A new energy output risk early warning method and system based on power mutation feature recognition

Abstract

The application discloses a new energy output risk early warning method and system based on power mutation feature recognition, comprising: the application collects output measurement values, meteorological forecast data and grid-connected operation state identifiers and generates operation observation records, reduces misjudgment caused by state switching, and improves early warning input reliability; the mutation feature vector is extracted by sliding window variable point detection and compared with the mutation threshold to generate a mutation trigger identifier, sudden climbing, sudden drop and other abnormalities are positioned in time, and the trigger condition is provided for subsequent risk inference; the mutation feature vector and the meteorological forecast data are input into the Bayesian linear network after triggering to output the future 60-minute output prediction interval and the out-of-limit probability, the quantitative expression of the mutation subsequent evolution uncertainty is realized; the risk intensity is calculated from the out-of-limit probability and the interval width, the risk grade is mapped, the early warning result and the disposal strategy are output, and the operation risk under high proportion of new energy is reduced.

Description

Technical Field

[0001] This invention relates to the technical field of risk assessment for monitoring the grid-connected operation of new energy sources, and in particular to a method and system for early warning of new energy output risks based on power mutation feature identification. Background Technology

[0002] In recent years, the installed capacity of new energy sources such as wind power and photovoltaics has grown rapidly. The influence of new energy output on the power system's supply and demand balance and frequency stability has continued to rise. As the proportion of new energy increases, the grid operation is gradually shifting from being dominated by controllable power sources to involving a high proportion of random power sources. Its output is driven by natural processes such as rapid cloud evolution, gust shear, and sudden changes in temperature and radiation, exhibiting more significant volatility and non-stationarity. Correspondingly, the dispatching side's management of new energy output has expanded from traditional power forecasting and planning to real-time risk assessment and graded early warning oriented towards safety constraints. On the one hand, it is necessary to identify power output mutation events in a short period of time to avoid their chain effects on AGC regulation, standby start-up and shutdown, and power flow constraints. On the other hand, the data acquisition system and numerical weather prediction (NWP) products of new energy power plants are gradually being standardized. Output measurements, meteorological forecast data, and grid-connected operation status indicators can form structured observation records within a unified sampling period, providing a data foundation for risk early warning methods that integrate statistical detection and probabilistic inference.

[0003] However, existing technologies still have shortcomings in early warning of risks associated with renewable energy output: First, common threshold comparison, moving average, or single change point detection methods mostly focus on binary discrimination of whether a sudden change has occurred, lacking characterization of the uncertainty of subsequent evolution after the change. This makes it difficult to translate early warnings into actionable backup configurations, power rationing recommendations, or AGC priority adjustments. Especially in scenarios where weather disturbances intensify rapidly or grid connection status switching is frequent, single criteria are easily affected by sampling gaps and the superposition of control actions, resulting in both false alarms and missed alarms. Second, although some prediction methods can provide predictions of future power points, they usually do not output the prediction range and... First, risk quantities such as the probability of exceeding limits make it difficult to map prediction errors into risks of exceeding limits, thus failing to support tiered early warning and differentiated handling. Second, existing solutions often ignore the constraint role of grid-connected operation status indicators and fail to distinguish the output change mechanism under states such as grid-limited, power-limited, or protection action, resulting in a lack of operational context for mutation identification and risk assessment, thereby weakening the interpretability of early warning results and their ability to be transferred across stations. Based on the above shortcomings, there is an urgent need for an early warning method that can continue to quantify future short-term risks after a mutation is triggered and form a tiered output and strategy association to meet the requirements of real-time safety margin management for high-proportion renewable energy power grids.

[0004] Given that existing new energy power output early warning technologies generally suffer from problems such as only performing binary discrimination of sudden changes, lacking quantification of subsequent uncertainties, and being unable to form executable risk levels and response strategies, this invention is proposed. Therefore, the problem to be solved by this invention is how to integrate power output measurements, meteorological forecast data, and grid-connected operation status indicators under a unified sampling period. First, sudden change features are extracted and sudden change trigger indicators are generated through sliding window change point detection. Then, under the condition of sudden change triggering, a Bayesian linear network is introduced to output the power output prediction interval and the probability of exceeding the limit for the next 60 minutes. The risk intensity index is calculated by combining the interval width, the risk level is mapped, and the response strategy is associated. As mentioned in the background technology above, existing methods are difficult to directly transform sudden change events into scheduling-executable hierarchical early warning results. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0006] In view of the aforementioned existing problems, the present invention is proposed.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for early warning of new energy output risks based on power mutation feature identification, comprising: Within a preset sampling period, collect power output measurements, meteorological forecast data, and grid-connected operation status indicators of new energy power stations to generate operation observation records; The running observation records are input into the mutation detection model constructed based on sliding window change point detection, the mutation feature vector is extracted, and the mutation feature vector is compared with the mutation threshold to generate a mutation trigger identifier; When the mutation triggering identifier is valid, the mutation feature vector and the weather forecast data are input into a Bayesian linear network for calculation, and the output prediction interval and the probability of exceeding the limit for the next 60 minutes from the mutation triggering time are output. The risk intensity index is calculated based on the probability of exceeding the limit and the width of the output prediction interval, and the risk intensity index is mapped to obtain the risk level. The warning result containing the risk intensity index and the risk level, as well as the handling strategy associated with the risk level, are output.

[0008] Secondly, the present invention provides a new energy output risk early warning system based on power mutation feature identification, comprising: The data acquisition module is used to acquire power output measurement values, meteorological forecast data and grid-connected operation status identifiers of new energy power stations within a preset sampling period, and write the power output measurement values, meteorological forecast data and grid-connected operation status identifiers of new energy power stations into the storage module to generate operation observation records; The mutation detection module is used to read the running observation record from the storage module, input the running observation record into the mutation detection model constructed based on sliding window change point detection to extract mutation feature vectors, and generate mutation trigger identifiers based on the comparison of the mutation feature vectors with the mutation threshold. The inference module is used to obtain the mutation feature vector and the weather forecast data when the mutation triggering identifier is in an effective state, and input them into a Bayesian linear network for inference calculation, and output the power output prediction interval and the probability of exceeding the limit for the next 60 minutes from the mutation triggering time. The risk assessment module is used to calculate a risk intensity index based on the probability of exceeding the limit and the width of the output prediction interval, and to map the risk intensity index to obtain a risk level. The strategy output module is used to generate an early warning result that includes the risk intensity index and the risk level, and output a handling strategy associated with the risk level; The storage module is used to store the output measurement value, the meteorological forecast data and the grid-connected operation status identifier in the order of fields when the acquisition module writes, and associate them with the sampling period index to generate the operation observation record; and when the mutation detection module and the inference module call it, it provides the operation observation record matching the sampling period index, as well as the mutation feature vector associated with the mutation trigger identifier and the meteorological forecast data.

[0009] Thirdly, the present invention provides a computer device, comprising: One or more processors; The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the process described above for the new energy output risk early warning method based on power mutation feature identification.

[0010] Fourthly, the present invention provides a computer-readable medium for storing software, the software including instructions executable by one or more computers, the instructions causing the one or more computers to perform operations, the operations including the process of the aforementioned method for early warning of new energy output risks based on power mutation feature identification.

[0011] The beneficial effects of this invention are as follows: By collecting power output measurements, meteorological forecast data, and grid-connected operation status identifiers and generating operation observation records, this invention achieves structured alignment of multi-source information within the same sampling period, reducing misjudgments caused by state switching and improving the reliability of early warning input; by extracting abrupt change feature vectors through sliding window change point detection and comparing them with abrupt change thresholds to generate abrupt change trigger identifiers, it enables timely location of anomalies such as sudden upslopes and sudden drops, providing triggering conditions for subsequent risk inference; by inputting the abrupt change feature vectors and meteorological forecast data into a Bayesian linear network after triggering, it outputs the power output prediction interval and the probability of exceeding the limit for the next 60 minutes, thereby achieving a quantitative expression of the uncertainty of the subsequent evolution of the abrupt change; by calculating the risk intensity from the probability of exceeding the limit and the interval width and mapping the risk level, it outputs early warning results and disposal strategies, reducing the operational risks under high proportion of new energy. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the new energy output risk early warning method based on power mutation feature identification as shown in this invention. Detailed Implementation

[0013] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0014] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0015] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0016] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates a method for early warning of new energy output risks based on power mutation feature identification, which specifically includes the following steps: S1. Collect power output measurements, meteorological forecast data, and grid-connected operation status indicators from the new energy power station within a preset sampling period, and generate an operation observation record. Note that the following should be noted in this step: S1.1 At the beginning of each preset sampling period, the acquisition module reads the power output measurement value of the new energy power station from the power output metering channel, reads the meteorological forecast data corresponding to the preset sampling period from the meteorological forecast interface, and reads the grid connection operation status identifier from the grid connection monitoring interface.

[0017] In a preferred embodiment, the preset sampling period is 60 seconds. The sampling period index is generated incrementally at the beginning of each preset sampling period according to the local clock of the acquisition module, and is written into the running observation record together with the current sampling time.

[0018] As an example, the output measurement value of a new energy power station should at least include the measurement time and the instantaneous active power value.

[0019] As an example, meteorological forecast data includes at least the start time, forecast step index, and a set of meteorological element fields corresponding to the station location; for example, the set of meteorological element fields includes 10m wind speed, wind direction, temperature, air pressure, cloud cover, and shortwave irradiance; the start time is 08:00, and the forecast step index is +10min, +20min, ..., +120min.

[0020] As an example, the grid-connected operation status identifiers include: off-grid status code = 0, grid-connected status code = 1, power generation limitation status code = 2, and maintenance status code = 3; when the read value is 1, the corresponding grid-connected operation status identifier is in the grid-connected state.

[0021] S1.2 Perform field standardization and integrity verification on the power output measurement value, meteorological forecast data and grid connection operation status identifier of the new energy power station, and write them into the same record entry in the order of the power output measurement value, meteorological forecast data and grid connection operation status identifier of the new energy power station to obtain the operation observation record; Field normalization includes aligning and writing the start time, forecast step size index, and sampling period index of meteorological forecast data; integrity verification includes range boundary verification of the output measurement values ​​of new energy power plants, non-empty verification of the start time field and forecast step size index field of meteorological forecast data, and enumeration value verification of the grid-connected operation status identifier, and writing the operation observation record when the verification passes.

[0022] In a preferred embodiment, integrity verification is performed after field normalization and before writing to the record, and the verification rules are as follows: Range boundary verification: The acquisition module reads the rated capacity and allowable negative deviation and allowable positive deviation from the power station configuration to form the allowable output range; for example, if the rated capacity is 50MW, the allowable negative deviation is 0MW, and the allowable positive deviation is +2MW, then the allowable output range is [0MW, 52MW]; when the measured output value of the new energy power station exceeds this allowable output range, the record entry corresponding to this sampling period is not written into the operation observation record; Non-empty check: The acquisition module performs a non-empty check on the start time field and forecast step index field of the meteorological forecast data. If either field is empty, the record entry corresponding to that sampling period will not be written into the running observation record. Enumeration value verification: The acquisition module compares the grid-connected operation status identifier with the enumeration set {0,1,2,3} item by item. If it does not belong to {0,1,2,3}, the record entry corresponding to this sampling period will not be written into the operation observation record.

[0023] When all the above verifications pass, the acquisition module writes the data into the same record entry in the order of the fields: the power output measurement value of the new energy power station, the meteorological forecast data, and the grid connection operation status identifier.

[0024] It should be noted that, since the measured output values ​​of new energy power plants reflect the actual output fluctuations, meteorological forecast data provides the exogenous prior background before and after abrupt changes, and grid-connected operation status indicators reflect the differences in operating conditions such as grid connection, power restriction, and maintenance; by standardizing the fields, verifying the integrity of the data, and writing them into a unified record entry within the same sampling period, a sliding window observation sequence with time alignment is obtained in the subsequent abrupt change detection stage, thereby reducing the probability of confusion between output changes caused by power restriction and output abrupt changes driven by meteorology; compared with the existing technology that directly thresholds the output time series, this embodiment incorporates meteorological forecasts and grid connection conditions into the operation observation record, making the generation of abrupt change trigger indicators have a clearer source of operating condition constraints, reducing false triggers and missed triggers.

[0025] S2. Input the observed data into the mutation detection model built based on sliding window change point detection, extract the mutation feature vector, and compare the mutation feature vector with the mutation threshold to generate a mutation trigger identifier. Note that the following should be noted in this step: S2.1. Form a sliding window observation sequence by indexing the sampling period in the record entries corresponding to N (N=12) consecutive preset sampling periods, and generate a window output sequence by reading the power output measurement value of each record entry from the sliding window observation sequence.

[0026] In a preferred embodiment, the acquisition module extracts the 12 most recent record entries from the running observation records in ascending order of the sampling period index to form a sliding window observation sequence. When there are missing entries in the window, the acquisition module fills in the missing entries with placeholder entries and writes the missing reason code into the placeholder entries, which do not participate in subsequent feature calculations. Subsequently, the acquisition module reads the power output measurement values ​​of the new energy power station one by one from the sliding window observation sequence and writes them into an array structure in order of the sampling period index to obtain the window power output sequence.

[0027] S2.2 Perform sliding window change point detection on the window output sequence, calculate the differential amplitude feature, piecewise slope transition feature and residual fluctuation feature of the window output sequence, and concatenate them in the order of differential amplitude feature, piecewise slope transition feature and residual fluctuation feature to obtain the change feature vector.

[0028] The mutation detection model in this embodiment consists of a candidate change point localization unit and a feature calculation unit: Candidate variable point location unit: Traverse each time step in the window output sequence as a candidate variable point position, calculate the piecewise fitting error on both sides of the candidate variable point position, and select the candidate position with the smallest fitting error as the variable point position within the window. Feature calculation unit: Calculates the differential amplitude feature, piecewise slope transition feature and residual fluctuation feature around the variable point position within the window, and splices them in order to obtain the abrupt change feature vector.

[0029] In a preferred embodiment, the differential amplitude characteristics, piecewise slope transition characteristics, and residual fluctuation characteristics can be calculated using the following formulas: in, This represents the number of record entries corresponding to the sliding window observation sequence. The serial number within the window; For the window output sequence in sequence number Power output measurement values ​​of the new energy power station at the location; The position number of the variable point within the window selected for the candidate variable point positioning unit; It is a differential amplitude characteristic; The slope of the piecewise linear fit to the left of the variable point position; The average of the segment numbers on the left; This represents the average output of the segment on the left. The slope of the piecewise linear fit to the right of the variable point position; The average of the segment numbers on the right; This represents the average output of the segment on the right side; It is a piecewise slope transition characteristic; For piecewise linear fitting of output; , The two segments are the fitting intercepts; For residuals; The mean of the residuals; This is a characteristic of residual fluctuation.

[0030] S2.3. Read the differential amplitude feature, piecewise slope transition feature, and residual fluctuation feature from the mutation feature vector, and compare them with the first mutation threshold, the second mutation threshold, and the third mutation threshold, respectively: When the differential amplitude feature is greater than or equal to the first mutation threshold and the segmented slope transition feature is greater than or equal to the second mutation threshold, a first candidate triggering identifier is generated and set to an effective state; otherwise, it will be set to an invalid state. When the residual fluctuation characteristic is greater than or equal to the third mutation threshold and the grid-connected operation status identifier is equal to the grid-connected status code (grid-connected status code = 1), a second candidate trigger identifier is generated and set to the valid state; otherwise, it is set to the invalid state. The first candidate trigger identifier and the second candidate trigger identifier are encoded as binary status codes, where the valid state corresponds to the first status code and the invalid state corresponds to the second status code. When the binary status code of the first candidate trigger identifier is equal to the first status code and the binary status code of the second candidate trigger identifier is equal to the first status code, a mutation trigger identifier is generated and the mutation trigger identifier is set to the valid state; otherwise, the mutation trigger identifier is set to the invalid state.

[0031] In a preferred embodiment, the first mutation threshold, the second mutation threshold, and the third mutation threshold are obtained through offline calibration: the acquisition module selects at least 30 days of operational observation records from the same site, divides the sample set into two categories: manually verified mutation event windows and normal windows without mutations, and statistically analyzes the distribution intervals of differential amplitude features, segmented slope transition features, and residual fluctuation features within the normal window. The threshold is selected as the high quantile of the normal window distribution, and fine-tuned in conjunction with the recall rate of the mutation event window; for example, the first mutation threshold is set to 6.0 MW, the second mutation threshold is set to 0.35 MW / minute, and the third mutation threshold is set to 2.5 MW.

[0032] It should be noted that the single output threshold judgment is easily affected by short-term dispatch instructions and limited power generation conditions, leading to false alarms in sudden change warnings. This embodiment first locates the change point within the sliding window, then calculates the differential amplitude characteristics, segmented slope transition characteristics, and residual fluctuation characteristics, and introduces the grid-connected operation status identifier as the triggering gating condition. The generation of the sudden change trigger identifier is correlated with the abnormal sudden change mode under the grid-connected state, reducing false triggers caused by limited power generation, off-grid, maintenance and other conditions; at the same time, the three-feature joint judgment improves the coverage of the three types of sudden change modes: sudden drop, sudden rise and sudden vibration.

[0033] S3. When the mutation trigger flag is valid, the mutation feature vector and meteorological forecast data are input together into a Bayesian linear network for calculation, outputting the power output prediction interval and the probability of exceeding the limit for the next 60 minutes from the mutation trigger time. Note that the following should be noted in this step: S3.1 When the mutation triggering flag is in an active state, read the mutation feature vector associated with the mutation triggering flag, and read the meteorological forecast data corresponding to the mutation triggering time, and concatenate them according to the preset field order to obtain the input vector.

[0034] For example, the preset field order is: the mutation feature vector associated with the mutation trigger identifier, and the meteorological forecast data corresponding to the mutation trigger time.

[0035] S3.2 Write the input vector into the input layer of the Bayesian linear network, perform sampling inference based on the weight posterior distribution of the Bayesian linear network, and obtain the output prediction distribution set corresponding to each prediction step size in the next 60 minutes from the moment of mutation triggering.

[0036] In a preferred embodiment, the Bayesian linear network is a linear structure with a single hidden layer. The number of nodes in the input layer is consistent with the dimension of the input vector, and the output layer outputs the data step by step according to the prediction step size within the next 60-minute period. The input vector is obtained by sequentially concatenating the mutation feature vector with the meteorological forecast data aligned with the mutation trigger time.

[0037] In one example, the prediction step size is 5 minutes for each prediction step within the next 60 minutes, corresponding to 12 prediction steps. The Bayesian linear network assigns a set of linear output parameters to each prediction step size and sets independent parameters for the output noise, thus forming the prediction distribution for each prediction step size. During sampling and inference, multiple sets of weight samples are extracted from the weight posterior distribution, and each set of weight samples is operated with the input vector to obtain a set of output prediction samples. All output prediction samples are aggregated for the same prediction step size to obtain the output prediction distribution for that prediction step size. The distributions of all prediction steps together constitute the set of output prediction distributions.

[0038] For example, its mathematical calculation formula is as follows: in, This is the prediction step number, and its value range corresponds to each prediction step within the next 60-minute time period; The sampling sequence number; The input vector; To predict step size In sampling sequence number The following power output prediction samples; To predict step size The corresponding weighted samples; This is an approximate distribution of the weighted posterior distribution; To predict step size Corresponding biased samples; This is an approximate distribution of the biased posterior distribution; This is the output noise sample; To predict step size The corresponding noise variance.

[0039] For example, the sampling number is 200 times, and the prediction step size corresponding to the 20th minute in the future is used to obtain 200 power output prediction samples, the distribution center of which is 28MW, and the dispersion increases with the increase of the mutation feature vector.

[0040] S3.3 Perform quantile statistics on the power output prediction distribution set to generate the power output prediction interval for the next 60 minutes, and calculate the probability of the power output exceeding the preset power output threshold on the power output prediction distribution set to obtain the over-limit probability.

[0041] In a preferred embodiment, for the output prediction distribution of each prediction step, the low quantile and high quantile are statistically analyzed as the upper and lower limits of the output prediction interval for that prediction step, and the intervals of each prediction step within the next 60 minutes are spliced ​​together to form a set of output prediction intervals for the next 60 minutes. In this embodiment, the 5% quantile is used as the lower limit (e.g., 18MW) and the 95% quantile is used as the upper limit (e.g., 31MW).

[0042] In this embodiment, the preset output threshold is given by the station scheduling constraint; for example, if the scheduling requires the output to be no less than 25MW in the next 60 minutes, then the preset output threshold is 25MW, and the over-limit event is defined as the predicted output being less than 25MW; at each prediction step, the over-limit probability is obtained by statistically analyzing the proportion of over-limit events in the prediction samples.

[0043] For example, its mathematical calculation formula is as follows: in, Number of samples; For quantile operators; The quantile parameter is set to 0.05; To predict step size The lower limit of the output prediction range; To predict step size The upper limit of the output prediction range; To predict step size The probability of exceeding the limit; For indicator functions; This is the preset output threshold.

[0044] It should be noted that when the mutation trigger flag is in an effective state, the mutation feature vector and meteorological forecast data are input into a Bayesian linear network to obtain the output prediction distribution set for the next 60 minutes. Then, the output prediction interval is generated by quantile statistics and the probability of exceeding the limit is calculated. This method transforms risk warning from single-point prediction to interval and probability description. In the case of mutation, it can give the magnitude of uncertainty and the possibility of exceeding the limit, thereby realizing subsequent risk level mapping and disposal strategy selection. Compared with common models that only output a single prediction value, this embodiment provides two types of quantitative basis on the model output side: interval width and probability of exceeding the limit, so that the source of risk intensity indicators is consistent with the mutation characteristics.

[0045] S4. Calculate the risk intensity index based on the probability of exceeding limits and the width of the output prediction interval, map the risk intensity index to the risk level, and output the early warning result containing the risk intensity index and the risk level, as well as the handling strategy associated with the risk level. Note that the following should be noted in this step: S4.1. Perform a difference calculation on the upper and lower limits of the power output prediction interval to obtain the interval width (e.g., 13MW). Multiply the over-limit probability by the first weight coefficient (e.g., 0.70) to obtain the first weighted amount. Multiply the interval width by the second weight coefficient (0.30) to obtain the second weighted amount. Then perform a summation operation on the first weighted amount and the second weighted amount to obtain the risk intensity index (e.g., 4.11).

[0046] In a preferred embodiment, the upper and lower limits of the output prediction interval are taken from the upper and lower limits generated by S3.3 for each prediction step. In this embodiment, the prediction step with the highest probability of exceeding the limit in the next 60 minutes is selected as the risk assessment step, and the upper and lower limits of the output prediction interval corresponding to the prediction step are read.

[0047] S4.2 Compare the risk intensity index with the preset first-level threshold (e.g., 2.50) and second-level threshold (e.g., 4.00): When the risk intensity index is less than the first-level threshold, the risk level is level one risk; the handling strategy includes maintaining the current active power output setting and the reporting code as a level one risk warning indicator; When the risk intensity index is greater than or equal to the first-level threshold and less than the second-level threshold, the risk level is level two; the handling strategy includes issuing a suggested value for reserve power and reporting a warning code indicating level two risk. When the risk intensity index is greater than or equal to the second-level threshold, the risk level is determined to be level three risk; the handling strategy includes issuing recommended values ​​for power limitation and AGC adjustment priority, and reporting a warning code indicating level three risk.

[0048] In a preferred embodiment, the first-level threshold and the second-level threshold are determined using historical operational data: at least 90 days of early warning samples are selected, and the samples are divided into three categories according to the retrospective scheduling intervention level: "no intervention required, standby required, and limited issuance and AGC priority adjustment required". The risk intensity index distribution of the corresponding samples is statistically analyzed, and the first-level threshold is selected as the dividing position between the two categories of no intervention required and standby required, and the second-level threshold is selected as the dividing position between the two categories of standby required and limited issuance and AGC priority adjustment required.

[0049] In a preferred example, a 50MW wind farm triggers a sudden change trigger flag at 10:12. Step S3 outputs the power output prediction range for the next 30 minutes as [18MW, 31MW], with an over-limit probability of 0.72. Step S4 obtains a risk intensity index of 4.11 and a risk level of three. The warning result then writes: risk intensity index = 4.11, risk level = three-level risk, and outputs the handling strategy associated with the three-level risk.

[0050] For example, the handling strategy can be written as follows: Recommended power generation limit: Adjust the active power limit of power plants from the current 45MW to 32MW; AGC priority adjustment recommendation: In the set of adjustable sites in the same area, adjust the priority of the site to high priority and control, and simultaneously report a warning indicator with a level 3 risk code.

[0051] If the risk intensity index is 3.10 in the same scenario, the risk level is level 2. The handling strategy can be written as: the recommended value of the backup power is +6MW, and the reporting code is a level 2 risk warning indicator.

[0052] If the risk intensity index is 1.80, the risk level is Level 1. The handling strategy can be written as: maintain the current active power output setting and report a warning code of Level 1 risk.

[0053] In a preferred embodiment, the probability of exceeding the limit and the output prediction range serve as probabilistic constraints for risk assessment: when the probability of exceeding the limit continues to be higher than the probability threshold and the range width continues to expand, the risk level tends to increase, and the corresponding handling strategy shifts from maintaining the setting to a backup suggestion or a limited release and AGC priority suggestion; the handling side can take the probability of exceeding the limit not being higher than the probability threshold as the operational constraint target, and take the risk intensity index as low as possible as the ranking basis for strategy selection, so as to form a traceable constraint and ranking relationship between the early warning result and the scheduling action.

[0054] Preferably, within a short window after a power output mutation is triggered, the risk intensity is characterized by both probability and interval uncertainty. This addresses the problem that single-point predictions alone cannot reflect the uncertainty of the mutation and make it difficult to quantify the response level. By using graded thresholds, the risk intensity indicators are transformed into actionable risk levels, and priority suggestions for reserve, limited power generation, and AGC are output in conjunction with these. This allows the dispatching side to obtain early warning indicators and strategy suggestions with numerical evidence, thereby reducing the probability of power deviation expansion and over-limit events caused by the mutation, and improving the credibility of early warnings and the consistency of response.

[0055] In applying the above embodiments, other aspects of the present invention also propose a new energy output risk early warning system based on power mutation feature identification, including: The data acquisition module is used to collect power output measurement values, meteorological forecast data and grid-connected operation status identifiers of new energy power stations within a preset sampling period, and write the power output measurement values, meteorological forecast data and grid-connected operation status identifiers of new energy power stations into the storage module to generate operation observation records; The mutation detection module is used to read the running observation records from the storage module and input the running observation records into the mutation detection model built based on sliding window change point detection to extract mutation feature vectors. Based on the comparison between the mutation feature vectors and the mutation threshold, a mutation trigger identifier is generated. The inference module is used to obtain the mutation feature vector and meteorological forecast data when the mutation triggering flag is in an effective state, and input them into the Bayesian linear network for inference calculation, and output the power output prediction interval and the probability of exceeding the limit for the next 60 minutes from the mutation triggering time. The risk assessment module is used to calculate the risk intensity index based on the probability of exceeding the limit and the width of the output prediction interval, and to map the risk intensity index to obtain the risk level. The strategy output module is used to generate early warning results that include risk intensity indicators and risk levels, and output the handling strategies associated with the risk levels. The storage module is used to store the output measurement value, meteorological forecast data and grid-connected operation status identifier in the order of fields when the acquisition module writes, and associate them with the sampling period index to generate operation observation records; and when the mutation detection module and inference module call, it provides operation observation records that match the sampling period index as well as mutation feature vectors and meteorological forecast data associated with mutation trigger identifiers.

[0056] Other aspects disclosed in the embodiments of the present invention also provide a computer device including one or more processors and a memory.

[0057] The memory is used to store operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the new energy output risk early warning method based on power mutation feature identification in the foregoing embodiments, especially... Figure 1 The flowchart of the method is shown.

[0058] Other aspects disclosed in the embodiments of the present invention also propose a computer-readable medium for storing software including instructions executable by one or more computers, which, upon execution, cause the one or more computers to perform operations including the flow of the new energy output risk early warning method based on power mutation feature identification of the foregoing embodiments, particularly... Figure 1 The flowchart of the method is shown.

[0059] It should be recognized that embodiments of the present invention may be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium.

[0060] The method can be implemented using standard programming techniques, including a non-transitory computer-readable storage medium configured with a computer program in the computer program, wherein the storage medium is configured such that the computer operates in a specific and predefined manner.

[0061] Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system; however, if necessary, the program can be implemented in assembly or machine language.

[0062] In any case, the language can be either compiled or interpreted.

[0063] Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit.

[0064] The processes described herein (or variations and / or combinations thereof) can be executed under the control of one or more computer systems configured with executable instructions, and can be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program includes a plurality of instructions executable by one or more processors.

[0065] Furthermore, the method can be implemented in any suitable computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices.

[0066] Various aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether portable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein.

[0067] Furthermore, machine-readable code, or parts thereof, can be transmitted via wired or wireless networks.

[0068] When such media includes instructions or programs that combine with a microprocessor or other data processor to implement the steps described above, the invention described herein includes these and other different types of non-transitory computer-readable storage media.

[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for early warning of new energy output risks based on power mutation feature identification, characterized in that, include: Within a preset sampling period, collect power output measurements, meteorological forecast data, and grid-connected operation status indicators of new energy power stations to generate operation observation records; The running observation records are input into the mutation detection model constructed based on sliding window change point detection, the mutation feature vector is extracted, and the mutation feature vector is compared with the mutation threshold to generate a mutation trigger identifier; When the mutation triggering identifier is valid, the mutation feature vector and the weather forecast data are input into a Bayesian linear network for calculation, and the output prediction interval and the probability of exceeding the limit for the next 60 minutes from the mutation triggering time are output. The risk intensity index is calculated based on the probability of exceeding the limit and the width of the output prediction interval, and the risk intensity index is mapped to obtain the risk level. The warning result containing the risk intensity index and the risk level, as well as the handling strategy associated with the risk level, are output.

2. The method for early warning of new energy output risk based on power mutation feature identification according to claim 1, characterized in that, The method for generating the operational observation records includes: At the beginning of each preset sampling period, the acquisition module reads the power output measurement value of the new energy power station from the power metering channel, reads the meteorological forecast data corresponding to the preset sampling period from the meteorological forecast interface, and reads the grid-connected operation status identifier from the grid-connected monitoring interface. The power output measurement value of the new energy power station, the meteorological forecast data, and the grid-connected operation status identifier are subjected to field normalization and integrity verification, and written into the same record entry in the order of the power output measurement value of the new energy power station, the meteorological forecast data, and the grid-connected operation status identifier to obtain the operation observation record; The field normalization includes aligning and writing the start time, forecast step index, and sampling period index of the meteorological forecast data; the integrity verification includes range boundary verification of the output measurement value of the new energy power station, non-empty verification of the start time field and forecast step index field of the meteorological forecast data, and enumeration value verification of the grid-connected operation status identifier, and writing the operation observation record when the verification passes.

3. The method for early warning of new energy output risk based on power mutation feature identification according to claim 2, characterized in that, The method for extracting the mutation feature vector includes: A sliding window observation sequence is formed by indexing the sampling period in the record entries corresponding to N consecutive preset sampling periods, and the power output measurement value of the new energy power station of each record entry is read from the sliding window observation sequence to generate a window power output sequence; Sliding window change point detection is performed on the window output sequence, and the differential amplitude feature, piecewise slope transition feature and residual fluctuation feature of the window output sequence are calculated. The change feature vector is obtained by concatenating the differential amplitude feature, the piecewise slope transition feature and the residual fluctuation feature in the order of the residual fluctuation feature.

4. The method for early warning of new energy output risk based on power mutation feature identification according to claim 3, characterized in that, Read the differential amplitude feature, the piecewise slope transition feature, and the residual fluctuation feature from the mutation feature vector, and compare them with the first mutation threshold, the second mutation threshold, and the third mutation threshold, respectively: When the differential amplitude feature is greater than or equal to the first mutation threshold and the segmented slope transition feature is greater than or equal to the second mutation threshold, a first candidate triggering identifier is generated and set to an effective state; otherwise, it is set to an invalid state. When the residual fluctuation characteristic is greater than or equal to the third mutation threshold and the grid-connected operation status identifier is equal to the grid-connected status code, a second candidate trigger identifier is generated and set to an effective state; otherwise, it is set to an invalid state. The first candidate trigger identifier and the second candidate trigger identifier are respectively encoded as binary status codes, wherein the valid state corresponds to the first status code and the invalid state corresponds to the second status code; when the binary status code of the first candidate trigger identifier is equal to the first status code and the binary status code of the second candidate trigger identifier is equal to the first status code, the mutation trigger identifier is generated and the mutation trigger identifier is set to the valid state; otherwise, the mutation trigger identifier is set to the invalid state.

5. The method for early warning of new energy output risk based on power mutation feature identification according to claim 1, characterized in that, The method for generating the output prediction interval and the over-limit probability is as follows: When the mutation trigger identifier is in a valid state, the mutation feature vector associated with the mutation trigger identifier is read, and the meteorological forecast data corresponding to the mutation trigger time is read, and the input vector is obtained by concatenating them according to the preset field order; The input vector is written into the input layer of the Bayesian linear network, and sampling inference is performed based on the weight posterior distribution of the Bayesian linear network to obtain the output prediction distribution set corresponding to each prediction step size in the next 60-minute period from the moment of mutation triggering. Quantile statistics are performed on the power output prediction distribution set to generate the power output prediction interval for the next 60 minutes, and the probability of the power output exceeding the preset power output threshold is calculated on the power output prediction distribution set to obtain the over-limit probability.

6. The method for early warning of new energy output risk based on power mutation feature identification according to claim 5, characterized in that, The calculation method for the risk intensity index includes: The interval width is obtained by performing a difference operation on the upper and lower limits of the power output prediction interval. The probability of exceeding the limit is multiplied by the first weighting coefficient to obtain the first weighted amount. The interval width is multiplied by the second weighting coefficient to obtain the second weighted amount. The first weighted amount and the second weighted amount are then summed to obtain the risk intensity index.

7. The method for early warning of new energy output risk based on power mutation feature identification according to claim 6, characterized in that, The risk intensity index is mapped to the risk level, including: The risk intensity index is compared with preset first-level thresholds and second-level thresholds: When the risk intensity index is less than the first level threshold, the risk level is level one risk; the handling strategy includes maintaining the current active power output setting and the warning indicator of the reporting code as level one risk. When the risk intensity index is greater than or equal to the first level threshold and less than the second level threshold, the risk level is level two risk; the handling strategy includes issuing a backup power recommendation value and a warning indicator with a reporting code indicating level two risk; When the risk intensity index is greater than or equal to the second level threshold, the risk level is determined to be level three risk; the handling strategy includes issuing a power limitation recommendation value and an AGC adjustment priority recommendation value and reporting a warning code indicating level three risk.

8. A new energy output risk early warning system based on power mutation feature identification, used in the new energy output risk early warning method based on power mutation feature identification as described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to acquire power output measurement values, meteorological forecast data and grid-connected operation status identifiers of new energy power stations within a preset sampling period, and write the power output measurement values, meteorological forecast data and grid-connected operation status identifiers of new energy power stations into the storage module to generate operation observation records; The mutation detection module is used to read the running observation record from the storage module, input the running observation record into the mutation detection model constructed based on sliding window change point detection to extract mutation feature vectors, and generate mutation trigger identifiers based on the comparison of the mutation feature vectors with the mutation threshold. The inference module is used to obtain the mutation feature vector and the weather forecast data when the mutation triggering identifier is in an effective state, and input them into a Bayesian linear network for inference calculation, and output the power output prediction interval and the probability of exceeding the limit for the next 60 minutes from the mutation triggering time. The risk assessment module is used to calculate a risk intensity index based on the probability of exceeding the limit and the width of the output prediction interval, and to map the risk intensity index to obtain a risk level. The strategy output module is used to generate an early warning result that includes the risk intensity index and the risk level, and output a handling strategy associated with the risk level; The storage module is used to store the output measurement value, the meteorological forecast data and the grid-connected operation status identifier in the order of fields when the acquisition module writes, and associate them with the sampling period index to generate the operation observation record; and when the mutation detection module and the inference module call it, it provides the operation observation record matching the sampling period index, as well as the mutation feature vector associated with the mutation trigger identifier and the meteorological forecast data.

9. A computer device, characterized in that, include: One or more processors; The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the new energy output risk early warning method based on power mutation feature identification as described in any one of claims 1 to 7.

10. A computer-readable medium for storing software, characterized in that: The software includes instructions executable by one or more computers, which cause the one or more computers to perform operations, including the flow of the new energy output risk early warning method based on power mutation feature identification as described in any one of claims 1 to 7.

Images