A cold front type wind power climbing ultra-short-term prediction method

Abstract

This application discloses an ultra-short-term prediction method for cold front-type wind power ramping, relating to the field of wind power generation technology. The method includes: acquiring target wind field, surrounding candidate observation nodes, and numerical weather prediction data before the prediction release time, and performing quality control and screening; identifying candidate cold front signals and their occurrence times at each node based on the coordinated changes of multiple meteorological elements; determining the cold front propagation direction by combining numerical weather prediction and signal occurrence timing, dynamically screening effective upstream nodes, and constructing a consistent cold front meteorological observation propagation chain; and fusing upstream node signals and propagation time delays to predict the cold front passage time at the target wind field, using this time as a time anchor point to perform event-level correction on the basic wind power prediction curve. This significantly improves the time positioning and amplitude accuracy of cold front-type ramping prediction, enhances the reliability and timeliness of early warnings, and can effectively support the safe and stable dispatching and operation of the power grid.

Description

Technical Field

[0001] This application relates to the field of wind power generation technology, and in particular to a short-term prediction method for cold front-type wind power ramp-up. Background Technology

[0002] Short-term wind power forecasting is a crucial foundation for grid-connected dispatching of new energy sources, power generation planning, and reserve capacity allocation. Rapid wind power ramp-up events triggered by cold fronts pose a significant challenge to the safe and stable operation of the power grid.

[0003] Many existing wind power forecasting methods implicitly incorporate the impact of cold fronts into independent meteorological features such as wind speed and temperature, failing to explicitly model them as meteorological disturbances with propagation direction and speed. This makes it difficult to accurately determine the arrival time of the cold front at the target wind field. Furthermore, current multi-source observation fusion methods employ fixed site sets and fixed ranges, failing to dynamically identify effective upstream nodes based on the real-time propagation direction of the cold front. This easily introduces interference from lateral, downstream, and local disturbance nodes into the forecast results. In addition, these methods lack a consistency verification mechanism for the cold front propagation chain, making it difficult to distinguish between actual cold front propagation and local meteorological anomalies. They also lack a causal propagation relationship between upstream cold front signals and the time of target wind field passage, and a mechanism for correcting event-level power curves using the time of cold front passage as a time anchor. This results in insufficient time positioning and amplitude accuracy in cold front-type uphill predictions, failing to meet the core requirement of power grid dispatch for early warning of uphill events. Summary of the Invention

[0004] This application provides an ultra-short-term prediction method for cold front-type wind power ramp-up. It can solve the problems of insufficient time positioning and amplitude accuracy in cold front-type wind power ramp-up prediction caused by the failure to explicitly model the cold front as a meteorological disturbance object, the use of fixed sites to screen multi-source data, the lack of propagation chain consistency verification, the failure to establish the causal relationship between upstream signals and the passing time of the target wind field, and the lack of event-level power correction mechanism.

[0005] According to a first aspect of this application, a method for ultra-short-term prediction of cold front-type wind power ramp-up is provided, comprising:

[0006] Acquire data on the target wind field, surrounding candidate observation nodes, and numerical weather prediction prior to the forecast release time, and perform data quality control and screening; Based on the coordinated changes of various meteorological elements at the candidate observation nodes, the candidate cold front signals for each candidate observation node are identified, and the occurrence time of the cold front signal is determined. Based on the data from the numerical weather forecast and the order of the occurrence times of the cold front signals at multiple candidate observation nodes, the current propagation direction of the cold front is determined. Based on the propagation direction of the cold front, effective upstream nodes located in the effective upstream direction of the target wind field are dynamically selected from the candidate observation nodes; The effective upstream nodes are sorted according to their relative positions with the target wind field in the propagation direction to construct a cold front meteorological observation propagation chain. The propagation chain is then checked for consistency, and nodes that do not meet the check requirements are removed or downgraded. Based on the occurrence time of the cold front signal of the effective upstream node and its propagation time delay to the target wind field, the passage time of the cold front of the target wind field is predicted. Using the time of the cold front passing through the target wind field as the time anchor point, the basic wind power prediction curve is modified at the event level related to the cold front-type climbing event to obtain the modified wind power prediction result.

[0007] This application addresses the shortcomings of related technologies in terms of time positioning and amplitude accuracy in cold front wind power ramp prediction. By employing multi-factor collaborative change to identify candidate cold front signals and determine their occurrence time, combining numerical weather prediction with the cold front signal occurrence sequence to determine the propagation direction, dynamically screening effective upstream nodes and constructing a consistent cold front meteorological observation propagation chain, and integrating upstream node signals with propagation time delays to predict the cold front passage time of the target wind field, and using this passage time as a time anchor point to perform event-level correction on the basic power prediction curve, this approach solves the problems of insufficient time positioning and amplitude accuracy in cold front-type wind power ramp prediction caused by not explicitly modeling cold fronts as meteorological disturbances, using fixed stations to screen multi-source data, lacking propagation chain consistency verification, not establishing a causal relationship between upstream signals and the target wind field passage time, and lacking an event-level power correction mechanism. This significantly improves the time positioning accuracy and amplitude prediction accuracy of ultra-short-term cold front-type wind power ramp prediction, enhances the reliability and timeliness of ramp event early warning, and better supports the safe and stable operation of the power grid.

[0008] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0009] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating a short-term prediction method for cold front-type wind power ramp-up provided in an embodiment of this application. Detailed Implementation

[0011] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0012] The following describes an embodiment of the cold front-type wind power ramp-up ultra-short-term prediction method of this application with reference to the accompanying drawings.

[0013] Figure 1 This is a flowchart illustrating a short-term prediction method for cold front-type wind power ramp-up provided in an embodiment of this application.

[0014] like Figure 1 As shown, the method includes the following steps: S1 acquires data on the target wind field, surrounding candidate observation nodes, and numerical weather prediction prior to the forecast release time, and performs data quality control and screening.

[0015] In some embodiments, acquiring multi-source data prior to the forecast release time is fundamental for conducting ultra-short-term forecasts of cold front-induced wind power ramp-up. All data includes only observational data obtained before the forecast release time t0 and numerical weather prediction data available at time t0, without using actual power, wind speed, or cold front passage results within the future forecast period.

[0016] The historical observation window Wh is defined as [t0-H,t0], and the future prediction period Wf is defined as (t0,t0+F). Here, H is the length of the historical window, typically 3 hours; F is the future prediction length, typically 4 hours in ultra-short-term prediction scenarios, and the sampling interval is uniformly 15 minutes. Simultaneously, the spatial range ΩR={Ni|D(Ni,F0)≤R} of candidate observation nodes around the target wind field is defined, where F0 represents the target wind field, Ni represents the i-th candidate observation node, R represents the search radius of the candidate observation node, and D(Ni,F0) represents the spatial distance between the candidate observation node Ni and the target wind field F0.

[0017] The acquired data covers four categories: actual power, basic predicted power, available capacity, grid-connected capacity, average wind speed and direction at the target wind farm, and unit operating status indicators; location, wind speed, wind direction, temperature, air pressure, gusts, precipitation, timestamps, and data quality indicators of candidate observation nodes; power, available capacity, wind speed, wind direction, and operating status indicators of neighboring wind farms; and meteorological field information related to hub height, wind speed and direction, temperature, air pressure, temperature gradient, air pressure gradient, gusts, precipitation, and cold front movement direction in numerical weather prediction.

[0018] The acquired data underwent quality control and screening. First, time alignment and sampling frequency standardization were achieved. Missing, duplicate, and outlier values ​​were processed, and abnormal communication data was marked. Wind direction variables were calculated using a circular variable method to avoid misinterpreting jumps near 0° and 360° as sudden wind direction changes. For the target wind farm and adjacent wind farms, power anomalies caused by non-meteorological factors such as power curtailment, outages, maintenance, and sudden changes in available capacity were eliminated.

[0019] Calculate the data quality factor Qi for each candidate observation node, with a value ranging from 0 to 1. Set a data quality screening threshold Qmin; when a node's Qi ≥ Qmin, it is included in the set of valid candidate observation nodes Nvalid. For example, in practical applications, a meteorological station with good data quality can have a Qi of 0.96, while a lateral node with slight data gaps has a Qi of approximately 0.72.

[0020] This step, through standardized acquisition and strict quality control of multi-source data, can effectively eliminate invalid data and non-meteorological interference, ensuring the reliability of the basic data for subsequent cold front signal identification and propagation analysis, and improving the accuracy of prediction results from the source.

[0021] S2, based on the coordinated changes of various meteorological elements of the candidate observation nodes, identify the cold front candidate signals of each candidate observation node and determine the time when the cold front signal occurs.

[0022] In some embodiments, this step is carried out within a historical observation window. Identifying candidate cold front signals and determining their occurrence time based on the coordinated changes of multiple meteorological elements at candidate observation nodes is a core prerequisite for constructing cold front propagation chains and conducting subsequent forecasts. The passage of a cold front is not a change in a single meteorological element, but rather a significant and coordinated abrupt change in multiple elements such as temperature, wind speed, wind direction, air pressure, and gusts. Local fluctuations of a single element are easily affected by local meteorological interference, and multi-element coordinated analysis can significantly improve the accuracy of cold front signal identification.

[0023] First, calculate the cold front characteristic quantities for each node, setting the calculation window length Lc, typically 60 minutes. The temperature drop ΔTi(t) is the temperature difference between the start of the window and the current time, calculated using the following formula: ΔTi(t) = Ti(t-Lc) - Ti(t); Where Lc represents the length of the cold front feature calculation window, and Ti(t) represents the temperature of the candidate observation node Ni at time t.

[0024] The wind speed increase ΔVi(t) is the difference in wind speed between the current time and the start time of the window, and is calculated using the following formula: ΔVi(t) = Vi(t) - Vi(t - Lc).

[0025] Where Lc represents the length of the cold front feature calculation window, and Vi(t) represents the wind speed at candidate observation node Ni at time t.

[0026] The angle of wind direction change is calculated using a circular variable method, and the calculation formula is as follows: ΔWDi(t)=min(|WDi(t)-WDi(t-Lc)|,360°-|WDi(t)-WDi(t-Lc)|); Where WDi(t) represents the wind direction of candidate observation node Ni at time t.

[0027] This can prevent the jump at the boundary between 0° and 360° from being misjudged as a sudden change in wind direction.

[0028] Simultaneously calculate the pressure change ΔPair,i(t) = |Pair,i(t) - Pair,i(t - Lc)| and the gust increase ΔGi(t) = Vgust,i(t) - Vi(t).

[0029] Where Pair,i(t) represents the air pressure at candidate observation node Ni at time t, and Vgust,i(t) represents the gust wind speed at candidate observation node Ni at time t, which usually refers to the instantaneous maximum wind speed within a short period of time (such as 3 seconds or 1 minute).

[0030] After standardizing the above characteristics to eliminate the influence of different dimensions, the single-node cold front signal intensity Fi(t) is constructed, and the calculation formula is as follows: Fi(t)=a1 ZΔTi(t)+a2 ZΔVi(t)+a3 ZΔWDi(t)+a4 ZΔPi(t)+a5 ZΔGi(t); Where a1 to a5 are the weighting coefficients of each cold front feature, which can be determined through statistical optimization of historical cold front events; ZΔTi(t), ZΔVi(t), ZΔWDi(t), ZΔPi(t), and ZΔGi(t) represent the standardized features of temperature drop, wind speed increase, sudden wind direction change, air pressure change, and gust enhancement, respectively.

[0031] When the cold front signal intensity Fi(t) is not less than the preset threshold θF, and at least m cold front features exceed the corresponding single threshold, the candidate observation node is determined to have a cold front candidate signal. For example, in practical applications, θF can be 0.75 and m can be 3. At the same time, the cold front signal confidence factor Bi is output, with a value ranging from 0 to 1, which is used for subsequent screening of valid upstream nodes.

[0032] To determine the occurrence time of a cold front signal, firstly, obtain the initial trigger time of the cold front signal, ti_first, which is the earliest time satisfying Fi(t) ≥ θF; then, obtain the peak time of the cold front signal within the peak search window, ti_peak. Finally, obtain the occurrence time of the cold front signal, ti_front, through weighted fusion, using the following formula: ti_front=λ ti_first+(1-λ) ti_peak; Where λ is the time-fusion weight, with a value ranging from 0 to 1. Simultaneously, the uncertainty σt,i of the cold front signal occurrence time is calculated and used for weight allocation in subsequent propagation delay fusion.

[0033] This step effectively filters out non-cold front interference signals such as local gusts and nighttime radiative cooling through multi-factor collaborative analysis, improving the reliability of cold front signal identification. By weighted fusion of the initial trigger and peak time, the actual time of the cold front's arrival at the observation node is accurately characterized, providing an accurate time reference for subsequent propagation direction estimation and transit time prediction.

[0034] S3. Based on the data from the numerical weather forecast and the order of the occurrence times of the cold front signals at multiple candidate observation nodes, determine the current propagation direction of the cold front.

[0035] In some embodiments, this step integrates numerical weather prediction information with the spatiotemporal distribution characteristics of cold front signals from observation nodes, and explicitly models the cold front as a meteorological disturbance object with a clear propagation direction. This provides a directional reference for subsequent dynamic screening of effective upstream nodes and calculation of propagation time delay, solving the problem that existing technologies implicitly contain the impact of cold fronts in independent meteorological features and cannot accurately capture their propagation patterns.

[0036] First, the predicted propagation direction of the cold front is extracted based on numerical weather prediction data. Numerical weather prediction provides spatiotemporal distribution information of large-scale meteorological fields. By analyzing the dynamic changes in temperature gradient, pressure gradient, low-level wind field, the position of the cold air advance front, and the cold front boundary in the forecast field, the predicted propagation direction of the cold front, e_NWP, can be directly obtained. The direction of the temperature gradient reflects the main advance direction of the cold air mass, the dominant wind direction of the low-level wind field is highly correlated with the direction of cold front movement, and the distribution of the pressure gradient further confirms the movement trend of the front.

[0037] Secondly, the propagation direction is inverted based on the cold front signal occurrence sequence of candidate observation nodes. The geographical coordinates of the candidate observation nodes are converted to local planar coordinates. Using the cold front signal occurrence time of each node as the dependent variable and its planar coordinates as the independent variable, the cold front arrival time plane is fitted using the following formula: ti_front=α xi+β yi+γ; Where xi and yi are the local plane coordinates of the candidate observation node Ni, and α, β, and γ are the fitting parameters of the cold front arrival time plane.

[0038] The propagation direction e_obs and initial propagation velocity Vf_obs of the cold front, derived from the observed parameters, are calculated using the following formula: e_obs=(α,β) / Sqrt(α²+β²); Vf_obs=1 / Sqrt(α²+β²); The direction of e_obs is the direction in which the cold front advances from far to near, and Vf_obs reflects the average moving speed of the cold front within the observation area.

[0039] Finally, the multi-source information is fused to obtain the final propagation direction of the current cold front. Combining the empirical propagation direction e_hist obtained from historical statistics of similar cold front events, e_NWP, e_obs, and e_hist are weighted, fused, and normalized, using the following formula: e_f=normalize(η1 e_NWP+η2 e_obs+η3 e_hist); η1, η2, and η3 are dynamic weights, which are dynamically adjusted based on the number of observation nodes, node data quality, reliability of cold front signals, consistency between numerical weather prediction and observation direction, and reliability of historical samples. For example, if the numerical weather prediction shows that the cold front is moving from northwest to southeast, and the signal occurrence order of nodes N1 and N2 is consistent with this direction, the final fused cold front propagation direction will be northwest to southeast.

[0040] This step achieves accurate estimation of the cold front propagation direction through multi-source information fusion, laying the directional foundation for subsequent dynamic screening of effective upstream nodes, ensuring the spatial rationality of the subsequent propagation chain construction, and fundamentally improving the accuracy of cold front passage time prediction.

[0041] S4. Based on the propagation direction of the cold front, dynamically select effective upstream nodes located in the effective upstream direction of the target wind field from the candidate observation nodes.

[0042] In some embodiments, this step dynamically filters observation nodes located in the effective upstream direction of the target wind field based on the current propagation direction of the cold front. This abandons the traditional method of using a fixed set of stations and a fixed distance range for filtering, and can accurately identify upstream observation points that have early warning value for the current cold front process, while excluding interference from lateral, downstream and local disturbance nodes.

[0043] First, calculate the projected distance and lateral offset distance of the candidate node relative to the target wind field. Let the position vector of the target wind field be x0, the position vector of the candidate observation node Ni be xi, and the position vector from Ni to the target wind field be ri = x0 - xi. Project the position vectors onto the cold front propagation direction ef to obtain the projected distance Di_proj = ri along the propagation direction. ef. A positive projected distance indicates that the node is located upstream of the target wind field, while a negative distance indicates it is located laterally or downstream. Simultaneously, the lateral offset distance of the node relative to the main path of the cold front propagation is calculated as Di_perp = |ri - Di_proj|. ef| reflects the degree to which a node deviates from the main propagation path of the cold front.

[0044] Further calculate the effective upstream node score Ui to comprehensively evaluate the node's early warning value for the current cold front process. The calculation formula is as follows: Ui=Ci Qi Ri Hi Bi exp(-Di_perp / Lperp) exp(-Di_proj / LD); Where Ci is the directional consistency factor, reflecting the degree of matching between the node's location and the direction of cold front propagation; Qi is the data quality factor, characterizing the completeness and accuracy of the node's observation data; Ri is the historical propagation reliability factor, obtained based on the statistical consistency of the node's signal propagation in historical cold front events; Hi is the terrain channel consistency factor, considering the influence of terrain on the cold front propagation path; Bi is the cold front signal credibility factor; Di_proj is the projected distance, Di_perp is the lateral offset distance, Lperp and LD are the lateral offset distance attenuation scale and the along-projection distance attenuation scale, respectively, controlling the attenuation effect of distance on node scores.

[0045] A valid upstream node scoring threshold θU is set. When a node's Ui ≥ θU and its projection distance is within a preset reasonable range, it is determined to be a valid upstream node under the current cold front process and included in the valid upstream node set Nup. For example, in practical applications, θU can be 0.70, Ui of far upstream nodes with good data quality and located on the main path of the cold front can reach 0.78, and Ui of mid-upstream nodes can reach 0.84. The valid upstream node set is dynamically updated according to the cold front propagation direction, path, real-time signal strength of nodes, and data quality, rather than a fixed list of sites.

[0046] This step effectively eliminates nodes with excessive lateral offset, those located downstream, or those with poor data quality through a dynamic screening mechanism. This ensures that the observational data input into subsequent propagation chain analysis all come from upstream regions that have causal early warning significance for the current cold front process, significantly improving the accuracy of subsequent propagation speed estimation and cold front passage time prediction.

[0047] S5. Sort the effective upstream nodes according to their relative positions with the target wind field in the propagation direction, construct the cold front meteorological observation propagation chain, and perform consistency verification on the propagation chain, removing or downgrading nodes that do not meet the verification.

[0048] In some embodiments, this step of constructing the cold front meteorological observation propagation chain and performing consistency verification is a key step in verifying the spatiotemporal propagation law of cold front signals, filtering local disturbances and anomalous data, and ensuring that subsequent propagation delay calculations and transit time predictions are based on the actual cold front propagation process.

[0049] The nodes in the effective upstream node set are sorted in descending order of their projected distances from the cold front to the target wind field along the propagation direction, constructing a cold front meteorological observation propagation chain that includes far-upstream nodes, near-upstream nodes, adjacent wind farm nodes, and the target wind field. The chain structure of the propagation chain intuitively reflects the spatial path of the cold front advancing towards the target wind field from far to near. For example, in the process of a cold front propagating from northwest to southeast, the far-upstream meteorological station with a projected distance of 220 km, the mid-upstream meteorological station with a projected distance of 150 km, the near-upstream node with a projected distance of 85 km, and the target wind field are arranged sequentially to form a complete propagation link.

[0050] A dual consistency check is performed on the propagation chain. The first is a timing consistency check, verifying that the occurrence time of the cold front signal between adjacent nodes meets the requirement that the signal occurrence time of the far upstream node is no later than that of the near upstream node, allowing for a time error not exceeding a preset value εt, ensuring that the signal time sequence is consistent with the cold front propagation logic. The second is a local propagation speed rationality check, calculating the local propagation speed between adjacent nodes using the following formula: Vlocal=(Dkm_proj-Dkm1_proj) / (t_km1_front-t_km_front); Where Dkm_proj and Dkm1_proj are the projected distances from two adjacent nodes along the propagation direction to the target wind field, respectively, and t_km_front and t_km1_front are the times when the cold front signal occurs at the corresponding nodes. The system determines whether the local propagation speed is within a preset reasonable range of 30km / h to 120km / h. For example, in a typical cold front process, a local propagation speed of 76.4km / h between adjacent nodes is within this reasonable range.

[0051] The number of adjacent node pairs that simultaneously satisfy both verification criteria is counted, and the propagation chain consistency factor Mchain = Nconsistent / Npair is calculated, where Nconsistent is the number of node pairs that meet the criteria, and Npair is the total number of adjacent node pairs in the propagation chain. When Mchain is close to 1, it indicates that the current signal has a clear upstream-to-downstream propagation characteristic; when Mchain is lower than a preset threshold, it indicates the presence of abnormal nodes or local disturbances, and nodes that do not meet the verification criteria should be removed or downweighted, and the propagation chain should be reconstructed if necessary.

[0052] This step, through dual consistency verification, effectively distinguishes between the actual cold front propagation and local meteorological and data anomalies, significantly reducing the risk of false alarms and missed alarms for climbing events, ensuring the spatiotemporal rationality of the propagation chain, and providing a reliable foundation for subsequent accurate prediction of the cold front's passage time.

[0053] S6. Based on the occurrence time of the cold front signal of the effective upstream node and its propagation time delay to the target wind field, predict the passage time of the cold front of the target wind field.

[0054] In some embodiments, this step accurately predicts the time of cold front passage of the target wind field by fusing multi-source propagation time delay information. This time serves as the core time anchor point for subsequent power ramping event prediction and curve correction, effectively solving the problem of time positioning deviation of ramping events in traditional prediction.

[0055] First, the three types of propagation delays for each effective upstream node are calculated. The real-time observation propagation delay τi_real is calculated based on the real-time propagation velocity of adjacent nodes in the propagation chain and the projected distance from that node to the target wind field, directly reflecting the actual advancing speed of the current cold front. The historical propagation delay τi_hist for similar cold fronts is obtained by statistically analyzing the average propagation delay from that node to the target wind field in historical cold front events of similar intensity and season, reflecting the statistical regularity of regional cold front propagation. The numerical weather prediction propagation delay τi_NWP is estimated based on the cold front position, movement speed, cooling zone, or advection velocity of the low-level wind field in the numerical weather prediction, providing the propagation trend under a large-scale meteorological background.

[0056] The three types of propagation delays are dynamically fused to obtain the fused propagation delay from the effective upstream node to the target wind field, as shown in the formula: τi=β1,i τi_real+β2,i τi_hist+β3,i τi_NWP; Wherein β1,i, β2,i, and β3,i are dynamic fusion weights, and their sum is 1. The weights are dynamically adjusted based on propagation chain consistency, the number of effective upstream nodes, the stability of real-time propagation speed, historical time delay dispersion, and the consistency between numerical weather prediction and observation direction. For example, in a scenario where the propagation chain consistency factor is 1.00 and the node data quality is good, the real-time propagation time delay weight can be set to 0.5 or higher.

[0057] Based on the occurrence time of the cold front signal from the effective upstream node and the fusion propagation time delay, the estimated time of the cold front passage of the target wind field from this node is obtained: t0i_front_hat=ti_front+τi; Where ti_front represents the time when the cold front signal occurs at that node. For example, if the signal from a distant upstream node occurs at 06:50 and the fusion propagation delay is 173.8 minutes, then the estimated transit time is approximately 09:44.

[0058] Finally, the prediction results from multiple valid upstream nodes are weighted and fused to obtain the final target wind field cold front passage time: t0_front_hat=Σ(Wi t0i_front_hat) / ΣWi; Where Wi is the fusion weight of each node, which is determined by the effective upstream node score, the credibility of the cold front signal, and the propagation delay uncertainty.

[0059] This step, through multi-source time-delay fusion, fully combines the timeliness of real-time observation, the regularity of historical statistics, and the globality of numerical forecasting. It explicitly establishes the causal relationship between upstream signals and the passing time of the target wind field, enabling accurate prediction of the arrival time of cold fronts tens of minutes in advance, and providing a reliable time reference for power grid dispatch.

[0060] S7. Using the time of the cold front passing through the target wind field as the time anchor point, perform event-level correction on the basic wind power prediction curve related to the cold front-type climbing event to obtain the corrected wind power prediction result.

[0061] In some embodiments, this step uses the predicted time of the cold front passing through the target wind field as the core time anchor point to perform event-level correction on the basic wind power prediction curve. Unlike the uniform scaling, fixed amplitude limit or threshold correction of traditional methods, the correction process has a clear meteorological propagation causal relationship and accurate time positioning, which can significantly improve the time accuracy and amplitude accuracy of power prediction during the cold front passing period.

[0062] When the cold front passes through the target wind farm within the predicted future timeframe, and the cold front intensity, propagation chain consistency, and future wind speed changes at the target wind farm meet preset conditions, a cold front-type wind power ramp-up risk is identified. Based on the cold front's passage time and combined with historical statistics of similar cold front events, a typical pre-climb correction value of 15 minutes is determined, and a typical post-climb correction value of 60 minutes is determined. The ramp-up start time, end time, and duration are then calculated.

[0063] When calculating the predicted ramp amplitude, the representative wind speeds before and after the cold front are first obtained based on the future predicted meteorological sequence of the target wind field. Then, using the wind turbine power curve mapping function and the current available capacity, the equivalent power before and after the cold front and the power change ΔPcf based on the cold front meteorological response are obtained. Finally, the predicted ramp amplitude is obtained by combining the power change ΔPbase of the baseline prediction curve within the corresponding window. The calculation formula is as follows: ΔP_ramp_hat=(1-μ) ΔPbase+μ ΔPcf; Where μ is the cold front correction weight. The stronger the cold front signal and the higher the consistency of the propagation chain, the larger the value of μ. In high-confidence scenarios, it can be taken as 0.65. At the same time, the predicted maximum ramp rate is calculated. When the ramp amplitude or the maximum ramp rate exceeds the preset threshold, it is judged as a significant cold front ramp event.

[0064] A cold front-type ramp correction window is defined as the predicted ramp start and end time interval, and corrections are only performed within this window. A normalized ramp shape function is constructed, preferentially using an S-shaped function that can fit smooth power changes. A power reference curve for the cold front event is constructed using the baseline predicted average power before the cold front as the power anchor point. The cold front correction intensity coefficient λcf is determined based on factors such as the overall cold front intensity and propagation chain consistency. Within the correction window, the baseline predicted power and the reference curve are weighted and fused to obtain the corrected output power.

[0065] Physical boundary constraints are applied to the corrected power results to ensure that the output power is not less than 0 and does not exceed the estimated available capacity for future periods. The final output is the corrected power prediction curve, along with scheduling information such as the cold front passage time, ramp-up event parameters, and prediction confidence levels.

[0066] This step accurately locates the time interval for power ramp-up, effectively correcting the problem of the overall advance or lag of the basic prediction curve, while improving the accuracy of ramp-up amplitude prediction, providing reliable power prediction results and event-level early warning information for power grid dispatch.

[0067] Compared with related technologies, in this embodiment, data on the target wind field, surrounding candidate observation nodes, and numerical weather prediction prior to the forecast release time are acquired, and data quality control and screening are performed. Based on the coordinated changes of various meteorological elements of the candidate observation nodes, cold front candidate signals for each candidate observation node are identified, and the occurrence time of the cold front signal is determined. According to the numerical weather prediction data and the order of the occurrence times of the cold front signals of multiple candidate observation nodes, the propagation direction of the current cold front is determined. Based on the propagation direction of the cold front, the candidate observation nodes located at the... Effective upstream nodes in the effective upstream direction of the target wind field are identified; these effective upstream nodes are sorted according to their relative positions to the target wind field in the propagation direction to construct a cold front meteorological observation propagation chain, and the propagation chain is checked for consistency, eliminating or downweighting nodes that do not meet the check; based on the occurrence time of the cold front signal of the effective upstream nodes and its propagation time lag to the target wind field, the cold front passage time of the target wind field is predicted; using the cold front passage time of the target wind field as the time anchor point, the basic wind power prediction curve is modified at the event level related to the cold front-type climbing event to obtain the modified wind power prediction result. This technology can solve the problems of insufficient time positioning and amplitude accuracy in cold front wind power ramp prediction caused by the failure to explicitly model cold fronts as meteorological disturbances, the use of fixed sites to screen multi-source data, the lack of propagation chain consistency verification, the failure to establish the causal relationship between upstream signals and the passing time of the target wind field, and the lack of event-level power correction mechanisms. It can significantly improve the time positioning accuracy and amplitude prediction accuracy of ultra-short-term cold front wind power ramp prediction, enhance the reliability and timeliness of ramp event early warning, and better support the safe and stable dispatch and operation of the power grid.

[0068] As a specific implementation of this application, based on the basic scheme, the method further defines the coordinated changes of multiple meteorological elements based on the candidate observation nodes to identify cold front candidate signals for each candidate observation node, including: For each candidate observation node, calculate its temperature drop, wind speed increase, wind direction change angle, air pressure change, and gust increase, and standardize the above features. The standardized features are weighted and summed to construct the single-node cold front signal intensity. When the intensity of the cold front signal is not less than a preset threshold and the number of features that undergo coordinated changes exceeds a preset number, it is determined that the candidate observation node has a cold front candidate signal.

[0069] Specifically, this implementation defines a method for identifying candidate cold front signals. By quantifying the coordinated changes of multiple meteorological elements, it achieves accurate identification of cold front signals, effectively filters out local interference caused by fluctuations of a single element, and provides a reliable node signal foundation for the subsequent propagation chain construction.

[0070] For each candidate observation node, five core cold front characteristic quantities are calculated within a historical observation window, with the calculation window length set at 60 minutes. The temperature drop ΔTi(t) is the temperature difference between the start of the window and the current time, expressed as ΔTi(t) = Ti(t-Lc) - Ti(t), reflecting the cooling effect brought by the passage of the cold front. The wind speed increase ΔVi(t) is the wind speed difference between the current time and the start of the window, expressed as ΔVi(t) = Vi(t) - Vi(t-Lc), reflecting the wind speed enhancement characteristic accompanying the cold front. The wind direction change angle is calculated using a circular variable method, expressed as ΔWDi(t) = min(|WDi(t) - WDi(t-Lc)|, 360° - |WDi(t) - WDi(t-Lc)|), to avoid numerical jumps at the 0° and 360° boundaries being misjudged as sudden wind direction changes. Simultaneously, the pressure change ΔPair,i(t) = |Pair,i(t) - Pair,i(t - Lc)| and the gust enhancement ΔGi(t) = Vgust,i(t) - Vi(t) are calculated to characterize the sudden pressure change and gust enhancement phenomena during the passage of a cold front, respectively.

[0071] The five types of features mentioned above are standardized to eliminate the influence of different physical dimensions and numerical ranges, resulting in a standardized feature sequence. The standardized features are then weighted and summed to construct the single-node cold front signal intensity, using the formula: Fi(t) = a1 ZΔTi(t)+a2 ZΔVi(t)+a3 ZΔWDi(t)+a4 ZΔPi(t)+a5 ZΔGi(t); Among them, a1 to a5 are the weight coefficients of each cold front feature, which can be determined by statistical optimization through historical cold front event samples. Usually, the weights of temperature drop and wind speed increase are relatively high to highlight the core meteorological features of the cold front.

[0072] A cold front signal intensity threshold θF = 0.75 is set, along with a threshold of 3 for the number of co-variation features. A candidate observation node is considered to have a cold front candidate signal when its cold front signal intensity Fi(t) is not less than 0.75 and at least three cold front features exceed their respective thresholds. For example, in a practical application, a distant upstream meteorological station experiences a 4.8℃ temperature drop, a 5.5m / s wind speed increase, and a 78° abrupt wind direction change within 60 minutes. The calculated cold front signal intensity is 0.93, meeting the criteria and thus identified as having a cold front candidate signal.

[0073] This implementation effectively avoids misjudgments caused by fluctuations in a single meteorological element through a multi-element collaborative judgment mechanism. It can distinguish between the actual passage of a cold front and non-cold front meteorological phenomena such as local gusts and nighttime radiative cooling, significantly improving the reliability and stability of cold front signal identification.

[0074] As a specific implementation of this application, based on the basic solution, the determination of the occurrence time of the cold front signal is further defined, including: The determination is based on a weighted fusion of one or more of the following: the initial trigger time of the cold front signal, the peak time of the cold front signal, and the trigger time of multi-feature fusion.

[0075] Specifically, this implementation defines a method for determining the occurrence time of a cold front signal. By weighted fusion of multiple time points, it accurately represents the actual time when the cold front arrives at the observation node, solving the problem that single-time point determination is easily affected by signal fluctuations and has large time positioning deviations. This provides an accurate time reference for subsequent propagation direction estimation, propagation delay calculation, and target wind field passage time prediction.

[0076] The first trigger time of the cold front signal, ti_first, refers to the moment when the intensity of the cold front signal Fi(t) at the candidate observation node first reaches the preset threshold θF, i.e., ti_first = min{t|Fi(t)≥θF}, reflecting the initial time when the cold front begins to have a significant impact on the node. The peak time of the cold front signal, ti_peak, refers to the moment when the intensity of the cold front signal reaches its maximum value within the peak search window, i.e., ti_peak = argmaxFi(t), t∈Wpeak. The peak search window Wpeak typically starts from the first trigger time and lasts for 60 minutes, reflecting the time when the cold front has the strongest impact on the node. The multi-feature fusion trigger time, ti_multi, is a comprehensive trigger time obtained by extracting the trigger times of five features—temperature drop, wind speed increase, sudden wind direction change, air pressure change, and gust enhancement—and then using the median or weighted average. This reduces the random error of trigger times based on single features.

[0077] Based on the actual application scenario, one or more of the above times are selected for weighted fusion to obtain the final cold front signal occurrence time ti_front. When only the initial trigger time and peak time are fused, the calculation formula is ti_front = λ. ti_first+(1-λ) ti_peak, where λ is the time-based fusion weight, ranging from 0 to 1. When the cold front signal is strong and the change is steep, λ can be 0.3-0.4, emphasizing the accuracy of the peak time; when the cold front signal is weak and the change is gradual, λ can be 0.6-0.7, emphasizing the timeliness of the warning at the first trigger time. If a multi-feature fusion trigger time is introduced, it can be expanded to ti_front=λ1 ti_first+λ2 ti_peak+λ3 ti_multi, where λ1+λ2+λ3=1, and the weights are dynamically adjusted based on the historical recognition reliability of each feature.

[0078] Simultaneously, the uncertainty σt,i of the cold front signal occurrence time is calculated. Its value is positively correlated with the signal strength fluctuation and the dispersion of the trigger time; the more stable the signal, the smaller the uncertainty. This uncertainty is used for node weight allocation in subsequent propagation delay fusion. Nodes with smaller uncertainties have higher weights in the prediction of the passage time.

[0079] This implementation, through the fusion and optimization of multiple time dimensions, comprehensively reflects the initiation and peak characteristics of the cold front's impact, effectively reducing the time deviation caused by signal noise and single feature anomalies, controlling the identification error of the cold front signal occurrence time to within 10 minutes, and significantly improving the time accuracy of subsequent propagation analysis.

[0080] As a specific implementation of this application, based on the basic scheme, the step of dynamically selecting effective upstream nodes located in the effective upstream direction of the target wind field from the candidate observation nodes according to the propagation direction of the cold front includes: Calculate the score of the effective upstream node. When the score of the effective upstream node Ui is not less than a preset threshold, the node is determined to be a valid upstream node. The formula for calculating the effective upstream node score is as follows: Ui=Ci·Qi·Ri·Hi·Bi·exp(-Di_perp / Lperp)·exp(-Di_proj / LD); Where Ci is the directional consistency factor, Qi is the data quality factor, Ri is the historical propagation reliability factor, Hi is the terrain channel consistency factor, Bi is the cold front signal reliability factor, Di_proj is the projection distance, Di_perp is the lateral offset distance, and Lperp and LD are the lateral and along-the-path distance attenuation scales, respectively.

[0081] Specifically, this implementation defines a dynamic screening method for effective upstream nodes. By constructing a comprehensive scoring model through multi-dimensional quantitative indicators, it accurately identifies upstream observation nodes that have early warning value for the current cold front process, solving the problem that the traditional fixed-site screening mode cannot adapt to the dynamic propagation characteristics of cold fronts and is prone to introducing invalid interference nodes.

[0082] First, calculate the projected distance and lateral offset distance of the candidate node relative to the target wind field. Project the position vector of the candidate node pointing to the target wind field onto the cold front propagation direction to obtain the projected distance Di_proj along the propagation direction. A positive projected distance indicates that the node is located upstream. The lateral offset distance Di_perp reflects the degree to which the node deviates from the main propagation path of the cold front. The larger the value, the lower the warning value of the node for the current cold front process.

[0083] Construct an effective comprehensive scoring model for upstream nodes, and calculate the formula as follows: Ui=Ci Qi Ri Hi Bi exp(-Di_perp / Lperp) exp(-Di_proj / LD); Ci is the directional consistency factor, ranging from 0 to 1, representing the degree of matching between the node's location and the direction of cold front propagation. Nodes located on the main propagation axis of the cold front have a Ci value close to 1. Qi is the data quality factor, calculated based on node data integrity, anomaly rate, and communication stability. Meteorological stations with good data quality can achieve a Qi of 0.96, while lateral nodes with slight data gaps have a Qi of approximately 0.72. Ri is the historical propagation reliability factor, obtained by statistically analyzing the temporal correlation between the node's signal and the cold front's passage in historical cold front events. Hi is the terrain channel consistency factor, considering the guiding effect of terrain such as mountains and valleys on the cold front propagation path; nodes located within the terrain channel have a higher Hi. Bi is the cold front signal credibility factor, ranging from 0 to 1, determined by the cold front signal intensity and the number of co-variation features. Di_proj is the projected distance, Di_perp is the lateral offset distance, and Lperp and LD are the attenuation scales for lateral offset distance and along-projection distance, respectively, with typical values ​​of 50km and 300km, controlling the attenuation effect of distance on node scores.

[0084] A threshold of θU = 0.70 is set for the scoring of effective upstream nodes. A node is considered an effective upstream node in the current cold front process if its Ui ≥ 0.70 and its projected distance is within the range of 0 to 250 km. For example, in a cold front propagating from northwest to southeast, the far upstream node with a projected distance of 220 km has a Ui = 0.78, the mid-upstream node with a Ui = 0.84 (150 km), and the near upstream node with a Ui = 0.76 (85 km), all meeting the threshold requirements and included in the effective upstream node set. Nodes with a lateral offset of 82 km have a Ui = 0.21, and downstream nodes have a Ui = 0.00, and are directly removed. The effective upstream node set is dynamically updated according to the cold front propagation direction, path, real-time signal strength, and data quality of the nodes.

[0085] This implementation achieves precise and dynamic screening of effective upstream nodes through a multi-dimensional comprehensive scoring mechanism, effectively eliminating interference from lateral, downstream, and low-quality nodes, ensuring that subsequent propagation chain analysis is based on upstream observation data with causal early warning significance, and significantly improving the accuracy of cold front passage time prediction.

[0086] As a specific implementation of this application, based on the basic scheme, the following further limitation is made to the consistency verification of the propagation chain, and the removal or demotion of nodes that do not meet the verification requirements, including: Verify whether the occurrence time of the cold front signal of adjacent nodes in the propagation chain satisfies the condition that the occurrence time of the signal of the far upstream node is not later than the occurrence time of the signal of the near upstream node; Calculate the local propagation speed between adjacent nodes in the propagation chain, and verify whether the local propagation speed is within the preset reasonable speed range; Only adjacent node pairs that satisfy both of the above conditions are considered to satisfy the propagation chain consistency.

[0087] Specifically, this implementation defines a consistency verification method for the cold front meteorological observation propagation chain. Through a dual verification mechanism of timing and velocity, it rigorously verifies the spatiotemporal propagation logic of the cold front signal, effectively distinguishes the real cold front propagation process from local meteorological disturbances and data anomalies, and fundamentally reduces the risk of false alarms and missed alarms for climbing events.

[0088] First, a time-series consistency check is performed. As a moving meteorological front, the signal of a cold front inevitably follows a spatiotemporal pattern of appearing sequentially from far upstream to near upstream. The timing of the cold front signal occurrence at each pair of adjacent nodes in the propagation chain is checked, requiring that the signal occurrence time of the far upstream node is no later than that of the near upstream node, with a time error allowed to be no more than 5 minutes, to compensate for sampling deviations in the observation data and minor delays in signal identification. If, in a pair of adjacent nodes, the near upstream node's signal occurs earlier than the far upstream node's, it indicates that the signal at at least one node does not originate from the current cold front propagation, and may be caused by local gusts, nighttime radiative cooling, or anomalies in station data.

[0089] Next, the rationality of the local propagation speed is verified. The local propagation speed between adjacent nodes is calculated using the following formula: Vlocal=(Dkm_proj-Dkm1_proj) / (t_km1_front-t_km_front); Where Dkm_proj and Dkm1_proj are the projected distances from two adjacent nodes along the cold front propagation direction to the target wind field, respectively, and t_km_front and t_km1_front are the times when the cold front signal occurs at the corresponding nodes. Based on the regional cold front climatological statistics, a reasonable propagation speed range is preset to be 30 km / h to 120 km / h. For example, for a winter cold front propagating from northwest to southeast, a local propagation speed of 76.4 km / h between adjacent nodes is within the reasonable range; if the calculated local propagation speed is lower than 30 km / h or higher than 120 km / h, it is determined to be inconsistent with physical laws, excluding the possibility of a real cold front propagation.

[0090] Only adjacent node pairs that simultaneously satisfy both temporal consistency and reasonable local propagation speed are considered to meet the propagation chain consistency criteria. The number of node pairs meeting the criteria is counted, and the propagation chain consistency factor Mchain = Nconsistent / Npair is calculated, where Nconsistent is the number of qualified node pairs, and Npair is the total number of adjacent node pairs in the propagation chain. Nodes that do not meet the criteria are directly removed from the propagation chain or given extremely low weight in subsequent calculations. If Mchain is below a preset threshold of 0.6, the current signal is determined not to have clear cold front propagation characteristics, and a high-confidence ramp-up warning is not triggered.

[0091] This implementation effectively filters out non-cold front interference signals through dual physical constraint verification, ensuring the spatiotemporal rationality of the propagation chain construction. It provides a reliable chain-like data foundation for subsequent propagation delay calculation and cold front passage time prediction, significantly improving the credibility of the prediction results.

[0092] As a specific implementation of this application, based on the basic scheme, the calculation method for the propagation time delay of the target wind field is further defined as follows: τi=β1, i ·τi_real+β2, i ·τi_hist+β3, i ·τi_NWP; Where τi_real is the real-time observation propagation time lag, τi_hist is the historical propagation time lag of similar cold fronts, τi_NWP is the numerical weather prediction propagation time lag, and β1, i β2, i β3, i The weights are dynamically fused, and the sum of the three is 1.

[0093] Specifically, this implementation defines a fusion calculation method for the propagation time delay from the effective upstream node to the target wind field. By integrating three types of time delay information—real-time observation, historical statistics, and numerical weather prediction—it solves the problem that time delays from a single source are susceptible to accidental factors and lack robustness, providing reliable time parameters for accurately predicting the passage time of cold fronts in the target wind field.

[0094] The real-time propagation delay τi_real directly reflects the actual advancement state of the current cold front, calculated based on the real-time propagation speed of adjacent nodes in the propagation chain and the projected distance from that node to the target wind field. For example, in a cold front propagating from northwest to southeast, the projected distance from the far upstream node N1 to the target wind field is 220 km, and the measured real-time propagation speed of the propagation chain is 76.4 km / h. Therefore, τi_real = 220 / 76.4h ≈ 173 min. This delay has the strongest timeliness and can accurately capture the dynamic changes of this cold front.

[0095] The historical propagation delay τi_hist of similar cold fronts reflects the statistical regularity of cold front propagation in the region. It is obtained by selecting cold front events of the same season, similar intensity, and similar propagation direction in history, and calculating the average propagation delay from the node to the target wind field. For example, during a winter cold front in this region, the historical average propagation delay from node N1 to the target wind field is approximately 176 minutes. This delay has good stability and can compensate for errors caused by fluctuations in real-time observation data.

[0096] The propagation lag τi_NWP of numerical weather prediction provides the propagation trend under a large-scale meteorological background. It is estimated based on the cold front boundary position, frontal movement speed, cooling belt advance speed, or low-level wind field advection speed output by the numerical weather prediction. For example, in this process, the numerical weather prediction forecasts a cold front propagation lag of 175 minutes from node N1 to the target wind field. This lag can cover areas with sparse observation nodes and supplement the deficiencies of local observations.

[0097] The three types of time delays are dynamically weighted and fused using the following formula: τi=β1,i τi_real+β2,i τi_hist+β3,i τi_NWP; Where β1,i+β2,i+β3,i=1. The dynamic weights are adaptively adjusted based on propagation chain consistency, the number of effective upstream nodes, the stability of real-time propagation speed, historical time delay dispersion, and the consistency between numerical weather prediction and observation directions. When the propagation chain consistency factor is 1.00 and the node data quality is good, the real-time observation time delay weight can be set to 0.55, the historical time delay weight to 0.25, and the numerical weather prediction time delay weight to 0.20. For example, the propagation time delay τi after the fusion of the above N1 nodes is approximately 173.8 min, which is 0.55×173+0.25×176+0.20×175.

[0098] This implementation, through the dynamic fusion of multi-source time delays, fully combines the timeliness of real-time observation, the regularity of historical statistics, and the globality of numerical forecasting. It effectively reduces the error of single-source time delays, controls the calculation error of propagation time delays to within 5%, and significantly improves the accuracy and robustness of predicting the passage time of cold fronts in the target wind field.

[0099] As a specific implementation of this application, based on the basic scheme, the parameters of the cold front-type wind power ramping event are further defined as follows: Climb start time, climb end time, climb direction, climb amplitude, and maximum climb rate; The formula for predicting the slope amplitude is as follows: ΔP_ramp_hat=(1-μ)·ΔPbase+μ·ΔPcf, Wherein, ΔP_ramp_hat is the ramp amplitude, ΔPbase is the power change of the basic prediction curve within the window before and after the cold front, ΔPcf is the power change estimated based on the meteorological response of the cold front, and μ is the cold front correction weight.

[0100] Specifically, this implementation defines the core parameters and prediction methods for the ramp amplitude of cold front-type wind power ramp events. By integrating basic prediction results with cold front meteorological response characteristics, it achieves accurate quantification of key parameters of ramp events, providing a clear basis for subsequent event-level correction of power curves.

[0101] The core parameters for cold front-type wind power ramping events include ramp start time, ramp end time, ramp direction, ramp amplitude, and maximum ramp rate. The ramp start and end times are calculated based on the time of the cold front's passage through the target wind farm. Combining this with statistical patterns from historical similar cold front events, a 15-minute pre-ramp correction and a 60-minute post-ramp correction are used. That is, the ramp start time is the time of the cold front's passage minus 15 minutes, and the ramp end time is the time of the cold front's passage plus 60 minutes; the difference between the two is the ramp duration. The ramp direction is determined by the sign of the equivalent power difference before and after the cold front; a positive difference indicates an upward ramp, and a negative difference indicates a downward ramp. The maximum ramp rate is the ratio of the predicted ramp amplitude to the ramp duration, typically expressed in megawatts per 15 minutes, for easy reference by the grid dispatching system.

[0102] The prediction of the slope amplitude comprehensively considers both the basic forecast results and the characteristics of the cold front meteorological response. The calculation formula is as follows: ΔP_ramp_hat=(1-μ) ΔPbase+μ ΔPcf; ΔPbase represents the power change based on the prediction curve within the window before and after a cold front, reflecting the overall assessment of power changes by the existing prediction model. ΔPcf represents the power change estimated based on the meteorological response to the cold front. It is calculated by first obtaining the representative wind speed before and after the cold front based on the future predicted meteorological sequence of the target wind field, and then using the wind turbine power curve mapping function combined with the current available capacity to obtain the equivalent power before and after the cold front. The difference between the two is ΔPcf. For example, in a typical uphill climb, the representative wind speed before the cold front is 4.8 m / s, corresponding to an equivalent power of 12 MW, and the representative wind speed after the cold front is 11.2 m / s, corresponding to an equivalent power of 82 MW. Therefore, ΔPcf = 70 MW.

[0103] μ is the cold front correction weight, ranging from 0 to 1, and is dynamically adjusted based on the cold front signal strength, propagation chain consistency, and the number of effective upstream nodes. The stronger the cold front signal and the higher the propagation chain consistency, the larger the value of μ, which can be 0.65 in high-confidence scenarios. When the propagation chain consistency is low or the cold front signal is weak, μ decreases accordingly, relying more on the base prediction results. For example, in a certain process, ΔPbase=45MW, ΔPcf=70MW, μ=0.65, then the predicted ramp amplitude ΔP_ramp_hat=(1-0.65)×45+0.65×70=61.25MW.

[0104] This implementation effectively corrects the underestimation or overestimation of the power change amplitude of cold fronts by integrating the statistical regularity of basic forecasts with the physical characteristics of cold front meteorological response. It controls the prediction error of the ramp amplitude within 15%, and provides a complete ramp event parameter system for power grid dispatch, significantly improving the scientificity and accuracy of dispatch decisions.

[0105] As a specific implementation of this application, based on the basic scheme, the event-level correction of the base wind power prediction curve related to cold front-type ramping events is further defined, including: Construct a cold front-type climbing correction window Wcf=[t_start_hat,t_end_hat], where t_start_hat and t_end_hat are the predicted start and end times of the climb, respectively; Construct a normalized climbing shape function Kn( t ), and construct a cold front event power reference curve Pcf( based on the ramp amplitude ΔP_ramp_hat and the power anchor point Panchor. t =Panchor + ΔP_ramp_hat·Kn( t ); Within the correction window Wcf, the base predicted power Ppred is adjusted according to the cold front correction intensity coefficient λcf. t ) and the cold front event power reference curve Pcf( t The corrected output predicted power Pout is obtained by performing a fusion correction. t )=(1-λcf)·Ppred( t )+λcf·Pcf( t ).

[0106] Specifically, this implementation defines an event-level correction method for the basic wind power forecast curve. It constructs a precise correction window and power reference curve with the passing time of a cold front as the core anchor point, and achieves targeted correction of the basic forecast through weighted fusion. This solves the problems of traditional uniform scaling and fixed threshold correction methods lacking meteorological causal basis and having large time positioning deviations.

[0107] First, a cold front-type ramp correction window is constructed. The time range of the correction window is from the predicted ramp start time to the ramp end time, i.e., Wcf=[t_start_hat,t_end_hat]. The correction is only performed within this window; the power prediction results outside the window remain unchanged from the base prediction value, avoiding unnecessary interference with the prediction results during non-cold front periods. For example, in a typical uphill ramp process, if the predicted ramp start time is 09:28 and the ramp end time is 10:43, then the correction window is set to [09:28,10:43].

[0108] Next, a power reference curve for the cold front event is constructed. A normalized ramp shape function Kn(t) is used, with an sigmoid function preferred to fit the smooth power change, accurately simulating the gradual power variation during the cold front's passage. The power anchor point Panchor is selected as the average power predicted before the cold front, representing the baseline power level before the cold front's influence. Combined with the predicted ramp amplitude ΔP_ramp_hat, a power reference curve for the cold front event is constructed, with the formula Pcf(t) = Panchor + ΔP_ramp_hat. Kn(t). For example, if the power anchor point is 15MW and the ramp amplitude is 61.25MW, and the normalized ramp shape function value is 0.65 at 10:15, then the reference power at that time is 15 + 61.25 × 0.65 = 54.81MW.

[0109] Finally, weighted fusion correction is performed within the correction window. The cold front correction intensity coefficient λcf is determined based on factors such as the overall cold front intensity, propagation chain consistency, and prediction confidence, with a value ranging from 0 to 1. The stronger the cold front signal and the higher the propagation chain consistency, the larger the value of λcf, which can be taken as 0.78 in high-confidence scenarios. The correction formula is Pout(t) = (1 - λcf). Ppred(t)+λcf Pcf(t). For example, if the base predicted power at 10:15 is 47MW and λcf=0.78, then the corrected output power is 0.22×47+0.78×54.81≈53.09MW. Physical boundary constraints are applied to the corrected power to ensure that it is not less than 0 and does not exceed the estimated available capacity for future periods.

[0110] This implementation precisely adjusts the power prediction curve during cold front passage periods through event-level targeted correction, while retaining the advantages of basic prediction during non-cold front periods. It significantly improves the timing and amplitude prediction accuracy of ramp events, providing more reliable power prediction results for power grid dispatch.

[0111] As a specific implementation of this application, the method embodiment further includes, based on the basic solution, the following additional features: Based on the number of effective upstream nodes, node data quality, cold front signal strength, propagation chain consistency factor, cold front direction consistency, and propagation time delay uncertainty, the confidence level of cold front-type ramp prediction is calculated, and the intensity or method of ramp warning is adjusted accordingly.

[0112] Specifically, this implementation adds a calculation and early warning adjustment mechanism for the confidence level of cold front-type ramp prediction, which can quantify the reliability of the prediction results, provide hierarchical early warning information for power grid dispatch, and avoid mis-dispatch or unnecessary reserve capacity configuration caused by low confidence signals.

[0113] The prediction confidence score is calculated based on six core dimensions, ranging from 0 to 1. A larger number of effective upstream nodes and a more complete coverage of the cold front propagation path result in higher prediction confidence. Typically, at least two high-quality effective upstream nodes are required to trigger a high-confidence warning. Node data quality is weighted based on the data quality factor Qi; better data integrity and lower anomaly rates contribute more to the confidence score. Higher cold front signal strength and a greater number of features satisfying coordinated changes result in stronger signal credibility and a correspondingly higher confidence score. The propagation chain consistency factor Mchain is a core influencing factor; the closer Mchain is to 1, the more the signal conforms to the spatiotemporal propagation patterns of the actual cold front, and the higher the confidence score. Cold front direction consistency reflects the degree of matching between numerical weather prediction, observational inversion, and historical experience propagation directions; better consistency leads to higher prediction reliability. Propagation time delay uncertainty is calculated based on the dispersion of the target wind field's passage time estimated by each effective upstream node; smaller dispersion indicates higher confidence.

[0114] For example, in a typical uphill climb, there are two effective upstream nodes with data quality of 0.96 and 0.94 respectively, cold front signal strength of 0.93 and 0.88 respectively, propagation chain consistency factor Mchain=1.00, numerical weather prediction and observation direction are completely consistent, the transit time estimated by the two nodes differs by only 1 minute, and the prediction confidence is calculated to be 0.86.

[0115] The intensity and method of the ramp warning are dynamically adjusted based on the prediction confidence level. When the confidence level is ≥0.8, it is judged as a high-confidence warning, and a formal cold front ramp warning is issued, using a higher cold front correction intensity coefficient to correct the power curve; when the confidence level is 0.5≤confidence<0.8, it is judged as a medium-confidence concern, and a cold front concern alert is issued, using a medium-intensity correction; when the confidence level is <0.5, it is judged as a low-confidence signal, and no warning is issued, only the basic prediction result is retained or a very weak correction is made.

[0116] This implementation achieves a hierarchical early warning and correction strategy by quantifying the reliability of prediction results, effectively reducing the risk of false alarms caused by local disturbances, while ensuring the timeliness of early warning for high-confidence ramp events, and improving the scientific and economical nature of power grid dispatching decisions.

[0117] As a specific implementation of this application, based on the basic scheme, the candidate observation nodes are further limited to meteorological stations, wind measurement towers, and lidar wind measurement points; After the rapid changes in power of nearby wind farms coincided with the meteorological characteristics of cold fronts and non-meteorological anomalies were ruled out, they served as special observation nodes in the cold front propagation chain.

[0118] Specifically, this implementation clarifies the composition of candidate observation nodes for predicting the power ramp-up of cold fronts, introduces adjacent wind farms as special observation nodes, fully integrates existing meteorological observation resources and wind power operation data, and constructs a multi-dimensional cold front observation network, effectively improving the spatial coverage and signal sensing accuracy of the cold front propagation chain.

[0119] Conventional candidate observation nodes include meteorological stations, anemometer towers, and lidar anemometer points. Meteorological stations provide long-term, continuous, and wide-coverage multi-element meteorological observation data, covering core indicators for cold front identification such as temperature, air pressure, wind direction, wind speed, and precipitation, serving as the fundamental data source for large-scale cold front signal identification. Anemometer towers can acquire refined wind field data at the height of wind turbine hubs, including wind speed, wind direction, and turbulence intensity, more closely reflecting the actual meteorological environment of wind turbine operation and providing more accurate perception of wind resource changes caused by cold fronts. Lidar anemometer points enable long-distance, high spatiotemporal resolution wind field observations, supplementing areas not covered by conventional anemometer towers and capturing the refined wind field evolution characteristics of the cold front front.

[0120] Adjacent wind farms can serve as special observation nodes in the cold front propagation chain, requiring the simultaneous fulfillment of two core conditions. First, the rapid power changes at the wind farm must occur synchronously with the meteorological characteristics of the cold front; that is, the moment of the power surge must closely coincide with the occurrence of core meteorological features of the cold front, such as temperature drops, sudden increases in wind speed, and wind direction changes, without significant temporal misalignment. Second, all power anomalies caused by non-meteorological factors must be completely ruled out, including power-limited operation, unit failure shutdowns, planned maintenance, communication data anomalies, sudden changes in available capacity, and mandatory control by dispatch commands.

[0121] Nearby wind farms offer unique advantages as special observation nodes. Their power data directly reflects the actual response of the wind power system to cold front meteorological processes, providing a more accurate representation of the true impact of cold fronts on wind power compared to simple meteorological observations. For example, during a cold front moving from north to south, a nearby upstream wind farm experienced a significant power decrease of 50MW within 45 minutes at 13:02, accompanied by a 4.6℃ drop in temperature and a sudden 75° change in wind direction. After investigation, no non-meteorological anomalies were found. Including this in the propagation chain allows for a 17-minute advance prediction of the cold front's passage time at the target wind farm.

[0122] This implementation fully leverages the value of existing data resources through the collaborative application of multiple types of observation nodes. Introducing nearby wind farms as special observation nodes effectively supplements the shortcomings of conventional meteorological observations in wind power response, enhances the signal reliability and spatiotemporal continuity of the cold front propagation chain, and further improves the accuracy and timeliness of cold front-type wind power ramp-up predictions.

[0123] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0124] The above are merely specific embodiments of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for ultra-short-term prediction of wind power ramp-up in cold front scenarios, characterized in that, include: Acquire data on the target wind field, surrounding candidate observation nodes, and numerical weather prediction prior to the forecast release time, and perform data quality control and screening; Based on the coordinated changes of various meteorological elements at the candidate observation nodes, the candidate cold front signals for each candidate observation node are identified, and the occurrence time of the cold front signal is determined. Based on the data from the numerical weather forecast and the order of the occurrence times of the cold front signals at multiple candidate observation nodes, the current propagation direction of the cold front is determined. Based on the propagation direction of the cold front, effective upstream nodes located in the effective upstream direction of the target wind field are dynamically selected from the candidate observation nodes; The effective upstream nodes are sorted according to their relative positions with the target wind field in the propagation direction to construct a cold front meteorological observation propagation chain. The propagation chain is then checked for consistency, and nodes that do not meet the check requirements are removed or downgraded. Based on the occurrence time of the cold front signal of the effective upstream node and its propagation time delay to the target wind field, the passage time of the cold front of the target wind field is predicted. Using the time of the cold front passing through the target wind field as the time anchor point, the basic wind power prediction curve is modified at the event level related to the cold front-type climbing event to obtain the modified wind power prediction result.

2. The ultra-short-term prediction method for cold front-type wind power ramping according to claim 1, characterized in that, The identification of cold front candidate signals for each candidate observation node based on the coordinated changes of multiple meteorological elements at the candidate observation nodes includes: For each candidate observation node, calculate its temperature drop, wind speed increase, wind direction change angle, air pressure change, and gust increase, and standardize the above features. The standardized features are weighted and summed to construct the single-node cold front signal intensity. When the intensity of the cold front signal is not less than a preset threshold and the number of features that undergo coordinated changes exceeds a preset number, it is determined that the candidate observation node has a cold front candidate signal.

3. The ultra-short-term prediction method for cold front-type wind power ramping according to claim 1, characterized in that, Determining the time of occurrence of the cold front signal includes: The determination is based on a weighted fusion of one or more of the following: the initial trigger time of the cold front signal, the peak time of the cold front signal, and the trigger time of multi-feature fusion.

4. The ultra-short-term prediction method for cold front-type wind power ramping according to claim 1, characterized in that, The step of dynamically selecting effective upstream nodes located in the effective upstream direction of the target wind field from the candidate observation nodes according to the propagation direction of the cold front includes: Calculate the score of the effective upstream node. When the score Ui of the effective upstream node is not less than a preset threshold, the node is determined to be a valid upstream node. The formula for calculating the effective upstream node score is as follows: Ui = Ci·Qi·Ri·Hi·Bi·exp(-Di_perp / Lperp)·exp(-Di_proj / LD); Where Ci is the directional consistency factor, Qi is the data quality factor, Ri is the historical propagation reliability factor, Hi is the terrain channel consistency factor, Bi is the cold front signal reliability factor, Di_proj is the projection distance, Di_perp is the lateral offset distance, and Lperp and LD are the lateral and along-the-path distance attenuation scales, respectively.

5. The ultra-short-term prediction method for cold front-type wind power ramping according to claim 1, characterized in that, The process of performing consistency checks on the propagation chain, removing or downgrading nodes that do not meet the checks, includes: Verify whether the occurrence time of the cold front signal of adjacent nodes in the propagation chain satisfies the condition that the occurrence time of the signal of the far upstream node is not later than the occurrence time of the signal of the near upstream node; Calculate the local propagation speed between adjacent nodes in the propagation chain, and verify whether the local propagation speed is within the preset reasonable speed range; Only adjacent node pairs that satisfy both of the above conditions are considered to satisfy the propagation chain consistency.

6. The ultra-short-term prediction method for cold front-type wind power ramping according to claim 1, characterized in that, The propagation time delay of the target wind field is calculated as follows: τi = β1, i ·τi_real + β2, i ·τi_hist + β3, i ·τi_NWP; Where τi_real is the real-time observation propagation lag, τi_hist is the historical propagation lag of similar cold fronts, τi_NWP is the numerical weather prediction propagation lag, and β1, i β2, i β3, i The weights are dynamically fused, and the sum of the three is 1.

7. The ultra-short-term prediction method for cold front-type wind power ramping according to claim 1, characterized in that, The parameters of the cold front-type wind power ramp-up event include: Climb start time, climb end time, climb direction, climb amplitude, and maximum climb rate; The formula for predicting the slope amplitude is as follows: ΔP_ramp_hat = (1-μ)·ΔPbase + μ·ΔPcf, Wherein, ΔP_ramp_hat is the ramp amplitude, ΔPbase is the power change of the basic prediction curve within the window before and after the cold front, ΔPcf is the power change estimated based on the meteorological response of the cold front, and μ is the cold front correction weight.

8. The ultra-short-term prediction method for cold front-type wind power ramping according to claim 1, characterized in that, The event-level correction of the base wind power prediction curve related to cold front-type ramping events includes: Construct a cold front-type ramp correction window Wcf = [t_start_hat, t_end_hat], where t_start_hat and t_end_hat are the predicted ramp start and end times, respectively; Construct a normalized climbing shape function Kn( t ), and construct a cold front event power reference curve Pcf based on the ramp amplitude ΔP_ramp_hat and the power anchor point Panchor. t ) = Panchor + ΔP_ramp_hat · Kn( t ); Within the correction window Wcf, the base predicted power Ppred is adjusted according to the cold front correction intensity coefficient λcf. t ) and the cold front event power reference curve Pcf( t The fusion correction is performed to obtain the corrected output predicted power Pout. t ) = (1-λcf)·Ppred( t ) + λcf·Pcf( t ).

9. The ultra-short-term prediction method for cold front-type wind power ramping according to claim 1, characterized in that, Also includes: Based on the number of effective upstream nodes, node data quality, cold front signal strength, propagation chain consistency factor, cold front direction consistency, and propagation time delay uncertainty, the confidence level of cold front-type ramp prediction is calculated, and the intensity or method of ramp warning is adjusted accordingly.

10. The ultra-short-term prediction method for cold front-type wind power ramping according to any one of claims 1-9, characterized in that, The candidate observation nodes include meteorological stations, wind measurement towers, and lidar wind measurement points; After the rapid changes in power of nearby wind farms coincided with the meteorological characteristics of cold fronts and non-meteorological anomalies were ruled out, they served as special observation nodes in the cold front propagation chain.

Images