An extreme low temperature seasonal prediction method and system based on time scale separation and a medium

Abstract

The application discloses a kind of based on time scale separation's extremely low temperature season prediction method, system and medium, it is related to temperature prediction technical field, the method includes: obtaining the daily minimum temperature observation data of each grid in target area, using relative threshold method obtains the time series of the average extreme low temperature day number of target area, and respectively with multiple groups amplitude controlled Gaussian white noise superposition, constructs noise disturbance sample set;Noise disturbance sample set is carried out ensemble empirical mode decomposition and maximum entropy spectrum analysis, and using time lag correlation analysis obtains the prediction factor that each order stable intrinsic mode function has physical meaning;Based on random forest algorithm, the prediction model of each order stable intrinsic mode function is constructed, based on the trained prediction model, each order stable intrinsic mode function is predicted, and based on the prediction result, the predicted value of the average extreme low temperature day number of target area is obtained by reconstruction.The application can significantly improve the stability and accuracy of extreme low temperature season prediction.

Description

Technical Field

[0001] This invention relates to the field of temperature prediction technology, and in particular to a method, system and medium for predicting extreme low temperatures during the season based on time scale separation. Background Technology

[0002] Spring, as a crucial period of transition from winter to summer, is characterized by large temperature fluctuations and frequent extreme low-temperature events. Extreme low temperatures in spring can easily cause frost damage to crops; therefore, accurate seasonal-scale forecasting of the number of days with low temperatures in spring is of significant practical importance.

[0003] Existing methods for seasonal temperature forecasting are mostly based on overall time series modeling, typically using historical temperature sequences or climatic factors to construct statistical or dynamic forecasting models. However, spring low-temperature days sequences often exhibit significant nonlinearity, non-stationarity, and multi-timescale superposition characteristics, with marked differences in the causes and dominant mechanisms of variation across different time scales. Using a single model to predict the overall sequence makes it difficult to simultaneously characterize high-frequency fluctuations, low-frequency changes, and long-term trends. The forecast results are also susceptible to noise interference, resulting in insufficient stability and reliability.

[0004] Furthermore, traditional empirical mode decomposition (EMD) methods are highly sensitive to the distribution of extreme points when processing climate observation data containing significant noise, easily leading to mode aliasing. This results in the intrinsic mode functions obtained from the decomposition lacking clear physical meaning, thus affecting the selection of predictor factors and their physical interpretation capabilities in subsequent prediction models. This problem is particularly pronounced for climate phenomena such as spring low temperatures, which are influenced by multiple factors including air-sea interactions and large-scale circulation adjustments. On the other hand, existing prediction methods often focus on statistical correlation analysis during the predictor factor selection process, lacking specific processing of the changing characteristics at different time scales. This makes it difficult to establish the correspondence between predictor factors and the time-scale components of the number of days with low temperatures, thereby limiting the improvement of seasonal prediction accuracy. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, system and medium for predicting extreme low temperature seasons based on time scale separation. By introducing ensemble empirical mode decomposition and random forest algorithm, the climate evolution law at different time scales is fully explored, and the accuracy of low temperature day prediction is significantly improved.

[0006] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0007] In a first aspect, the present invention provides a method for predicting extreme low-temperature seasons based on time-scale separation, comprising:

[0008] Acquire daily minimum temperature observation data for each grid point within the target area, and obtain the time series of the average number of extreme low temperature days in the target area using the relative threshold method based on the daily minimum temperature observation data;

[0009] The time series of the average number of extreme low temperature days in the target area is superimposed with multiple sets of amplitude-controlled Gaussian white noise to construct a noise perturbation sample set.

[0010] The noise disturbance sample set was subjected to ensemble empirical mode decomposition and maximum entropy spectrum analysis to obtain the stable intrinsic mode functions of each order and their dominant mode time scales;

[0011] Based on the stable intrinsic mode functions of each order and their dominant mode time scales, time delay correlation analysis is used to obtain the physically meaningful predictors of the stable intrinsic mode functions of each order.

[0012] A prediction model is constructed based on the random forest algorithm, and a training set is built according to the stable intrinsic mode functions of each order and the prediction factors to train the prediction model.

[0013] Based on the trained prediction model, stable intrinsic mode functions of each order are predicted, and based on the prediction results, ensemble empirical mode reconstruction is performed to obtain the predicted value of the average number of extreme low temperature days in the target area.

[0014] Optionally, the daily minimum temperature observation data of each grid point in the target area are the daily minimum temperature statistics of each day in a preset season of each natural year;

[0015] The time series of average extreme low temperature days in the target area obtained by using the relative threshold method based on the daily minimum temperature observation data includes:

[0016] Obtain the daily minimum temperature of each grid point within the target area during the historical reference period;

[0017] For each grid point, the daily minimum temperature of each day in the historical reference period is sorted from low to high, and the daily minimum temperature corresponding to the preset percentile of the sort is taken as the relative threshold of extreme low temperature of the grid point.

[0018] Based on the relative threshold of extreme low temperature and the daily minimum temperature observation data of each grid point, determine the number of extreme low temperature days for each grid point in the preset season of each natural year;

[0019] The average number of extreme low-temperature days for each grid point in each preset season of each calendar year is taken to obtain the average number of extreme low-temperature days in the target area in each preset season of each calendar year, and a time series is formed in calendar year units.

[0020] Optionally, the ensemble empirical mode decomposition and maximum entropy spectrum analysis of the noise perturbation sample set includes:

[0021] Perform ensemble empirical mode decomposition on the noise perturbation sample set:

[0022]

[0023] In the formula, For the k-th noise disturbance sample in the noise disturbance sample set, the first... Sample values ​​for each calendar year For the first eigenmode functions of order 1 For residual terms;

[0024] The mean value of the eigenmode functions of the same order is taken as the stable eigenmode functions of the corresponding order:

[0025]

[0026] In the formula, For the first Stable intrinsic mode functions, This represents the number of noise disturbance samples in the noise disturbance sample set.

[0027] Maximum entropy spectrum analysis was performed on the stable intrinsic mode functions of each order to obtain the maximum entropy power spectrum. , For the first Maximum entropy power spectrum of first-order stable intrinsic mode functions;

[0028] For the maximum entropy power spectrum Find the frequency corresponding to its maximum point. According to frequency Solving for the dominant mode time scale , For the first The dominant mode timescale of the first-order stable intrinsic mode function.

[0029] Optionally, the method of obtaining physically meaningful predictors for each order of stable intrinsic mode functions using time-delay correlation analysis includes:

[0030] Obtain the sea surface temperature anomaly field (SSTA) of the target region for each calendar year;

[0031] The sea surface temperature anomaly field (SSTA) is time-scale matched based on the dominant mode timescale of each stable intrinsic mode function, and the time-delay correlation coefficient of each grid point in the target region is calculated. :

[0032]

[0033] In the formula, For the first Stable intrinsic mode functions, The sea surface temperature anomaly field SSTA for the x-th row and y-th column grid point within the target area after time matching processing. The time lag is in months. For the calendar year, In the first The time lag in the preset seasons of a natural year ; Let covariance function be used. for standard deviation for Standard deviation;

[0034] The Student's t significance test was performed on the time lag correlation coefficients under different time lags to obtain their confidence levels, and the sea surface temperature anomaly areas were determined based on the confidence levels.

[0035] For the sea surface temperature anomaly regions corresponding to key months for each order of stable intrinsic mode functions, calculate the potential predictors for each order of stable intrinsic mode functions. :

[0036]

[0037] In the formula, This is a region with anomalies in sea surface temperature. For standardized processing, Latitude area weighting, , Let x be the latitude of the grid point in the x-th row and y-th column;

[0038] For potential predictors Perform physical consistency discrimination to screen predictors with clear physical significance.

[0039] Optionally, the time-scale matching process for the sea surface temperature anomaly field (SSTA) based on the dominant mode timescale of each stable intrinsic mode function includes:

[0040] If the first Dominant mode timescale of first-order stable intrinsic mode functions Then the first The first-order stable intrinsic mode function is defined as the component on the decadal and higher time scales;

[0041] like Then the first The first-order stable intrinsic mode function is defined as the interannual time scale component;

[0042] For the stable intrinsic mode functions belonging to the interdecadal and higher time scale components, the sea surface temperature anomaly field SSTA is subjected to low-pass filtering.

[0043] For the stable intrinsic mode functions belonging to the interannual timescale component, the sea surface temperature anomaly field SSTA is subjected to linear detrending processing.

[0044] Optionally, training the prediction model includes:

[0045] The predictor is used as the input to the prediction model, and the corresponding stable intrinsic mode function is used as the output label of the prediction model. The mean absolute error is used as the loss function. The prediction model is trained and the parameters are optimized by five-fold cross-validation and grid search to determine the optimal parameter combination.

[0046] Secondly, the present invention provides an extreme low temperature seasonal prediction system based on time scale separation, comprising:

[0047] The data sampling module is configured to acquire the daily minimum temperature observation data of each grid point in the target area, and to obtain the time series of the average extreme low temperature days in the target area using the relative threshold method based on the daily minimum temperature observation data.

[0048] The noise perturbation module is configured to superimpose the time series of the average extreme low temperature days in the target area with multiple sets of amplitude-controlled Gaussian white noise to construct a noise perturbation sample set;

[0049] The mode decomposition module is configured to perform ensemble empirical mode decomposition and maximum entropy spectrum analysis on the noise disturbance sample set to obtain the stable intrinsic mode functions of each order and their dominant mode time scales.

[0050] The predictor determination module is configured to obtain physically meaningful predictors for each stable intrinsic mode function based on the time scale of each stable intrinsic mode function and its dominant mode function, using time delay correlation analysis.

[0051] The model building and training module is configured to build a prediction model based on the random forest algorithm, construct a training set according to the stable intrinsic mode functions of each order and the prediction factors, and train the prediction model.

[0052] The predictive mode reconstruction module is configured to predict stable intrinsic mode functions of each order based on the trained prediction model, and to perform ensemble empirical mode reconstruction based on the prediction results to obtain the predicted value of the average number of extreme low temperature days in the target area.

[0053] Thirdly, the present invention provides an electronic device, including a processor and a storage medium;

[0054] The storage medium is used to store instructions;

[0055] The processor is configured to operate according to the instructions to perform the steps according to the method described above.

[0056] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0057] Fifthly, the present invention provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0058] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0059] This invention provides a method, system, and medium for predicting extreme low temperatures seasonally based on time scale separation. 1) By introducing the ensemble empirical mode decomposition method, the time series of low temperature days is adaptively decomposed into intrinsic mode functions of different time scales, which effectively suppresses the mode aliasing phenomenon that is prone to occur in traditional empirical mode decomposition and improves the stability and physical interpretability of the decomposition results.

[0060] 2) For intrinsic mode functions at different time scales, predictive factors with clear physical meanings are selected respectively, and the correspondence between predictive factors and multi-scale changes in the number of low-temperature days is established, avoiding the problem that a single predictive model is difficult to characterize complex climate change features.

[0061] 3) Using the random forest method to model and predict each intrinsic mode function separately can effectively characterize the nonlinear relationship between the predictor and the number of low-temperature days, and improve the generalization ability and prediction accuracy of the prediction model.

[0062] 4) By linearly reconstructing the prediction results of each intrinsic mode function, the final prediction result of the number of low-temperature days is obtained, which fully integrates information from different time scales and significantly improves the stability and accuracy of seasonal prediction of the number of low-temperature days.

[0063] 5) It has good versatility and scalability, and can be applied to the seasonal prediction of the number of low-temperature days in spring in the target area, as well as extended to the prediction of other seasons and other temperature-related extreme events. Attached Figure Description

[0064] Figure 1 This is a flowchart of the extreme low temperature seasonal prediction method based on time scale separation provided in the embodiments of the present invention. Detailed Implementation

[0065] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0066] Example 1

[0067] like Figure 1 As shown, this embodiment of the invention provides a method for predicting extreme low temperatures seasonally based on time scale separation, including the following steps:

[0068] Step S1: Obtain the daily minimum temperature observation data of each grid point in the target area, and use the relative threshold method to obtain the time series of the average extreme low temperature days in the target area based on the daily minimum temperature observation data.

[0069] The daily minimum temperature observation data of each grid point in the target area are the daily minimum temperature statistics of each natural year in a preset season; the preset season is spring (i.e., March, April and May) mentioned in the background technology of this invention, and the preset season or any time period can be freely adjusted as needed.

[0070] The time series of average extreme low temperature days in the target area, obtained using the relative threshold method based on daily minimum temperature observation data, includes:

[0071] 1.1 Obtain the daily minimum temperature of each grid point within the target area during the historical reference period;

[0072] 1.2 For each grid point, sort the daily minimum temperature of each day in the historical baseline period from low to high, and take the daily minimum temperature corresponding to the preset percentile (such as 90%) of the sort as the extreme low temperature relative threshold of the grid point.

[0073] 1.3. Based on the relative threshold of extreme low temperature and the daily minimum temperature observation data of each grid point, determine the number of extreme low temperature days for each grid point in each preset season of each natural year;

[0074] 1.4. Take the average number of extreme low temperature days for each grid point in each preset season of each calendar year to obtain the average number of extreme low temperature days in the target area in each preset season of each calendar year, and form a time series in calendar year units.

[0075] In this invention, a relative threshold method is used to define the number of low-temperature days, which effectively eliminates the influence of differences in climate background in different regions and makes the determination of low-temperature events more regionally adaptable. At the same time, regional averaging reduces local noise interference, highlights large-scale climate signals, and provides a stable and unified prediction object for subsequent time-scale decomposition and prediction modeling.

[0076] Step S2: The time series of the average extreme low temperature days in the target area is superimposed with multiple sets of amplitude-controlled Gaussian white noise to construct a noise perturbation sample set.

[0077] Specifically, in this embodiment, at least 100 sets of amplitude-controlled Gaussian white noise are constructed and superimposed with the time series of the average extreme low temperature days in the target area to obtain at least 100 noise disturbance samples.

[0078] Step S3: Perform ensemble empirical mode decomposition and maximum entropy spectrum analysis on the noise disturbance sample set to obtain the stable intrinsic mode functions of each order and their dominant mode time scales.

[0079] Ensemble empirical mode decomposition and maximum entropy spectrum analysis of the noise perturbation sample set include:

[0080] 3.1 Perform ensemble empirical mode decomposition on the noise-perturbed sample set:

[0081]

[0082] In the formula, For the k-th noise disturbance sample in the noise disturbance sample set, the first... Sample values ​​for each calendar year For the first eigenmode functions of order 1 For residual terms;

[0083] 3.2. Take the mean value of the eigenmode functions of the same order as the stable eigenmode functions of the corresponding order:

[0084]

[0085] In the formula, For the first Stable intrinsic mode functions, This represents the number of noise disturbance samples in the noise disturbance sample set.

[0086] 3.3. Perform maximum entropy spectrum analysis on the stable intrinsic mode functions of each order to obtain the maximum entropy power spectrum. , For the first Maximum entropy power spectrum of first-order stable intrinsic mode functions;

[0087] 3.4. Regarding the maximum entropy power spectrum Find the frequency corresponding to its maximum point. According to frequency Solving for the dominant mode time scale , For the first The dominant mode timescale of the first-order stable intrinsic mode function.

[0088] This invention effectively suppresses the mode aliasing problem that easily occurs in traditional empirical mode decomposition by adopting an ensemble empirical mode decomposition method, and significantly improves the stability and physical interpretability of the decomposition results. At the same time, this method can adaptively extract the variation features of different time scales in the low temperature day series, laying the foundation for the targeted construction of multi-scale prediction models.

[0089] Step S4: Based on the stable intrinsic mode functions of each order and their dominant mode time scales, use time delay correlation analysis to obtain the predictive factors with physical significance for each stable intrinsic mode function.

[0090] The physically meaningful predictors of each order of stable intrinsic mode functions obtained using time-delay correlation analysis include:

[0091] 4.1 Obtain the sea surface temperature anomaly field (SSTA) of the target area for each calendar year;

[0092] 4.2. Based on the dominant mode timescale of each stable intrinsic mode function, the sea surface temperature anomaly field (SSTA) is subjected to timescale matching processing, and the time-delay correlation coefficient of each grid point in the target region is calculated. :

[0093]

[0094] In the formula, For the first Stable intrinsic mode functions, The sea surface temperature anomaly field SSTA for the x-th row and y-th column grid point within the target area after time matching processing. The time lag is in months. For the calendar year, In the first The time lag in the preset seasons of a natural year ; Let covariance function be used. for standard deviation for Standard deviation;

[0095] 4.3. Perform Student's t significance test on the time lag correlation coefficients under different time lags to obtain their confidence levels, and determine the sea surface temperature anomaly areas based on the confidence levels;

[0096] For the sea surface temperature anomaly regions corresponding to key months for each order of stable intrinsic mode functions, calculate the potential predictors for each order of stable intrinsic mode functions. :

[0097]

[0098] In the formula, This is a region with anomalies in sea surface temperature. For standardized processing, Latitude area weighting, , Let x be the latitude of the grid point in the x-th row and y-th column;

[0099] For potential predictors Perform physical consistency discrimination to screen predictors with clear physical significance.

[0100] Physical consistency is used as a latent predictor. The corresponding sea surface temperature anomaly signals can induce or modulate known large-scale circulation indices (such as Northwest Pacific anticyclones, East Asian summer monsoon anomalies, etc.) that affect the target prediction area.

[0101] The time-scale matching process for the sea surface temperature anomaly field (SSTA) based on the dominant mode timescale of each stable intrinsic mode function includes:

[0102] If the first Dominant mode timescale of first-order stable intrinsic mode functions Then the first The first-order stable intrinsic mode function is defined as the component on the decadal and higher time scales;

[0103] like Then the first The first-order stable intrinsic mode function is defined as the interannual time scale component;

[0104] For stable intrinsic mode functions belonging to the interdecadal and higher time scales, the sea surface temperature anomaly field (SSTA) is subjected to low-pass filtering.

[0105] For the stable intrinsic mode functions belonging to the interannual timescale component, linear detrending processing is performed on the sea surface temperature anomaly field SSTA.

[0106] This invention achieves a one-to-one match between predictor selection and intrinsic mode functions, thus avoiding the spurious correlation problem caused by scale mixing in traditional methods. At the same time, the scientific validity and stability of the predictor are ensured through physical mechanism verification, which improves the interpretability and business credibility of the prediction model.

[0107] Step S5: Construct a prediction model based on the random forest algorithm, build a training set according to the stable intrinsic mode functions of each order and the prediction factors, and train the prediction model.

[0108] The predictor is used as the input to the prediction model, and the corresponding stable intrinsic mode function is used as the output label of the prediction model. The mean absolute error is used as the loss function. The prediction model is trained and the parameters are optimized by five-fold cross-validation and grid search to determine the optimal parameter combination.

[0109] This invention is based on the random forest algorithm, which can effectively characterize the complex nonlinear relationship and multivariate interaction between the predictor and the number of low-temperature days. Through cross-validation and parameter optimization, it improves the generalization ability and stability of the model, reduces the risk of overfitting, and makes the prediction of different time scale components more robust and reliable.

[0110] Step S6: Based on the trained prediction model, predict the stable intrinsic mode functions of each order, and based on the prediction results, perform ensemble empirical mode reconstruction to obtain the predicted value of the average number of extreme low temperature days in the target area.

[0111] Obtain the values ​​of each predictor in the target prediction year at the corresponding lead time, input them into the prediction model corresponding to each order of intrinsic mode function, and obtain the prediction results of each order of intrinsic mode function in the target prediction year; according to the reconstruction principle of ensemble empirical mode decomposition, perform scale synthesis on the prediction results of each order of intrinsic mode function to obtain the reconstructed prediction value of the number of low temperature days; output the reconstructed prediction value as the seasonal prediction result of the number of low temperature days in the target prediction year.

[0112] This invention fully integrates high-frequency fluctuations and low-frequency background changes by performing scale synthesis on prediction results at different time scales, avoiding information loss caused by single-model prediction. This reconstruction method ensures the mathematical consistency and physical integrity of the prediction results, thereby significantly improving the overall accuracy and stability of seasonal predictions for the number of low-temperature days. This embodiment achieves high-precision seasonal prediction of extreme low temperatures in spring for target areas by combining ensemble empirical mode decomposition, physically constrained prediction factor selection, and random forest multi-scale modeling, demonstrating good prospects for business applications and promotional value.

[0113] Example 2

[0114] This invention provides an extreme low temperature seasonal prediction system based on time scale separation, comprising:

[0115] The data sampling module is configured to acquire the daily minimum temperature observation data of each grid point in the target area, and to obtain the time series of the average number of extreme low temperature days in the target area using the relative threshold method based on the daily minimum temperature observation data.

[0116] The noise perturbation module is configured to superimpose the time series of the average extreme low temperature days in the target area with multiple sets of amplitude-controlled Gaussian white noise to construct a noise perturbation sample set.

[0117] The mode decomposition module is configured to perform ensemble empirical mode decomposition and maximum entropy spectrum analysis on the noise disturbance sample set to obtain the stable intrinsic mode functions of each order and their dominant mode time scales.

[0118] The predictor determination module is configured to obtain physically meaningful predictors for each stable intrinsic mode function based on the time scale of each stable intrinsic mode function and its dominant mode function, using time delay correlation analysis.

[0119] The model building and training module is configured to build a prediction model based on the random forest algorithm, construct a training set based on the stable intrinsic mode functions of each order and the predictor factors, and train the prediction model.

[0120] The predictive mode reconstruction module is configured to predict stable intrinsic mode functions of each order based on the trained prediction model, and to perform ensemble empirical mode reconstruction based on the prediction results to obtain the predicted value of the average number of extreme low temperature days in the target area.

[0121] Example 3

[0122] Based on the extreme low temperature seasonal prediction method provided in Embodiment 1, this embodiment of the invention provides an electronic device, including a processor and a storage medium;

[0123] Storage media are used to store instructions;

[0124] The processor is used to perform operations according to instructions to execute the steps according to the method described above.

[0125] Example 4

[0126] Based on the extreme low temperature seasonal prediction method provided in Embodiment 1, this embodiment of the invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above method.

[0127] Example 5

[0128] Based on the extreme low temperature seasonal prediction method provided in Embodiment 1, this embodiment of the invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above method.

[0129] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0130] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0131] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0132] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0133] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting extreme low temperatures seasonally based on time-scale separation, characterized in that, include: Acquire daily minimum temperature observation data for each grid point within the target area, and obtain the time series of the average number of extreme low temperature days in the target area using the relative threshold method based on the daily minimum temperature observation data; The time series of the average number of extreme low temperature days in the target area is superimposed with multiple sets of amplitude-controlled Gaussian white noise to construct a noise perturbation sample set. The noise disturbance sample set was subjected to ensemble empirical mode decomposition and maximum entropy spectrum analysis to obtain the stable intrinsic mode functions of each order and their dominant mode time scales; Based on the stable intrinsic mode functions of each order and their dominant mode time scales, time delay correlation analysis is used to obtain the physically meaningful predictors of the stable intrinsic mode functions of each order. A prediction model is constructed based on the random forest algorithm, and a training set is built according to the stable intrinsic mode functions of each order and the prediction factors to train the prediction model. Based on the trained prediction model, stable intrinsic mode functions of each order are predicted, and based on the prediction results, ensemble empirical mode reconstruction is performed to obtain the predicted value of the average number of extreme low temperature days in the target area. The step of performing ensemble empirical mode decomposition and maximum entropy spectrum analysis on the noise perturbation sample set includes: Perform ensemble empirical mode decomposition on the noise perturbation sample set: ； In the formula, For the k-th noise disturbance sample in the noise disturbance sample set, the first... Sample values ​​for each calendar year For the first eigenmode functions of order 1 For the residual term, M For the total order; The mean value of the eigenmode functions of the same order is taken as the stable eigenmode functions of the corresponding order: ； In the formula, For the first Stable intrinsic mode functions, This represents the number of noise disturbance samples in the noise disturbance sample set. Maximum entropy spectrum analysis was performed on the stable intrinsic mode functions of each order to obtain the maximum entropy power spectrum. , For the first The maximum entropy power spectrum of the stable intrinsic mode function. f For frequency; For the maximum entropy power spectrum Find the frequency corresponding to its maximum point. According to frequency Solving for the dominant mode time scale , For the first The dominant mode timescale of the first-order stable intrinsic mode function; The predictive factors with physical meaning obtained from time-delay correlation analysis for each order of stable intrinsic mode function include: Obtain the sea surface temperature anomaly field (SSTA) of the target region for each calendar year; The sea surface temperature anomaly field (SSTA) is time-scale matched based on the dominant mode timescale of each stable intrinsic mode function, and the time-delay correlation coefficient of each grid point in the target region is calculated. : ； In the formula, For the first Stable intrinsic mode functions, The sea surface temperature anomaly field SSTA for the x-th row and y-th column grid point within the target area after time matching processing. The time lag is in months. For the calendar year, In the first The time lag in the preset seasons of a natural year ; Let covariance function be used. for standard deviation for Standard deviation; The Student's t significance test was performed on the time lag correlation coefficients under different time lags to obtain their confidence levels, and the sea surface temperature anomaly areas were determined based on the confidence levels. For the sea surface temperature anomaly regions corresponding to key months for each order of stable intrinsic mode functions, calculate the potential predictors for each order of stable intrinsic mode functions. : ； In the formula, This is a region with anomalies in sea surface temperature. For standardized processing, Latitude area weighting, , Let x be the latitude of the grid point in the x-th row and y-th column; For potential predictors Perform physical consistency discrimination to screen predictors with clear physical significance.

2. The extreme low temperature seasonal prediction method based on time scale separation according to claim 1, characterized in that, The daily minimum temperature observation data for each grid point within the target area are the daily minimum temperatures statistically analyzed for each preset season in each natural year; The time series of average extreme low temperature days in the target area obtained by using the relative threshold method based on the daily minimum temperature observation data includes: Obtain the daily minimum temperature of each grid point within the target area during the historical reference period; For each grid point, the daily minimum temperature of each day in the historical reference period is sorted from low to high, and the daily minimum temperature corresponding to the preset percentile of the sort is taken as the relative threshold of extreme low temperature of the grid point. Based on the relative threshold of extreme low temperature and the daily minimum temperature observation data of each grid point, determine the number of extreme low temperature days for each grid point in the preset season of each natural year; The average number of extreme low-temperature days for each grid point in each preset season of each calendar year is taken to obtain the average number of extreme low-temperature days in the target area in each preset season of each calendar year, and a time series is formed in calendar year units.

3. The extreme low temperature seasonal prediction method based on time scale separation according to claim 1, characterized in that, The time-scale matching process for the sea surface temperature anomaly field (SSTA) based on the dominant mode timescale of each stable intrinsic mode function includes: If the first Dominant mode timescale of first-order stable intrinsic mode functions Then the first The first-order stable intrinsic mode function is defined as the component on the decadal and higher time scales; like Then the first The first-order stable intrinsic mode function is defined as the interannual time scale component; For the stable intrinsic mode functions belonging to the interdecadal and higher time scale components, the sea surface temperature anomaly field SSTA is subjected to low-pass filtering. For the stable intrinsic mode functions belonging to the interannual timescale component, the sea surface temperature anomaly field SSTA is subjected to linear detrending processing.

4. The extreme low temperature seasonal prediction method based on time scale separation according to claim 1, characterized in that, Training the prediction model includes: The predictor is used as the input to the prediction model, and the corresponding stable intrinsic mode function is used as the output label of the prediction model. The mean absolute error is used as the loss function. The prediction model is trained and the parameters are optimized by five-fold cross-validation and grid search to determine the optimal parameter combination.

5. An extreme low temperature seasonal prediction system based on time scale separation, characterized in that, The extreme low temperature seasonal prediction system is configured to perform the steps of the method according to any one of claims 1-4, the extreme low temperature seasonal prediction system comprising: The data sampling module is configured to acquire the daily minimum temperature observation data of each grid point in the target area, and to obtain the time series of the average extreme low temperature days in the target area using the relative threshold method based on the daily minimum temperature observation data. The noise perturbation module is configured to superimpose the time series of the average extreme low temperature days in the target area with multiple sets of amplitude-controlled Gaussian white noise to construct a noise perturbation sample set; The mode decomposition module is configured to perform ensemble empirical mode decomposition and maximum entropy spectrum analysis on the noise disturbance sample set to obtain the stable intrinsic mode functions of each order and their dominant mode time scales. The predictor determination module is configured to obtain physically meaningful predictors for each stable intrinsic mode function based on the time scale of each stable intrinsic mode function and its dominant mode function, using time delay correlation analysis. The model building and training module is configured to build a prediction model based on the random forest algorithm, construct a training set according to the stable intrinsic mode functions of each order and the prediction factors, and train the prediction model. The predictive mode reconstruction module is configured to predict stable intrinsic mode functions of each order based on the trained prediction model, and to perform ensemble empirical mode reconstruction based on the prediction results to obtain the predicted value of the average number of extreme low temperature days in the target area.

6. An electronic device, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-4.

8. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-4.

Images