison between the output fluctuation direction and the meteorological forecast deviation trend, extracting all data index sets with opposite sign relationships in the same time interval, establishing an abnormal marking rule based on the set, and generating an output abnormal mark; To call the temperature and irradiance prediction deviation sequence and the output change association sequence, it is necessary to first obtain four data at each moment: photovoltaic output fluctuation, wind turbine output climbing rate, temperature prediction deviation value, and irradiance prediction deviation value, and respectively determine whether the photovoltaic fluctuation direction is opposite to the temperature deviation trend direction, that is, when the photovoltaic output increases, the temperature deviation is negative (the predicted temperature is higher than the actual measurement) or when the photovoltaic output decreases, the temperature deviation is positive (the predicted temperature is lower than the actual measurement). Such situations are recorded as opposite directions, and a set threshold is further introduced to avoid errors caused by small fluctuations. The photovoltaic output fluctuation threshold is set to ±10kW, and the temperature difference threshold is set to ±1℃. That is, only when the absolute value of the photovoltaic fluctuation is greater than 10kW and the absolute value of the temperature difference is greater than 1℃, it is judged whether the direction relationship is opposite. The same is true for the fan part. The fan output climbing rate threshold is set to ±0.3kW / min, and the irradiance difference threshold is set to ±30W / m². Only when the climbing rate and the difference meet the conditions, the direction is judged. If it is judged to be in the opposite direction, the moment is marked as an abnormal point. For example, at a certain time point, the photovoltaic fluctuation is +15kW and the temperature difference is , in the opposite direction and exceeding the threshold, it is marked as anomaly, and finally all the time numbers that meet the conditions are recorded to form an anomaly index set, such as 08:15, 09:30, 10:45, etc., and these point positions are assigned 1 in the overall time series according to the set, and other points are assigned 0, completing the output anomaly marking operation for the entire data segment and forming a complete anomaly sequence output.

[0029] The specific steps of S2 are: S201: Based on the abnormal output mark, the voltage measurement data and grid frequency data of the station grid connection point during the abnormal period are extracted, the voltage change value between consecutive sampling points is calculated, and converted into a percentage form to obtain the corresponding voltage fluctuation rate sequence, and the grid frequency value and the rated frequency are differenced to construct the corresponding frequency deviation sequence, and the voltage fluctuation rate and frequency deviation are generated; The calculation formula of the standardized voltage change value of the sampling point is as follows: ; in, Indicates The sampling point is relative to the The standardized voltage change value of the sampling points, Indicates The voltage measurement value of the sampling point, Indicates The voltage measurement value of the sampling point, Indicates The grid frequency value at each sampling point, Indicates the rated frequency of the power grid (usually 50Hz), It represents the absolute value sum of frequency deviations of all sampling points in the current abnormal period. Represents the total number of frequency sampling points, Indicates the 1st to 2nd abnormal period. The sum of the squares of the deviations between the voltage measurements and their average value, Represents the voltage mean of all sampling points during the abnormal period, is the voltage standard deviation as the normalization adjustment factor, Indicates the total number of voltage sampling points; parameter represents the voltage measurement value of the i-th sampling point, which is obtained by the voltage monitoring equipment at the station grid connection point with a sampling frequency of 1 Hz. The measurement value at 10:15 on April 8, 2025 is 10.32 kV; parameter Indicates the voltage measurement value at the previous moment, i.e. 10:14:59, obtained by the same monitoring equipment, which is 10.18 kV; parameter It represents the grid frequency value of the kth sampling point in the abnormal period. The sampling frequency of the monitoring equipment is 1Hz. Assuming that there are 120 sampling points in this period, the absolute value of the difference between its frequency value and the rated frequency is calculated and then summed; Detection value sequence such as: 49.89, 49.95, 50.02, 49.97 Hz, etc., rated frequency Set to 50.00Hz, set according to the standard rated frequency value of the power system, and perform the difference calculation: The absolute value of the frequency deviation sequence is |49.89 50.00|+|49.95 50.00|+|50.02 50.00|+|49.97 50.00|+…=0.11+0.05+0.02+0.03+…≈5.40Hz, total number of sampling points , the average frequency deviation is; Hz parameter It represents the jth voltage value in the abnormal period. There are 121 sampling points from 10:14:00 to 10:16:00, and the mean of the voltage sequence is; ; Among them, 1272.24kV is the sum of the square deviation of the voltage values ​​at 121 points; ; The standard deviation is: ; Substitute all parameters into the formula to calculate: ; The result shows that the standardized voltage change value of the i-th sampling point is 0.01208, that is, the voltage change at this point relative to the previous sampling point is 1.208%. This value will be used as the basic data for calculating the voltage fluctuation percentage and enter the subsequent sequence to further generate the voltage fluctuation rate and frequency deviation. This indicator directly reflects the severity of the voltage change in the current abnormal period. Continuous calculation in all time periods can construct a voltage change sequence for abnormal fluctuation judgment and adjustment strategy support.

[0030] S202: calling the voltage fluctuation rate and frequency deviation, performing cumulative integration operations on the two sequences according to the sampling time, calculating the voltage fluctuation integral value and the frequency deviation integral value in the current period, and aligning the two types of integrals by time to establish a joint sequence to obtain the time domain integral accumulation; Call the voltage fluctuation rate and frequency deviation data, perform cumulative integral operations on the two sequences, first unify the sampling time dimension to ensure that the two sequences have corresponding data at each time point. If there are missing values, perform linear interpolation on the missing points. For example, there are four time points from 08:15 to 09:00, and the corresponding voltage fluctuation rates are 1.2%, , 0.5%, 1.5%, the frequency deviation is , , 0.2Hz, 0.4Hz, then the voltage fluctuation integral value is 1.2 0.8+0.5+1.5=2.4%, the integral value of frequency deviation is 0.1+0.2+0.4=0.2Hz. In this way, the data of all abnormal periods are integrated separately. In this process, the length of the integration period needs to be set. Usually, every two hours is taken as a cycle. The integration result of the sampling points in the cycle is used as the integral value of the cycle. At the same time, the integration threshold is set to judge the validity. For example, when the voltage fluctuation integral value is less than or greater than 3% is considered a large fluctuation, and the frequency integral value is less than Or if it is greater than 0.5Hz, it is considered as a frequency anomaly. The integration results of all abnormal periods are sorted according to the time axis to form an integral cumulative quantity sequence divided by time period. Each time period contains two integral quantities: voltage fluctuation integral and frequency deviation integral. After being arranged in time sequence, a joint time domain sequence is formed for further correlation prediction error analysis.

[0031] S203: According to the time domain integral accumulation, the voltage fluctuation integral value and the frequency deviation integral value are respectively compared with the set voltage fluctuation reference value and the frequency deviation reference value, and the irradiance prediction value of each time period is corrected by linear interpolation according to the comparison result and a correction function is constructed to establish the prediction error suppression boundary under the time series and generate the output correction threshold; The voltage fluctuation integral value and frequency deviation integral value are read in each time period, and compared with the pre-set voltage fluctuation reference value and frequency deviation reference value. The reference value is set with reference to historical statistical data and technical specifications. The voltage fluctuation reference value is set to ±2.0%, and the frequency deviation reference value is set to ±0.3Hz. When the voltage fluctuation integral value exceeds 2.0% or is lower than , while the frequency deviation integral value exceeds 0.3Hz or is lower than When , it is considered that the power grid disturbance is significant in this time period. The original irradiance prediction value in this time period is affected by external disturbances. To correct the prediction value, a linear interpolation operation is performed in this time period. For example, the original predicted irradiance of a certain section is 480, 500, 520, and 510 W / m² respectively. When it is detected that the section needs to be corrected, the end value of the previous period of the current section and the starting value of the next period are used as boundaries, and the value of the section is adjusted smoothly according to the linear gradient. For example, the previous period is 470 W / m² and the next period is 500 W / m², then the four values ​​of the current section are adjusted to 475, 480, 490, and 495 W / m² respectively, and the correction function corresponding to the section is constructed. The function sequence is formed with the time point as the independent variable and the adjusted irradiance as the dependent variable, which is further used to construct the prediction error suppression boundary on the overall time series, and an output correction threshold is generated for each time period, which is taken as the absolute value of the maximum change in the prediction value before and after correction. For example, if the maximum adjustment amplitude is 25 W / m², the correction threshold of this section is set to 25 W / m².

[0032] The specific steps of S3 are: S301: Call the output correction threshold, obtain the AGC command power data and the actual output power curve of the station, calculate the difference between the AGC command power value and the output curve value in the same time interval, extract the continuous deviation sequence according to the sampling time and establish the power deviation curve, and generate the AGC command power deviation; First, it is necessary to determine the corrected threshold sequence of the current station in the analysis period, for example, the threshold from 09:00 to 10:00 is 30kW, and then retrieve the AGC command power data P_agc(t) and the actual output power data P_act(t) of the corresponding period to ensure that the two types of data are aligned according to the sampling time. Every 15 minutes is a sampling interval. For example, data are collected at 09:00, 09:15, 09:30, and 09:45 respectively. The AGC commands are 500kW, 520kW, 510kW, and 530kW, and the actual outputs are 480kW, 500kW, 490kW, and 510kW. The difference between the command and the actual output is 20kW and 20kW calculated point by point. , 20kW, 20kW, to form a power deviation sequence, and then draw a power deviation curve with time as the horizontal axis and power difference as the vertical axis. In this process, it is necessary to judge whether the difference at each time point exceeds the corresponding correction threshold. If the absolute value of the deviation value at a certain time point exceeds the threshold, such as the deviation is 35kW and the threshold is 30kW, then this point is marked as an over-deviation point. This judgment is achieved by taking the absolute value of the difference and comparing it with the threshold. At the same time, all deviation values ​​are sorted in time and the continuous change trend is recorded for subsequent trend analysis. The power deviation curve reflects the real-time error distribution between the AGC execution effect and the station response capability, and finally the AGC command power deviation is output in the form of a sequence.

[0033] S302: Based on the AGC command power deviation, extract the deviation value change trend of each time period and the slope direction of the station output curve for sign comparison to determine whether the change directions of the two types of data are consistent, align the deviation trend direction and the output change direction sign difference by time, and calculate the offset angle, and obtain the phase offset direction matching value based on the sign matching result of the offset angle and the set angle tolerance range; It is necessary to extract the deviation change trend in each time period. First, compare the adjacent points of the power deviation curve, calculate the difference between the current point and the previous point, and determine the direction of change. If the previous deviation is 20kW and the current deviation is 25kW, the trend is rising and the direction is positive. If it changes to 15kW, the trend is falling and the direction is negative, forming a trend direction sequence, such as [+1, ,+1], then obtain the slope change of the station output curve at the corresponding time point. The slope direction is obtained by comparing the adjacent actual output values. For example, if the actual output changes from 500kW to 490kW, the output direction is decreasing, and the negative sign is taken. Then, the deviation change direction and the output change direction constitute two direction sequences respectively. By comparing whether the two directions are the same at each moment, the direction consistency judgment is constructed. If the same sign is used, the direction is consistent, and the opposite sign is used. The two types of direction differences are further converted into offset angle calculation. The offset angle estimates the degree of directional deviation by taking the vector angle formed by the difference between the two values. Then, the set angle tolerance range θ_tol is introduced. For example, the tolerance is set to 30°. When the offset angle is less than 30°, it is considered to be phase matching, and the matching value is 1. If it is greater than 30°, it is considered to be mismatched, and the matching value is 0. The angle tolerance can be set according to the fluctuation of historical actual execution data. For example, the average offset angle between the deviation and output of a station in the past month is 25°, so it is more reasonable to set θ_tol to 30°. Finally, the matching values ​​at all moments are arranged according to time to obtain a complete phase offset direction matching value sequence.

[0034] S303: According to the phase offset direction matching value, the original wind speed data within the station sampling period is obtained, the rolling window length is set, the corrected wind speed mean is calculated, the rolling mean sequence of the time points is established, and the predicted wind speed value is replaced by the rolling mean sequence according to the time points to generate the output smoothing benchmark; The calculation formula for corrected wind speed mean is: ; in, Representatives The corrected wind speed mean in the rolling window with the last time point as the end, Represents the length of the rolling window, Representative The original wind speed value obtained by monitoring at each time point is Indicates The absolute value of the deviation between the AGC command power and the actual output power at a time point, Indicates The phase shift angle at each time point is represents the mean of all phase offset angles in the rolling window, It represents the sum of the absolute values ​​of the deviations of all wind speed values ​​in the window from their corresponding mean values; Tumbling window length Set to 15, according to the actual wind farm sampling frequency of the State Grid once a minute, select 15 minutes as the window period The time point is 10:30, the window corresponding to the interval is 10:16 to 10:30, once per minute, a total of 15 groups of monitoring data original wind speed values The monitoring data obtained through the anemometer are as follows: The wind speed sequence from 10:16 to 10:30 is as follows: 5.1, 5.3, 5.2, 5.0, 4.9, 5.1, 5.4, 5.2, 5.3, 5.3, 5.5, 5.4, 5.2, 5.1, 5.0, The difference between AGC command power and actual output Derived from the station dispatching system, the data is as follows: 20, 25, 22, 18, 21, 23, 19, 17, 20, 26, 30, 24, 21, 20, 19kW The wind speed deviation and mean are calculated as follows: ; ; Phase shift angle The matching angle value comes from the difference between the AGC command response angle and the output slope direction, which is calculated as follows (in degrees): 28, 32, 30, 31, 29, 30, 33, 34, 30, 28, 29, 31, 30, 32, 30, The average value is ; The 15th time point The calculation below is as follows: Wind speed value at point 15 , The AGC deviation is kW, ; ; The correction value is ; The same goes for the remaining 14 groups. All weighted values ​​are summed up and the average is calculated: Let the sum be , but ; The results show that the corrected wind speed average at the 15th time point is 14.71 meters per second, which shows a significant upward trend compared to the original predicted value of 5.0 meters per second. This corrected value is the smoothing benchmark used to replace the predicted wind speed. It is then brought into the prediction model for output recalculation to fully generate a prediction curve based on the smoothed wind speed.

[0035] The specific steps of S4 are: S401: extracting the grid inertia support data and frequency modulation response rate data within the matching time period according to the output smoothing benchmark, obtaining the time series of the inertia support time point and the corresponding power change rate, extracting the change interval data reflecting the system adjustment speed in the response rate, and generating the inertia support value and the response rate value; First, the inertia support data and frequency modulation response rate data consistent with the time period are extracted, and the time series analysis of the inertia support data is performed to obtain the specific time point of the actual triggering of the inertia support function, and the power change rate at the corresponding time is extracted as the response indicator. For example, during the period from 10:00 to 11:00, the inertia support triggering time recorded by the system is 10:15, 10:30, and 10:45, and the power change rate of each trigger point is +15kW / min, +15kW / min, and +15kW / min, respectively. , +12kW / min, it can be organized into a time series format and its direction and amplitude can be marked. At the same time, the frequency modulation response rate data can be called to extract the recorded response rate point by point within the same time range. For example, the rate from 10:15 to 10:30 is +8kW / min, and from 10:30 to 10:45 is , 10:45 to 11:00 is +9kW / min, and the data is divided into intervals to analyze the continuous changes in the system response. The response rate fluctuation range is determined by calculating the difference between the maximum and minimum values ​​of the rate data in each interval. For example, if the rate in a certain interval changes from +5kW / min to +12kW / min, the fluctuation amplitude is 7kW / min. By judging whether the amplitude exceeds the set threshold (the upper limit of the response rate fluctuation is set to 10kW / min here), the time period with drastic response fluctuations is screened out and further marked as a sensitive section. At the same time, the data is reorganized with a fixed time length such as 15 minutes as the interval unit and a two-way time series is established. The inertia support value corresponding to each time period (that is, the power change rate of the trigger point) and the frequency modulation response rate value are matched and combined into a structured data group, and finally a complete sequence of inertia support value and response rate value is output.

[0036] S402: calling the inertia support value and the response rate value, performing difference calculation on the delay difference between the inertia support demand time point and the actual frequency modulation response in the same time period, extracting the response delay length of each time period according to the sampling interval, comparing all interval delay values ​​with the system response delay tolerance value item by item, obtaining the response over-limit deviation of the interval, and obtaining the delay matching deviation degree; First, identify the actual frequency modulation response time point corresponding to each inertia support demand, that is, the time point when the system response rate changes by more than 2kW / min for the first time after the inertia support trigger point is taken as the frequency modulation response starting point. For example, if the inertia support demand occurs at 10:15, and the response rate is recorded to change from +3kW / min to +8kW / min at 10:17, then the response start time is considered to be 10:17, and the response delay of this section is calculated to be 2 minutes. In this way, the time difference between each inertia support demand point and its corresponding frequency modulation response is calculated to form a response delay sequence. Each delay value is compared with the maximum tolerable response delay set by the system, and the tolerance value is set to 3 minutes. If a certain time If the response delay of a time segment is 4 minutes, it is considered to be out of limit and recorded as 1, otherwise it is recorded as 0. The judgment result is constructed as an out-of-limit deviation label corresponding to each time delay in the time series. All time segments are arranged in sampling order, and the out-of-limit segments and their corresponding deviation values ​​are sorted out. Finally, the proportion or frequency of response out-of-limit in the entire time domain is counted. By setting the comparison logic, it is determined whether the response of each time segment meets the frequency modulation requirements. Finally, the delay matching deviation is output according to the time point. The value interval is divided into 0 for complete synchronization and 1 for complete mismatch. The intermediate interval is set by linear interpolation, such as 0.5 for 2 minutes and 0.8 for 3 minutes, etc., to ensure that the matching degree calculation result can be used for subsequent normalization processing.

[0037] S403: Based on the delay matching deviation, extract the irradiance prediction value sequence of the current period, set the dynamic attenuation coefficient range, normalize the matching deviation value according to the weight and use it as the attenuation coefficient input of each period, form a correction curve, establish the adjustment limit range at the time point, and generate the output adjustment boundary; The corresponding irradiance prediction value sequence is extracted for each time period. The corresponding prediction values ​​are 450W / m², 470W / m², 490W / m², and 510W / m² in the time period from 10:00 to 11:00, respectively. The dynamic attenuation coefficient range is set to 0.6 to 1.0. The specific weight distribution is calculated based on the delay matching deviation. The matching deviation is 0, corresponding to the attenuation coefficient of 1.0, and the deviation is 1, corresponding to the attenuation coefficient of 0.6. The intermediate value is calculated by linear interpolation. For example, if the deviation is 0.75, the attenuation coefficient is 0.7. The matching deviation of each time period is converted into an attenuation coefficient and then mapped in sequence to obtain a weight coefficient sequence, such as [1.0, 0.9, 0.8, 0.7]. Then the original irradiance prediction value is multiplied by A new correction curve is formed with the corresponding attenuation coefficient, such as 450W / m²×1.0=450W / m², 470W / m²×0.9=423W / m², and so on to form a corrected prediction sequence [450,423,392,357]. This sequence is used to generate the output regulation limit range, and the difference between it and the original predicted value is used as the boundary bandwidth. For example, if the original value is 490W / m² and the corrected value is 392W / m², then the adjustment boundary bandwidth is ±98W / m². The upper and lower limits of each moment are set based on this bandwidth, that is, 490±98W / m², corresponding to 392W / m² to 588W / m². Finally, the correction values ​​at all time points are combined with the boundary range to form a complete output regulation boundary.

[0038] The specific steps of S5 are: S501: Call the output regulation boundary, collect the power measurement values ​​on both sides of the grid tie line in the current period and perform time point difference to obtain the power fluctuation of the grid tie line, and extract the difference sequence between the station output plan value and the actual output value, construct the output plan deviation data, and generate the fluctuation and deviation value sequence; First, determine the time period currently being analyzed, for example, 10:00 to 11:00, and synchronously collect the power measurement value sequence on both sides of the grid tie line during this period from the main control system to ensure consistent sampling frequency. Every 15 minutes is a sampling point, which are recorded as the power value P_A(t) on the A side of the tie line and the power value P_B(t) on the B side. For example, at 10:00, P_A=1050kW and P_B=1075kW, and at 10:15, P_A=1080kW and P_B=1060kW. By calculating the difference component ΔP_grid(t)=P_A(t) point by point, P_B(t), and then extract the variation of the difference between adjacent moments to obtain the power fluctuation sequence, such as ΔP_grid(10:15) ΔP_grid(10:00)= , and the power fluctuation of the grid interconnection line is obtained accordingly. At the same time, the output plan value P_plan(t) and the actual output value P_real(t) of the station are retrieved to construct the output deviation data, that is, the deviation value is P_real(t) P_plan(t), if the planned output at a certain moment is 1000kW and the actual output is 960kW, then the deviation value is , record the output deviation at each sampling moment in chronological order to form a complete output plan deviation sequence, match and sort the above fluctuation sequence and deviation value sequence according to the time axis, and construct two column structure sequences for subsequent vector calculation and direction superposition judgment, and finally obtain the fluctuation and deviation value sequence.

[0039] S502: Based on the fluctuation amount and deviation value sequence, the two sequences are added item by item in the same time interval, the positive and negative directions of the superposition results are determined, the same direction intervals are extracted according to the direction consistency, the values ​​of the corresponding time periods are calculated, and the matching vector is established by combining the amplitude and the sign direction to obtain the direction and amplitude value of the superposition effect; Based on the fluctuation and deviation value sequence, the two are added item by item in the same time interval, that is, the power fluctuation of the grid tie line at each time point is added to the corresponding output plan deviation value to obtain the combined value. For example, the fluctuation is , the deviation is , then the superposition value is , and then determine the positive and negative directions of the superposition value. If it is a positive value, it is marked as +1, and if it is a negative value, it is marked as , zero value is marked as 0, and the sign direction of the fluctuation amount and the deviation value are extracted at the same time. If both are positive or negative, they are judged as the same direction interval. If the directions are opposite, they are different direction intervals. For example, if the fluctuation amount is +25kW and the deviation is +30kW, they are in the same direction. If one is +30kW and the other is For the same-direction interval, the amplitude of the actual superposition value is extracted and the corresponding time start and end points of the segment are recorded. At the same time, the fluctuation direction, deviation value direction and superposition direction are combined into a three-dimensional vector structure to describe the directional coupling situation in the current period. The directional consistency of each segment is further represented by a vector by constructing a matching vector. The vector value of 1 represents that the three directions are completely consistent. Represents completely opposite, and 0 represents no direction. In this way, the corresponding relationship between each vector and the superposition amplitude value in the current time period is obtained, and the matching vectors of all time periods and their corresponding superposition amplitude values ​​are structured and combined to form a complete sequence of superposition effect direction and amplitude values.

[0040] S503: Calculate the coverage interval according to the direction and amplitude of the superposition effect and the upper and lower limits of the output regulation boundary in the current period, make an interval cross judgment on the superposition deviation value and the regulation boundary coverage rate, adjust the dynamic attenuation range of the current prediction value according to the cross result, generate a prediction curve under the time series, and generate the station output prediction result; According to the direction and amplitude of the superposition effect, the upper and lower limits of the output regulation boundary of the current period are retrieved. For example, if the predicted output of a certain period is 1000kW, the lower limit of the regulation boundary is 950kW, and the upper limit is 1050kW, the coverage interval of the period is constructed as [950,1050]kW, and the superposition deviation value is cross-judged with this coverage interval. If the superposition value is within the interval, it is considered to fluctuate within the acceptable range. If it exceeds the upper and lower limits of the boundary, the predicted value needs to be adjusted. For example, the superposition deviation value at a certain moment is , the predicted value is 1000kW, then it is 925kW after adjustment, which is lower than the lower limit of 950kW. At this time, attenuation processing should be applied to the predicted value to limit its deviation trend. The dynamic attenuation range is set to [0.7, 1.0]. The specific value is determined according to the relationship between the superposition amplitude and the regulation boundary coverage rate. The coverage rate of 1 means that it is completely within the regulation interval, and the coverage rate of 0 means that it is completely out of bounds. If the deviation value exceeds the lower limit of 25kW and the boundary width is 100kW, the coverage rate is 0.75, and the corresponding attenuation coefficient can be set to 0.8. The predicted value is adjusted to the original value multiplied by the attenuation coefficient, such as 1000kW×0.8=800kW. All predicted values ​​are adjusted according to the corresponding coefficients in chronological order and the time series is reconstructed to finally form the adjusted output prediction curve as the new station output prediction result.

[0041] See also Figure 2 , the new energy station power prediction system based on the large meteorological forecast model includes: The meteorological difference recognition module detects the temperature and irradiance forecast and measured values, calculates the predicted difference, synchronously obtains the PV output fluctuation and wind turbine ramp rate, screens the trend reverse period, and generates an output abnormality mark; The electrical disturbance correction module calls the output abnormality mark, extracts the voltage fluctuation rate and frequency deviation, calculates the product of the two and compares them with the correction benchmark, interpolates and adjusts the irradiance prediction value, and generates the output correction threshold; The output benchmark construction module calls the output correction threshold, obtains the AGC command deviation and the station output curve, determines the phase offset direction of the two and matches the reference angle segment, replaces the wind speed prediction value of this segment with the measured wind speed sliding mean sequence, and generates an output smoothing benchmark; The dynamic boundary control module calls the output smoothing benchmark, extracts the inertia demand and frequency modulation response rate, determines their matching delay, and adjusts the irradiance prediction value according to the degree of deviation to generate the output regulation boundary; The output forecast generation module calls the output regulation boundary, collects the power fluctuation of the interconnection line and the planned deviation, analyzes their superposition effect and compares the adjustable margin, adjusts the predicted value attenuation range, and generates the station output forecast result.

[0042] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A new energy station power prediction method based on a large meteorological forecast model, characterized in that: The following steps are involved: S1: Detect the difference in temperature and irradiance forecast in the weather forecast of the area where the new energy station is located, synchronously collect the output fluctuation of photovoltaic modules and the output ramp rate data of wind turbines, screen out the abnormal period when the output fluctuation direction is opposite to the meteorological trend, and generate an output abnormality mark; S2: Based on the abnormal mark, extract the voltage fluctuation rate and grid frequency deviation of the station grid connection point, calculate the time domain integral accumulation, determine the deviation level by comparing with the preset reference value, and generate the output correction threshold; S3: calling the correction threshold, obtaining the AGC command power deviation and the actual output curve data, analyzing the matching of the phase offset direction and the preset angle, replacing the wind speed prediction value with the rolling mean sequence of the station measured wind speed, and generating an output smoothing benchmark; S4: extracting the grid inertia support demand and the frequency regulation response rate according to the smoothing benchmark, analyzing the delay matching, applying a dynamic attenuation constraint coefficient to the irradiance prediction value, and generating an output regulation boundary; S5: calling the regulation boundary, collecting the power fluctuation amount and output plan deviation data of the power grid interconnection line in real time, analyzing the direction of the superposition effect, dynamically adjusting the predicted value attenuation interval, and generating the station output prediction result.

2. The method for predicting power of new energy stations based on a large meteorological forecast model according to claim 1 is characterized in that: The output abnormality mark specifically includes the temperature prediction difference sequence, the irradiance difference fluctuation range, and the output fluctuation reverse time period identifier; the output correction threshold includes the voltage integral accumulation, the frequency deviation integral, and the irradiance dynamic correction value; the output smoothing benchmark specifically includes the AGC command phase offset angle, the wind speed rolling mean sequence, and the phase matching benchmark angle; the output regulation boundary includes the inertia response delay value, the irradiance attenuation constraint coefficient, and the frequency modulation rate deviation; the output prediction result specifically includes the interconnection line superimposed fluctuation amplitude, the planned deviation margin range, and the predicted attenuation range adjustment range.

3. The method for predicting power of new energy stations based on a large meteorological forecast model according to claim 1 is characterized in that: The specific steps of S1 are: S101: Detect the meteorological forecast temperature and the measured temperature, the irradiance prediction value and the measured irradiance in the area where the new energy station is located, match the data by time point, construct a hierarchical prediction difference sequence, calculate the mean square error of temperature and irradiance in the interval, and generate a temperature and irradiance prediction deviation sequence; S102: Acquire the output power of photovoltaic modules and wind turbines, calculate the power change values ​​at adjacent sampling moments, merge the photovoltaic output difference into the output fluctuation, merge the wind turbine output difference into the output ramp rate, and perform time alignment on the data to generate an output change correlation sequence; S103: Call the temperature and irradiance prediction deviation sequence and the output change association sequence, compare the signs of the output fluctuation direction and the meteorological forecast deviation trend, extract all data index sets with opposite signs in the same time interval, establish anomaly marking rules based on the set, and generate an output anomaly mark.

4. The method for predicting power of new energy stations based on a large meteorological forecast model according to claim 1 is characterized in that: The specific steps of S2 are: S201: Based on the output abnormality mark, extract the voltage measurement data and grid frequency data of the station grid connection point during the abnormal period, calculate the voltage change value between consecutive sampling points, and convert it into a percentage form to obtain a corresponding voltage fluctuation rate sequence, and at the same time perform a difference operation on the grid frequency value and the rated frequency to construct a corresponding frequency deviation sequence, and generate a voltage fluctuation rate and a frequency deviation; S202: calling the voltage fluctuation rate and frequency deviation, performing cumulative integration operations on the two sequences according to the sampling time, calculating the voltage fluctuation integral value and the frequency deviation integral value in the current period, and aligning the two types of integrals according to time to establish a joint sequence to obtain the time domain integral accumulation; S203: Based on the time domain integral accumulation, the voltage fluctuation integral value and the frequency deviation integral value are interval-compared with the set voltage fluctuation reference value and the frequency deviation reference value respectively, and the irradiance prediction value of each time period is linearly interpolated and corrected according to the comparison result, and a correction function is constructed to establish a prediction error suppression boundary under the time series and generate an output correction threshold.

5. The method for predicting power of new energy stations based on a large meteorological forecast model according to claim 4 is characterized in that: The calculation formula of the standardized voltage change value of the sampling point is specifically: ； in, Indicates The sampling point is relative to the The standardized voltage change value of the sampling points, Indicates The voltage measurement value of the sampling point, Indicates The voltage measurement value of the sampling point, Indicates The grid frequency value at each sampling point, Indicates the rated frequency of the power grid (usually 50Hz), It represents the absolute value sum of frequency deviations of all sampling points in the current abnormal period. Represents the total number of frequency sampling points, Indicates the 1st to 2nd abnormal period. The sum of the squares of the deviations between the voltage measurements and their average value, Represents the voltage mean of all sampling points during the abnormal period, is the voltage standard deviation as the normalization adjustment factor, Indicates the total number of voltage sampling points.

6. The method for predicting power of new energy stations based on a large meteorological forecast model according to claim 1 is characterized in that: The specific steps of S3 are: S301: calling the output correction threshold, obtaining the AGC command power data and the actual output power curve of the station, calculating the difference between the AGC command power value and the output curve value in the same time interval, extracting the continuous deviation sequence according to the sampling time and establishing the power deviation curve, and generating the AGC command power deviation; S302: Based on the AGC command power deviation, extract the deviation value change trend of each time period and the slope direction of the station output curve for sign comparison, determine whether the change directions of the two types of data are consistent, align the deviation trend direction and the output change direction sign difference by time, and calculate the offset angle, and obtain the phase offset direction matching value according to the sign matching result of the offset angle and the set angle tolerance range; S303: According to the phase offset direction matching value, the original wind speed data within the station sampling period is obtained, the rolling window length is set, the corrected wind speed mean is calculated, the rolling mean sequence of time points is established, and the predicted wind speed value is replaced by the rolling mean sequence according to the time point to generate an output smoothing benchmark.

7. The method for predicting power of new energy stations based on a large meteorological forecast model according to claim 1 is characterized in that: The modified wind speed mean value calculation formula is specifically: ； in, Representatives The corrected wind speed mean in the rolling window with the last time point as the end, Represents the length of the rolling window, Representative The original wind speed value obtained by monitoring at each time point is Indicates The absolute value of the deviation between the AGC command power and the actual output power at a time point, Indicates The phase shift angle at each time point is represents the mean of all phase offset angles in the rolling window, It represents the sum of the absolute values ​​of the deviations of all wind speed values ​​in the window from their corresponding mean.

8. The method for predicting power of new energy stations based on a large meteorological forecast model according to claim 1 is characterized in that: The specific steps of S4 are: S401: extracting the grid inertia support data and the frequency modulation response rate data within the matching time period according to the output smoothing benchmark, obtaining the time series of the inertia support time point and the corresponding power change rate, extracting the change interval data reflecting the system adjustment speed in the response rate, and generating the inertia support value and the response rate value; S402: calling the inertia support value and the response rate value, performing difference calculation on the delay difference between the inertia support demand time point and the actual frequency modulation response in the same time period, extracting the response delay length of each time period according to the sampling interval, comparing all interval delay values ​​with the system response delay tolerance value item by item, obtaining the response over-limit deviation of the interval, and obtaining the delay matching deviation degree; S403: Based on the time delay matching deviation, extract the irradiance prediction value sequence of the current time period, set the dynamic attenuation coefficient range, normalize the matching deviation value according to the weight and use it as the attenuation coefficient input for each time period to form a correction curve, establish the adjustment limit range at the time point, and generate the output adjustment boundary.

9. The method for predicting power of new energy stations based on a large meteorological forecast model according to claim 1 is characterized in that: The specific steps of S5 are: S501: calling the output regulation boundary, collecting the power measurement values ​​on both sides of the grid tie line in the current period and performing time point difference, obtaining the power fluctuation of the grid tie line, and extracting the difference sequence between the station output plan value and the actual output value, constructing the output plan deviation data, and generating the fluctuation and deviation value sequence; S502: Based on the fluctuation amount and deviation value sequence, the two sequences are added item by item in the same time interval, the positive and negative directions of the superposition results are determined, the same direction intervals are extracted according to the direction consistency, the values ​​of the corresponding time periods are calculated, and the matching vector is established in combination with the amplitude and the sign direction to obtain the direction and amplitude value of the superposition effect; S503: According to the direction and amplitude of the superposition effect, the coverage interval is calculated in combination with the upper and lower limits of the output regulation boundary in the current period, and the superposition deviation value and the regulation boundary coverage rate are cross-judged. The dynamic attenuation range of the current prediction value is adjusted according to the cross result, and a prediction curve under the time series is generated to generate the station output prediction result.

10. The new energy station power prediction system based on the weather forecast large model is characterized by: According to the method for predicting power of new energy stations based on a large meteorological forecast model according to any one of claims 1 to 9, the system comprises: The meteorological difference recognition module detects the temperature and irradiance forecast and measured values, calculates the predicted difference, synchronously obtains the PV output fluctuation and wind turbine ramp rate, screens the trend reverse period, and generates an output abnormality mark; The electrical disturbance correction module calls the output abnormality mark, extracts the voltage fluctuation rate and the frequency deviation, calculates the product of the two and compares them with the correction benchmark, interpolates and adjusts the irradiance prediction value, and generates an output correction threshold; The output benchmark construction module calls the output correction threshold, obtains the AGC command deviation and the station output curve, determines the phase offset direction of the two and matches the reference angle segment, replaces the wind speed prediction value of this segment with the measured wind speed sliding mean sequence, and generates an output smoothing benchmark; The dynamic boundary control module calls the output smoothing benchmark, extracts the inertia demand and the frequency modulation response rate, determines the matching delay and adjusts the irradiance prediction value according to the degree of deviation, and generates the output regulation boundary; The output forecast generation module calls the output regulation boundary, collects the power fluctuation of the interconnection line and the planned deviation, analyzes the superposition effect and compares the adjustable margin, adjusts the attenuation range of the predicted value, and generates the station output forecast result.

Images