Cascade low-level jet probability prediction method for environmental wind field

Abstract

The application provides a cascaded low-level jet probability prediction method for an environmental wind field, and belongs to the field of atmospheric environment and weather forecast. The application adopts a cascaded prediction method to realize driving by sea level pressure field data, and realizes a fast, efficient, lightweight, and rolling update low-level jet occurrence and non-occurrence probability prediction scheme and a dominant wind direction probability prediction scheme. The application comprises the following steps: S1, low-level jet identification and dominant wind direction extraction; S2, screening of key spatial modes significantly related to the low-level jet; S3, construction of a multi-classification logistic regression model; S4, construction of a time series prediction model; S5, real-time data assimilation and principal component value prediction; S6, low-level jet probability prediction; and S7, real-time and rolling update of the prediction result.

Description

Technical Field

[0001] This application relates to the field of atmospheric environment and meteorological forecasting, and specifically proposes a cascaded low-level jet probability prediction method for environmental wind fields. Background Technology

[0002] The low-level jet stream is a strong and narrow airflow band in the lower troposphere. As a key environmental wind field, its activity directly affects aviation safety and wind power generation efficiency, and dominates the medium- and long-distance transport paths and local convergence and diffusion processes of air pollutants. How to achieve rapid and accurate prediction of whether the low-level jet stream will occur and the dominant wind direction of the jet stream is a key link in improving environmental meteorological service capabilities.

[0003] Operational forecasting of low-level jets in existing technologies mainly relies on numerical weather prediction models. These models, based on complex physical equations, can provide multi-dimensional meteorological field forecasts, but their computational processes are lengthy and resource-intensive. From data assimilation to model integration, it typically takes several hours, making it difficult to support the demand for high-temporal-resolution real-time rolling early warnings and resulting in significant delays in responding to sudden environmental risks. As described in the previously published patent application (CN201811512736.1), entitled "Method and Apparatus for Low-Level Jet Identification Based on Wind Profiler Radar," this method acquires wind profiler data; converts the absolute values ​​of wind speed and wind speed gradient at each altitude in the wind profiler data at any given time t into dimensionless values; calculates the weighted average of the converted absolute values ​​of wind speed and wind speed gradient at each altitude according to preset weights, and uses this weighted average as the low-level jet index at each altitude in the wind profiler data at time t; judges the low-level jet index at each altitude according to a preset threshold, and obtains the low-level jet parameters at time t based on the judgment result. While the above methods can achieve accurate identification of low-level jet stream profiles, they are essentially real-time or post-event diagnoses of wind fields that have already occurred, and lack the ability to predict future conditions.

[0004] From a dynamic perspective, low-level jet streams are primarily driven by pressure gradient forces, and their generation and dissipation exhibit a stable and identifiable physical correlation with specific large-scale sea-level pressure field configurations. This indicates that the sea-level pressure field contains potential information characterizing low-level jet stream activity. This provides a solid physical foundation for bypassing complex full-physics numerical simulations and directly utilizing sea-level pressure field data to drive modeling, thereby enabling rapid and accurate prediction of low-level jet streams. However, despite the recognition that sea level pressure field data contains crucial information for predicting low-level jets, existing technologies still suffer from significant technical gaps in achieving this data-driven approach: First, at the modality extraction level, it fails to systematically identify and extract the key spatial modes most statistically correlated with low-level jet events from the sea level pressure field, thus hindering the construction of efficient forecasting factors; second, at the probability mapping level, there is a lack of a quantitative model that can map the principal component sequences of the aforementioned modes to the probability of occurrence of different categories of low-level jets (occurrence or non-occurrence, dominant wind direction of the jet); third, at the temporal prediction level, it fails to construct an effective temporal prediction model based on the principal component sequences of the aforementioned modes, resulting in the inability to transform historical or real-time-based diagnoses into advanced forecasts. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a cascaded low-level jet probability prediction method for environmental wind fields. In view of the lack of systematic connection between "modal extraction - probability mapping - time series prediction" in the existing technology, a cascaded prediction method is adopted to realize a fast, efficient, lightweight and rolling updated probability prediction scheme for the occurrence of low-level jet and its dominant wind direction, driven by sea level pressure field data.

[0006] To achieve the above-mentioned objective, the cascaded low-level jet probability prediction method for environmental wind fields includes the following steps:

[0007] Step S1: Identification of low-level jet stream and extraction of prevailing wind direction;

[0008] Based on historical wind field data, low-level jet streams are identified and the prevailing wind direction is extracted. The jet streams are then categorized according to their presence or absence and the prevailing wind direction, generating a jet stream category label sequence.

[0009] Step S2: Screening of key spatial modes significantly associated with the low-level jet stream;

[0010] The historical surface pressure field of the target area is decomposed using empirical orthogonal functions to screen key spatial modes and their corresponding principal component sequences that are significantly related to the jet stream category label sequence.

[0011] Step S3: Construct a multi-class logistic regression model;

[0012] Based on the principal component sequence and the jet stream category label sequence, a multi-class logistic regression model is trained to establish a mapping relationship from principal component values ​​to the probability of occurrence of each jet stream category, thus completing the "modality-probability" mapping.

[0013] Step S4: Construct a time series prediction model;

[0014] A time-series prediction model is constructed based on principal component sequences, which is achieved by mining the time-series characteristics of the principal component sequences themselves;

[0015] The Long Short-Term Memory (LSTM) network was selected as the time-series prediction model. This model is trained using a 24-hour historical principal component sequence as input and the principal component sequence for the next 6 hours as the prediction target. After training, inputting the latest 24-hour principal component sequence will output the predicted principal component values ​​for the next 6 hours, as shown in the following expression:

[0016]

[0017] in, Indicates based on Model prediction Principal component values ​​in the next 6 hours after the given time; express Principal component values ​​for the preceding 24 consecutive hours;

[0018] Step S5: Real-time data assimilation and principal component value prediction;

[0019] Using a time-series prediction model, the principal component values ​​at multiple future time points are predicted. This step is cascaded with steps S3 and S4 above to form a chain of "principal component value prediction of key spatial modes - jet probability prediction", so that the prediction results are based on the constraints of physical mechanisms.

[0020] Step S6: Low-level jet probability prediction;

[0021] The predicted values ​​of the future principal components obtained in step S5 are converted into the probability of low-level jet streams at each future time.

[0022] Step S7: Real-time and rolling updates of prediction results;

[0023] Whenever new real-time surface pressure field data is obtained, steps S5 to S6 are repeated to achieve dynamic and rolling updates of the low-level jet probability prediction results.

[0024] Step S1 includes,

[0025] The criteria for identifying the low-level jet stream profile are: within an altitude of 700 hPa, there exists a maximum wind speed exceeding a set threshold, and the wind speed above it exhibits negative shear characteristics. The threshold is set based on historical wind field data and regional characteristics.

[0026] By statistically analyzing the identified historical low-level jet events, their dominant wind direction characteristics are extracted, and the low-level jets are classified according to whether the jets occur and the dominant wind direction, generating a low-level jet category label sequence.

[0027] Step S2 includes,

[0028] The historical surface pressure field of the target region is decomposed using empirical orthogonal functions to extract its main spatial modes and corresponding temporal principal component sequences, as shown in the following formula:

[0029]

[0030] In the formula, The historical sea level pressure field is obtained through empirical orthogonal function decomposition. A spatial mode, Indicates the first A spatial mode, Indicates the corresponding principal component sequence, superscript Represents the transpose matrix. Represents the residual.

[0031] Step S2 involves identifying key modes through statistical testing using the following process.

[0032] First, the correlation coefficients between the principal component sequences of each modality and the low-level jet stream category label sequences are calculated;

[0033] Secondly, based on the 5% significance level test, i.e. p<0.05, statistically significant correlation modalities were selected;

[0034] Finally, the mode with the largest absolute value of the correlation coefficient was selected as the key spatial mode.

[0035] Step S3 involves mapping the principal component values ​​of key spatial modes to the probability of low-level jet occurrence using a multi-class logistic regression model. The constructed multi-class logistic regression model is configured with... There are three mutually exclusive low-level jet stream state categories. The "no low-level jet stream" state is set as the reference category. For the remaining... The log-dominance function for a state with a "low-level jet stream" is:

[0036]

[0037] In the formula, The value of the principal component time series at the corresponding time point. , The first determined by fitting historical data Class model parameters, Indicates the first Predicted probability of low-level jet stream-like events This indicates the predicted probability that no low-level jet stream will occur.

[0038] Transforming the above formula, the predicted probabilities of each type of low-level jet stream are expressed as follows:

[0039] ,

[0040]

[0041] Step S5 includes,

[0042] Step S5.1: Real-time principal component value calculation;

[0043] The real-time surface pressure field is projected onto the key spatial modes to obtain the real-time principal component values;

[0044] In this embodiment, the key spatial mode is EOF2, and the real-time principal component values ​​are obtained by projection, using the following formula:

[0045]

[0046] In the formula, Indicates real-time The principal component values ​​at time t are derived from the standardized real-time sea level pressure field. With key spatial modes The inner product is calculated;

[0047] Step S5.2: Principal component sequence update;

[0048] Real-time principal component values Add the data to the end of the historical principal component sequence, while deleting the earliest value to form the latest 24-hour continuous sequence. as follows:

[0049]

[0050] Step S5.3: Output of the time series prediction model;

[0051] The updated sequence Input the time series forecast model constructed in step S4, and output the PC2 forecast value for the next 6 hours. as follows:

[0052]

[0053] In step S6, the predicted principal components output in step S5 are substituted one by one into the multi-class logistic regression model constructed in step S3 to calculate the probability of occurrence of each category of low-level jet stream at each future time.

[0054] This application proposes a computer device comprising one or more processors and a storage device, wherein the storage device stores one or more programs, and when the one or more programs are executed by the one or more processors, the above-described cascaded low-level jet probability prediction method for environmental wind fields is implemented.

[0055] This application proposes a computer-readable storage medium storing one or more executable programs, which, when executed by the processor of a computer device, enable the implementation of the aforementioned cascaded low-level jet probability prediction method for environmental wind fields.

[0056] In summary, the beneficial effects and advantages of this application compared with the prior art include:

[0057] 1. The prediction method proposed in this application can overcome the timeliness bottleneck of numerical models and make up for the lack of prediction function of monitoring technology, and construct a low-level jet stream prediction tool that is oriented towards environmental wind fields and has both real-time response capability and probabilistic forecasting.

[0058] 2. This application focuses on environmental wind fields and employs a real-time, efficient cascaded low-level jet probability prediction method. By abandoning traditional complex numerical model simulations and relying solely on sea-level pressure field data, it achieves probabilistic prediction of the occurrence of low-level jets and their dominant wind direction through a cascaded process of "modal extraction—probability mapping—time-series prediction." This application can provide accurate low-level jet forecast support for wind energy dispatching and provide highly timely environmental wind field prediction data for atmospheric pollution transmission path analysis and collaborative prevention and control decisions.

[0059] 3. This application reduces the high-dimensional prediction problem involving multiple meteorological elements to a single-variable time-series task based on surface pressure field, realizing lightweight, efficient, and real-time rolling prediction of the probability of low-level jet stream, which can provide decision-making reference for improving applications such as environmental wind field analysis, pollutant diffusion simulation, and wind energy dispatch. Attached Figure Description

[0060] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. Some specific embodiments of this application will be described in detail below with reference to the accompanying drawings in an exemplary and non-limiting manner. The same reference numerals in the drawings designate the same or similar parts or components. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale.

[0061] Figure 1 This is a flowchart of the cascaded low-level jet probability prediction method for environmental wind fields described in this application;

[0062] Figure 2 It is a low-level jet stream wind rose diagram;

[0063] Figure 3 It is a principal component sequence diagram;

[0064] Figure 4 This is a graph showing the relationship between the category label sequence and the principal component sequence of the low-level jet stream. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0066] To fully illustrate the cascaded low-level jet probability prediction method for environmental wind fields proposed in this application, this embodiment uses ERA5 reanalysis data and a specific region in China as an example. The overall process is as follows: Figure 1 As shown, the implementation steps include the following:

[0067] Step S1: Identification of low-level jet stream and extraction of prevailing wind direction;

[0068] Based on historical wind field data, low-level jet streams are identified and the prevailing wind direction is extracted. The jet streams are then categorized according to their presence or absence and the prevailing wind direction, generating a jet stream category label sequence.

[0069] The criteria for identifying the low-level jet stream profile are: within an altitude of 700 hPa, there exists a maximum wind speed exceeding a set threshold, and the wind speed above it exhibits negative shear characteristics. The threshold is set based on historical wind field data and regional characteristics.

[0070] By statistically analyzing the identified historical low-level jet events, we extract their dominant wind direction characteristics and classify the low-level jets according to whether the jets occur and their dominant wind direction, generating a low-level jet category label sequence.

[0071] For example, to identify a low-level jet stream that occurred in a certain region of China in 2024, the wind speed threshold was set to 8 m / s based on local historical wind field characteristics and related research.

[0072] Based on the identified low-level jet stream profile, the jet stream wind direction characteristics were statistically analyzed; to visually analyze the dominant jet stream wind direction, the statistical results were plotted as a wind rose diagram; Figure 2 It is known that the prevailing wind direction of the jet stream in this region in 2024 was southwest and northeast. Therefore, the wind direction range SSE-NNW is classified as a southwest jet stream, and the remaining wind direction range is classified as a northeast jet stream.

[0073] To facilitate subsequent modeling, each time point is labeled with a category, generating a low-level jet stream category label sequence: no jet stream is labeled as 0, southwest jet stream is labeled as -1, and northeast jet stream is labeled as 1;

[0074] Step S2: Screening of key spatial modes significantly associated with the low-level jet stream;

[0075] The historical surface pressure field of the target area is decomposed using empirical orthogonal functions to screen key spatial modes and their corresponding principal component sequences that are significantly related to the jet stream category label sequence.

[0076] Surface pressure fields are important indicators of atmospheric circulation patterns. To extract the signals most relevant to the occurrence of low-level jets from massive amounts of information, it is necessary to perform empirical orthogonal function decomposition on the historical surface pressure fields of the target region (covering mesoscale to large-scale ranges) to extract its main spatial modes and corresponding temporal principal component sequences, as shown in the following formula:

[0077]

[0078] In the formula, The historical sea level pressure field is obtained through empirical orthogonal function decomposition. A spatial mode, Indicates the first A spatial mode, Indicates the corresponding principal component sequence, superscript Represents the transpose matrix. Represents the residual;

[0079] Given that not all spatial modes in the surface pressure field are directly related to the low-level jet dynamics, key modes need to be identified through statistical tests; including,

[0080] First, the correlation coefficients between the principal component sequences of each modality and the low-level jet stream category label sequences are calculated;

[0081] Secondly, statistically significant modalities were selected based on the 5% significance level test (p<0.05);

[0082] Finally, the mode with the largest absolute value of the correlation coefficient was selected as the key spatial mode;

[0083] The above screening process can eliminate random correlation interference at a 95% confidence level, extract the key spatial modes that can truly reflect the circulation pattern driving the low-level jet, and ensure its physical interpretability and statistical reliability.

[0084] In this embodiment, ERA5 sea level pressure field data is used as the surface pressure field data, covering the region and surrounding mesoscale to large-scale study areas. The 2024 sea level pressure field data is standardized and then subjected to empirical orthogonal function decomposition to obtain multiple spatial modes and their corresponding principal component time series. The correlation coefficients between the principal component sequences (PC1, PC2, PC3) of the first three modes (EOF1, EOF2, EOF3) and the low-level jet category label sequence are 0.3, 0.6, and 0.1, respectively, and all pass the 95% significance test. Therefore, in this embodiment, the second mode, EOF2, is identified as the key spatial mode favorable to low-level jets, and its corresponding principal component sequence is PC2. Figure 3 As shown; thereafter, this PC2 sequence is used as the core input for subsequent modeling and is denoted as "the principal component time series";

[0085] Step S3: Construct a multi-class logistic regression model;

[0086] Based on the principal component sequence and the jet stream category label sequence, a multi-class logistic regression model is trained to establish a mapping relationship from principal component values ​​to the probability of occurrence of each jet stream category, thus completing the "modality-probability" mapping.

[0087] A highly significant statistical association exists between the principal component sequences of the key spatial modes and the low-level jet category label sequences, indicating that the principal component sequences are strong predictors distinguishing different low-level jet categories. Multi-class logistic regression models are a classic statistical method for deriving the probability of low-level jet occurrence from the principal component values ​​of the key spatial modes.

[0088] By mapping the principal component values ​​of key spatial modes to the probability of low-level jet occurrence using a multi-class logistic regression model, the temporal evolution of these modes is characterized by the principal component sequences. Establishing a probabilistic mapping between the principal component sequences and the low-level jet label sequences, the multi-class logistic regression model is a classic statistical method for achieving this mapping.

[0089] The training method for the multi-class logistic regression model is as follows: Multi-class logistic regression is performed based on the principal component sequences of the key spatial modes and the low-level jet category label sequences to establish a mapping relationship from principal component values ​​to the occurrence probability of low-level jet categories. In application, the principal component values ​​at any given time are input into the multi-class logistic regression model to output the occurrence probability of each low-level jet category at that time. Specifically,

[0090] Using the key spatial modal principal component sequences selected in step S2 as input, a multi-class logistic regression model is constructed.

[0091] Multi-class logistic regression model: (The model is described below) There are three mutually exclusive low-level jet stream state categories, with the "no low-level jet stream" state set as the reference category (denoted as category 0); for the remaining categories... Similar to the state of "having a low-level jet stream" (i.e., different types of jet streams with different prevailing wind directions), Its log-dominance function relative to the reference class is:

[0092]

[0093] In the formula, The value of the principal component time series at the corresponding time (i.e., the principal component value). , The first determined by fitting historical data Class model parameters, Indicates the first Predicted probability of low-level jet stream-like events This indicates the predicted probability that no low-level jet stream will occur.

[0094] Transforming the above formula, the predicted probabilities of each type of low-level jet stream are expressed as follows:

[0095] ,

[0096]

[0097] The multi-class logistic regression model constructed above has clear meteorological significance: the phase changes of the principal component values ​​of the key spatial modes, which serve as inputs to the model, reflect the spatial modes that are conducive to the development of different types of low-level jets; the regression coefficients obtained by fitting the multi-class logistic regression model quantify the direction and intensity of this correspondence through their signs and magnitudes, thereby statistically verifying the physical mechanism of "key spatial modes driving low-level jets in specific wind directions".

[0098] To intuitively analyze the relationship between the low-level jet stream and key spatial modes and to establish a probabilistic mapping model, the following procedure was performed:

[0099] First, the jet stream direction and the principal component sequence (PC2) are plotted on a polar coordinate graph. Figure 4 );

[0100] Subsequently, based on Figure 4 Based on the statistical relationships revealed in the data, a multi-class logistic regression model was established with "no rapid flow" as the reference category.

[0101] In this embodiment, The specific category is: no low-level jet stream (the first category is missing from the original text). (type), southwest low-level jet stream ( ), northeasterly low-level jet stream ( By training a multi-class logistic regression model using the low-level jet stream category label sequence and the principal component sequence PC2 for the region in 2024, the model parameters are as follows:

[0102] The regression coefficient of the southwest low-level jet relative to the absence of a low-level jet is: , The regression coefficient of the northeasterly low-level jet stream relative to the absence of a low-level jet stream is: , .

[0103] Substituting the model parameters into the above probability model, we obtain the prediction formula for calculating the probability of each category of low-level jet stream from the PC2 value:

[0104]

[0105]

[0106]

[0107] In the formula, , , These correspond to the probabilities of no jet stream, southwest jet stream, and northeast jet stream, respectively. The values ​​represent the PC2 principal component values. When the PC2 phase of the principal component sequence is negative, the spatial mode favors the development of a southwest low-level jet; when it is positive, it favors the development of a northeast jet; a weak PC2 amplitude is unfavorable for the occurrence of a low-level jet, which is consistent with... Figure 4 The revealed statistical regularities and the physical mechanisms by which pressure field configurations drive jet streams in different wind directions are consistent.

[0108] Step S4: Construct a time series prediction model;

[0109] The time series prediction model is constructed based on the principal component sequence. Since step S3 realizes the mapping of "given the principal component value, calculate the current jet probability", this step is to predict the principal component value at future time. That is, it is achieved by mining the time series characteristics of the principal component sequence itself. The time series prediction model has the ability to capture the long-term and short-term dependencies in the time series.

[0110] The temporal evolution of key spatial modes driving the low-level jet stream is characterized by principal component sequences, which inherently contain untapped temporal evolutionary patterns. Based on the periodicity, trend, and other temporal characteristics of historical principal component sequences, future sequences can be predicted, and the predicted principal component values ​​will reflect the future evolutionary characteristics of the key spatial modes. Therefore, models capable of capturing long-term or short-term dependencies in time series are needed; this is the core model for transforming the historical evolution of key spatial modes into advanced predictions of future states.

[0111] In this embodiment, a Long Short-Term Memory Network (LSTM) is selected as the time series prediction model. Its unique gating mechanism can effectively capture long-term dependencies in time series, making it suitable for modeling and predicting the time series features of the principal component time series PC2. The model is trained using the historical principal component series PC2 of the past 24 hours as input and the PC2 series of the next 6 hours as the prediction target. After training, the latest 24-hour PC2 series is input to output the predicted PC2 values ​​for the next 6 hours, as shown in the following expression:

[0112]

[0113] in, Indicates based on Model prediction Principal component values ​​in the next 6 hours after the given time; express Principal component values ​​for the preceding 24 consecutive hours;

[0114] Based on this step, firstly, the prediction problem of high-dimensional meteorological fields is transformed into time series prediction of a single key index "principal component value", which greatly simplifies the complexity of the low-level jet forecasting task; secondly, the model framework is flexible and can be adapted to various time series models to meet the computation and data requirements of different scenarios.

[0115] Step S5: Real-time data assimilation and principal component value prediction;

[0116] Using a time-series prediction model, the principal component values ​​at multiple future time points are predicted. This step is cascaded with steps S3 and S4 above to form a chain of "principal component value prediction of key spatial modes - jet probability prediction", so that the prediction results are based on the constraints of physical mechanisms.

[0117] The core of this step is to complete the calculation process of "real-time surface pressure field observation - real-time principal component value calculation - principal component sequence extrapolation", which assimilates the latest real-time atmospheric state into the prediction system and generates the prediction sequence of key spatial mode principal components.

[0118] Specifically, it includes,

[0119] Step S5.1: Real-time principal component value calculation;

[0120] The real-time surface pressure field is projected onto the key spatial modes to obtain the real-time principal component values;

[0121] In this embodiment, the key spatial mode is EOF2, and the real-time principal component values ​​are obtained by projection, using the following formula:

[0122]

[0123] In the formula, Indicates real-time The principal component values ​​at time t are derived from the standardized real-time sea level pressure field. With key spatial modes The inner product is calculated; for example, 06:00 on March 2, 2025 as the real-time... At time 1, the real-time principal component values ​​are calculated. ;

[0124] Step S5.2: Principal component sequence update;

[0125] Real-time principal component values Add the data to the end of the historical principal component sequence, while deleting the earliest value to form the latest 24-hour continuous sequence. as follows:

[0126]

[0127] In this embodiment, the updated principal component sequence covering the period from 07:00 on March 1st to 06:00 on March 2nd, a total of 24 hours, is as follows: ;

[0128] Step S5.3: Output of the time series prediction model;

[0129] The updated sequence Input the time series forecast model constructed in step S4, and output the PC2 forecast value for the next 6 hours. as follows:

[0130]

[0131] In this embodiment, the predicted principal component values ​​for the next 6 hours (from 07:00 to 12:00 on March 2nd) This completes the process from assimilating real-time observation data to predicting principal component values ​​at future moments.

[0132] Step S6: Low-level jet probability prediction;

[0133] The predicted values ​​of the future principal components obtained in step S5 are converted into the probability of low-level jet streams at each future time.

[0134] Specifically, the predicted PC2 values ​​for each future time point output in step S5 are substituted into the multi-class logistic regression model constructed in step S3 to calculate the probability of occurrence of each category of low-level jet stream at each future time point.

[0135] In this embodiment, the principal component prediction values ​​for the next 6 hours output in step S5 are... Substituting the values ​​into the multi-class logistic regression probability model trained in step S3, the probability forecast for the low-level jet stream over the next 6 hours is calculated. Specifically, the principal component prediction value for the 6th hour, i.e., 12:00 on March 2nd, is 54.9. Substituting this value into the formula in step S3 yields:

[0136] , ,

[0137] The probability of a northeasterly low-level jet stream occurring in the region at that time is the highest, at approximately 95%, followed by a probability of no low-level jet stream at approximately 5%, and a probability of a southwesterly low-level jet stream at approximately 0%.

[0138] Step S7: Real-time and rolling updates of prediction results;

[0139] Whenever new real-time surface pressure field data is obtained, steps S5 to S6 are repeated to achieve dynamic and rolling updates of the low-level jet probability prediction results.

[0140] Example of prediction performance in this embodiment:

[0141] by Using the time (06:00 on March 2, 2025) as the real-time benchmark, Projecting the real-time sea level pressure field at time t to the key space mode ( ), to obtain the principal component values ​​at that moment. Based on this, a continuous 24-hour sequence containing the latest observations is obtained. The sequence is input into the time series prediction model ( Output the principal component prediction sequence for the next 6 hours (07:00 to 12:00). Furthermore, based on a multi-class logistic regression model, the probability forecast of the low-level jet stream at the corresponding time is calculated, with the probability of it occurring in the 6th hour (12:00) being:

[0142] , ,

[0143] This indicates that the probability of a northeasterly low-level jet stream occurring in the 6th hour is the highest (95%), while the probabilities of no low-level jet stream and southwesterly low-level jet stream are 5% and 0%, respectively. Actual observation confirmed that a northeasterly low-level jet stream occurred at this time, with a jet stream wind speed of 24 m / s and a jet stream wind direction of 43°. This example verifies the effective predictive ability of the method of this invention regarding the occurrence of low-level jet streams and the prevailing wind direction.

[0144] This application proposes a novel computer device comprising one or more processors and a storage device, wherein the storage device stores one or more programs, and when the one or more programs are executed by the one or more processors, the aforementioned cascaded low-level jet probability prediction method for environmental wind fields is implemented.

[0145] This application proposes a novel computer-readable storage medium storing one or more executable programs, which, when executed by the processor of a computer device, enable the implementation of the aforementioned cascaded low-level jet probability prediction method for environmental wind fields.

[0146] In summary, this application achieves lightweight, rapid, and efficient low-level jet probability prediction based solely on surface pressure field data through a cascaded "modal extraction—probability mapping—time series prediction" method. The prediction results can directly serve applications such as environmental wind field analysis and atmospheric pollutant diffusion simulation, providing probability-based low-level jet state information support for related fields.

[0147] It should be understood that the above embodiments are only for illustrating the technical solutions of the present invention. Those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and all such improvements and modifications should be covered within the protection scope of the present invention.

Claims

1. A cascaded low-level jet probability prediction method for environmental wind fields, characterized in that: Including, Step S1: Identification of low-level jet stream and extraction of prevailing wind direction; Based on historical wind field data, low-level jet streams are identified and the prevailing wind direction is extracted. The jet streams are then categorized according to their presence or absence and the prevailing wind direction, generating a jet stream category label sequence. Step S2: Screening of key spatial modes significantly associated with the low-level jet stream; The historical surface pressure field of the target area is decomposed using empirical orthogonal functions to screen key spatial modes and their corresponding principal component sequences that are significantly related to the jet stream category label sequence. Step S3: Construct a multi-class logistic regression model; Based on the principal component sequence and the jet stream category label sequence, a multi-class logistic regression model is trained to establish a mapping relationship from principal component values ​​to the probability of occurrence of each jet stream category, thus completing the "modality-probability" mapping. Step S4: Construct a time series prediction model; A time-series prediction model is constructed based on principal component sequences, which is achieved by mining the time-series characteristics of the principal component sequences themselves; The Long Short-Term Memory (LSTM) network was selected as the time-series prediction model. This model is trained using a 24-hour historical principal component sequence as input and the principal component sequence for the next 6 hours as the prediction target. After training, inputting the latest 24-hour principal component sequence will output the predicted principal component values ​​for the next 6 hours, as shown in the following expression: ； in, Indicates based on Model prediction Principal component values ​​in the next 6 hours after the given time; express Principal component values ​​for the preceding 24 consecutive hours; Step S5: Real-time data assimilation and principal component value prediction; Using a time-series prediction model, the principal component values ​​at multiple future time points are predicted. This step is cascaded with steps S3 and S4 above to form a chain of "principal component value prediction of key spatial modes - jet probability prediction", so that the prediction results are based on the constraints of physical mechanisms. Step S6: Low-level jet probability prediction; The predicted values ​​of the future principal components obtained in step S5 are converted into the probability of low-level jet streams at various future times. Step S7: Real-time and rolling updates of prediction results; Whenever new real-time surface pressure field data is obtained, steps S5 to S6 are repeated to achieve dynamic and rolling updates of the low-level jet probability prediction results.

2. The cascaded low-level jet probability prediction method for environmental wind fields according to claim 1, characterized in that: Step S1 includes, The criteria for identifying the low-level jet stream profile are: within an altitude of 700 hPa, there exists a maximum wind speed exceeding a set threshold, and the wind speed above it exhibits negative shear characteristics. The threshold is set based on historical wind field data and regional characteristics. By statistically analyzing the identified historical low-level jet events, their dominant wind direction characteristics are extracted, and the low-level jets are classified according to whether the jets occur and the dominant wind direction, generating a low-level jet category label sequence.

3. The cascaded low-level jet probability prediction method for environmental wind fields according to claim 1, characterized in that: Step S2 includes, The historical surface pressure field of the target region is decomposed using empirical orthogonal functions to extract its main spatial modes and corresponding temporal principal component sequences, as shown in the following formula: ； In the formula, The historical sea level pressure field is obtained through empirical orthogonal function decomposition. A spatial mode, Indicates the first A spatial mode, Indicates the corresponding principal component sequence, superscript Represents the transpose matrix. Represents the residual.

4. The cascaded low-level jet probability prediction method for environmental wind fields according to claim 3, characterized in that: Step S2 involves identifying key modes through statistical testing using the following process. First, the correlation coefficients between the principal component sequences of each modality and the low-level jet stream category label sequences are calculated; Secondly, based on the 5% significance level test, i.e. p<0.05, statistically significant correlation modalities were selected; Finally, the mode with the largest absolute value of the correlation coefficient was selected as the key spatial mode.

5. The cascaded low-level jet probability prediction method for environmental wind fields according to claim 1, characterized in that: Step S3 involves mapping the principal component values ​​of key spatial modes to the probability of low-level jet occurrence using a multi-class logistic regression model. The constructed multi-class logistic regression model is configured with... There are three mutually exclusive low-level jet stream state categories. The "no low-level jet stream" state is set as the reference category. For the remaining... The log-dominance function for a state with a "low-level jet stream" is: ； In the formula, The values ​​of the principal component time series at the corresponding time points. , The first determined by fitting historical data Class model parameters, Indicates the first Predicted probability of low-level jet stream-like events This indicates the predicted probability that no low-level jet stream will occur. Transforming the above formula, the predicted probabilities of each type of low-level jet stream are expressed as follows: ， ； 。 6. The cascaded low-level jet probability prediction method for environmental wind fields according to claim 1, characterized in that: Step S5 includes, Step S5.1: Real-time principal component value calculation; The real-time surface pressure field is projected onto the key spatial modes to obtain the real-time principal component values; The key spatial mode is EOF2. The real-time principal component values ​​are obtained by projection, and the formula is as follows: ； In the formula, Indicates real-time The principal component values ​​at time t are derived from the standardized real-time sea level pressure field. With key spatial modes The inner product is calculated; Step S5.2: Principal component sequence update; Real-time principal component values Add the data to the end of the historical principal component sequence, while deleting the earliest value to form the latest 24-hour continuous sequence. as follows: ； Step S5.3: Output of the time series prediction model; The updated sequence Input the time series forecast model constructed in step S4, and output the PC2 forecast value for the next 6 hours. as follows: 。 7. The cascaded low-level jet probability prediction method for environmental wind fields according to claim 1, characterized in that: In step S6, the predicted PC2 values ​​for each future time point output in step S5 are substituted one by one into the multi-class logistic regression model constructed in step S3 to calculate the probability of occurrence of each category of low-level jet stream at each future time point.

8. A computer device, characterized in that: The method includes one or more processors and a storage device, wherein the storage device stores one or more programs, and when the one or more programs are executed by the one or more processors, the cascaded low-level jet probability prediction method for environmental wind fields as described in any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium, characterized in that: It stores one or more executable programs, and when the programs are executed by the processor of a computer device, the cascaded low-level jet probability prediction method for environmental wind fields described in any one of claims 1 to 7 is implemented.

Images