Abnormal snowfall prediction method in winter based on meteorological data and statistical model

Abstract

The present invention discloses a method for predicting abnormal snowfall in winter based on meteorological data and statistical models, and belongs to the technical field of meteorological prediction. The method includes meteorological station monitoring, data preprocessing, feature engineering, prediction model design and prediction of abnormal snowfall in winter. The present invention adopts tuna optimization algorithm for feature selection, and improves the basic performance of the prediction model by efficiently searching and selecting the best feature combination; adopts a hybrid learning model method for predictive model design, and optimizes the prediction ability of the model for time series data by designing a convolutional long short-term memory machine learning model and a gated recurrent unit, and also improves the training efficiency by designing a gated recurrent unit; adopts a method of combining a random forest statistical model and further designing a hybrid learning model, which improves the interpretability of the model while supplementing the feature extraction performance, and improves the overall usability of the method.

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorological forecasting, and specifically refers to a method for forecasting abnormal snowfall in winter based on meteorological data and statistical models. Background Art

[0002] Abnormal snowfall in winter can cause traffic jams, water and power outages, and crop damage, which has a negative impact on people's lives and production, and threatens people's life and property safety. The abnormal snowfall prediction method in winter based on meteorological data and statistical models uses meteorological observation data and statistical analysis techniques to predict whether the winter snowfall is abnormal. It helps relevant departments to make emergency preparations in advance and take corresponding measures to reduce the impact of abnormal snowfall in winter, thereby improving society's ability to respond to extreme weather and ensuring people's safety.

[0003] However, in the existing process of forecasting abnormal snowfall in winter, there are technical problems that abnormal snowfall involves complex meteorological data, and artificial feature selection or traditional optimization methods may lead to model overfitting or underfitting, thus affecting the accuracy of subsequent forecasts; there is a technical problem that a single model cannot fully handle the diverse features in abnormal snowfall forecasting, and the complexity of meteorological data determines that abnormal snowfall forecasting requires both adaptive design for time series data and optimization of computing efficiency and training time, resulting in poor prediction results of traditional methods; there is a technical problem that the complex conditions of meteorological data lead to low model interpretability when using machine learning model methods alone, which increases the difficulty of understanding model decisions, and thus leads to the practicality of existing methods decreasing with the complexity and performance of machine learning models. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method for predicting abnormal snowfall in winter based on meteorological data and statistical models. In view of the technical problem that in the process of predicting abnormal snowfall in winter, abnormal snowfall involves complex meteorological data, and artificial feature selection or traditional optimization methods may cause model overfitting or underfitting, thereby affecting the accuracy of subsequent predictions, this solution creatively adopts the tuna optimization algorithm for feature selection, and improves the basic performance of the prediction model by efficiently searching and selecting the best feature combination; in the process of predicting abnormal snowfall in winter, there is a single model that cannot comprehensively handle the diverse features in abnormal snowfall prediction, and the complexity of meteorological data determines that abnormal snowfall prediction requires both adaptive design for time series data and optimization of computing efficiency and training time, which leads to poor prediction effect of traditional methods. In order to solve the technical problem of abnormal snowfall in winter, this solution creatively adopts the method of hybrid learning model to design the prediction model. By designing the convolutional long short-term memory machine learning model and the gated recurrent unit, the convolutional long short-term memory machine learning model is used to optimize the model's prediction ability for time series data, and the training efficiency is improved by designing the gated recurrent unit. In order to solve the technical problem that in the process of predicting abnormal snowfall in winter, the complex conditions of meteorological data lead to the low model interpretability of the machine learning model method alone, which increases the difficulty of understanding the model decision, and then leads to the practicality of the existing method. The method is reduced with the complexity and performance of the machine learning model. This solution creatively adopts the method of combining the random forest statistical model and further designing the hybrid learning model. While supplementing the feature extraction performance, it improves the interpretability of the model and improves the overall usability of the method.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides a method for predicting abnormal snowfall in winter based on meteorological data and statistical models, the method comprising the following steps:

[0006] Step S1: weather station monitoring;

[0007] Step S2: data preprocessing;

[0008] Step S3: feature engineering;

[0009] Step S4: prediction model design;

[0010] Step S5: Prediction of abnormal snowfall in winter.

[0011] Furthermore, in step S1, the weather station monitoring is used to collect the original data required for the prediction of abnormal snowfall in winter, specifically, collecting historical meteorological data through weather station monitoring to obtain a meteorological data set, wherein the meteorological data set includes precipitation, temperature, humidity, wind speed and atmospheric pressure, and the precipitation includes rainfall and snowfall.

[0012] Furthermore, in step S2, the data preprocessing is used to optimize the data set, specifically by performing data denoising, outlier deletion and missing value filtering operations on the meteorological data set to obtain a meteorological optimized data set.

[0013] Further, in step S3, the feature engineering is used to select the optimal features from the data set, specifically using the tuna optimization algorithm to perform feature selection to obtain the optimal features of abnormal snowfall in winter, including the following steps:

[0014] Step S31: Initializing a tuna group, specifically, randomly initializing a group of tunas in the feature space of the meteorological optimization data set, each tuna representing a feature subset;

[0015] Step S32: defining a fitness function for evaluating the quality of the feature subset represented by each tuna, specifically taking the accuracy of the winter abnormal snowfall prediction model as the fitness function;

[0016] Step S33: simulating the foraging behavior of tuna, specifically updating the position of tuna by simulating the foraging behavior of tuna, including the following steps:

[0017] Step S331: Design a spiral foraging strategy to simulate tuna chasing prey. Specifically, the tuna explores the feature subset through a spiral search path. The calculation formula is:

[0018] ;

[0019] Where j is the tuna index, t is the iteration index, is the position of the jth tuna in the next iteration, is the first weight coefficient, is the optimal individual position, specifically the position of the optimal tuna in the current iteration, is a constant used to adjust individual position updates, is the second weight coefficient, is the current individual position, specifically the position of the jth tuna in the current iteration, is the position of the j-1th tuna in the current iteration, N is the number of tunas, rand is a random number in the range (0, 1), is the random individual position, specifically the position of the tuna randomly generated in the current iteration, t max is the maximum number of iterations;

[0020] Step S332: Design a parabolic foraging strategy to simulate the activity of tuna groups. Specifically, tuna explores feature subsets through a parabolic search path. The calculation formula is:

[0021] ;

[0022] Where TF is a random number in the range of {-1, 1}, and p is the parabola adjustment parameter, which is used to adjust the amplitude of the parabola during its development;

[0023] Step S333: updating the position of the tuna according to the spiral foraging strategy and the parabolic foraging strategy;

[0024] Step S34: Select the optimal feature subset by repeatedly performing step S33 operation max Perform iterative calculations and select the tuna with the highest fitness value as the optimal feature subset;

[0025] Step S35: feature selection, specifically, performing feature selection by initializing the tuna population, defining the fitness function, simulating the tuna foraging behavior and selecting the optimal feature subset to obtain the optimal feature of abnormal snowfall in winter, wherein the optimal feature of abnormal snowfall in winter specifically refers to the optimal feature subset.

[0026] Further, in step S4, the prediction model design is used to establish a winter abnormal snowfall prediction model, specifically, based on the meteorological optimization data set and the optimal features of winter abnormal snowfall, a hybrid learning model combining a convolutional long short-term memory machine learning model, a gated recurrent unit and a decision tree statistical model is used to design the prediction model, so as to obtain a winter abnormal snowfall prediction model;

[0027] The hybrid learning model combining the convolutional long short-term memory machine learning model, the gated recurrent unit and the decision tree statistical model comprises a convolutional long short-term memory sub-model, a gated recurrent unit sub-model and a decision tree statistical sub-model;

[0028] The convolutional long short-term memory sub-model is used to extract hidden features from time series data;

[0029] The gated recurrent unit sub-model is used to optimize training parameters and training time;

[0030] The decision tree statistical sub-model is used to supplement the feature extraction capability and enhance the interpretability of the model through the tree structure;

[0031] The prediction model design comprises the following steps:

[0032] Step S41: constructing a convolutional long short-term memory sub-model, including the following steps:

[0033] Step S411: construct a convolutional subnet, specifically, setting a convolutional layer and a maximum pooling layer in the convolutional subnet, and the calculation formula of the convolutional layer is:

[0034] ;

[0035] Where k is the feature map index, is the kth feature map output by the convolutional layer, μ(·) is the nonlinear activation function, and W k is the convolution kernel weight, specifically the convolution kernel weight connected to the kth feature map, * is the convolution operator, X is the input of the convolution layer, specifically refers to the optimal feature of abnormal snowfall in winter, g k is the convolutional layer bias;

[0036] Step S412: constructing a long short-term memory subnet, specifically receiving the output of the maximum pooling layer as the input of the long short-term memory subnet, and setting the input gate, forget gate, output gate and cell state in the long short-term memory subnet. The calculation formula is:

[0037] ;

[0038] Where T is the time step index, which is used to represent the current moment, i T is the output vector of the input gate, is the S-type activation function, W xi is the input weight of the input gate, which is used to control the influence of the input vector at the current moment, x T is the input of the LSTM subnet at the current moment, W hi is the recurrent weight of the input gate, which is used to control the influence of the long short-term memory hidden state at the previous moment, h T-1 is the long short-term memory hidden state of the previous moment, b i is the input gate bias term, f T is the output vector of the forget gate, W xf is the input weight of the forget gate, W hf is the recurrent weight of the forget gate, b f is the forget gate bias term, o T is the output vector of the output gate, W xo is the input weight of the output gate, W ho is the recurrent weight of the output gate, b o is the output gate bias term, c T is the current cell state, c T-1 is the cell state at the previous time step, is a candidate cell state;

[0039] Step S413: setting a fully connected layer, specifically connecting the output of the long short-term memory sub-network to the fully connected layer, and calculating the prediction result of the convolutional long short-term memory sub-model through the fully connected layer;

[0040] Step S414: constructing a convolutional long short-term memory sub-model by constructing the convolutional sub-network, constructing the long short-term memory sub-network, and setting the fully connected layer, and predicting the convolutional long short-term memory sub-model prediction output P by constructing the convolutional long short-term memory sub-model. CL ;

[0041] Step S42: construct a gated recurrent unit sub-model and predict the gated recurrent unit sub-model prediction output P GR , the calculation formula for constructing the gated recurrent unit sub-model is:

[0042] ;

[0043] In the formula, z T is the update gate, W z is the update gate weight, H T-1 is the gated recurrent hidden state at the previous moment, X T is the input of the gated recurrent unit sub-model at the current moment, b z is the update gate bias term, r T is the reset gate, W r is to reset the gate weight, b r is the reset gate bias term, is the candidate hidden state at the current moment, φ(·) is the hyperbolic tangent function, W H is the weight of the gated recurrent hidden state, b H is the bias term of the gated recurrent hidden state, H T is the gated recurrent hidden state at the current moment;

[0044] Step S43: construct a random forest statistical sub-model and calculate the random forest statistical sub-model prediction output P TE , the calculation formula for constructing the random forest statistical sub-model is:

[0045] ;

[0046] Where P TE is the prediction output of the random forest statistical sub-model, B is the number of decision trees, b is the decision tree index, u is the input of the random forest statistical sub-model, and t b (·) is the calculation function of the bth decision tree in the random forest;

[0047] Step S44: hybrid learning model construction, the calculation formula is:

[0048] ;

[0049] Where P F is the prediction output of the hybrid learning model, which is used to represent the hybrid fusion prediction output of all sub-models. is the convolutional long short-term memory sub-model fusion weight, P CL is the convolutional long short-term memory sub-model prediction output, is the gated recurrent unit sub-model fusion weight, P GR is the predicted output of the gated recurrent unit sub-model, is the fusion weight of the random forest statistical sub-model, P TE is the prediction output of the random forest statistical sub-model;

[0050] Step S45: training a winter abnormal snowfall prediction model, specifically, training the model by constructing the winter abnormal snowfall prediction model, constructing the gated cyclic unit sub-model, constructing the random forest statistical sub-model and the hybrid learning model to obtain the winter abnormal snowfall prediction model.

[0051] Further, in step S5, the winter abnormal snowfall prediction is used to adopt a prediction model to perform winter abnormal snowfall prediction, specifically using the winter abnormal snowfall prediction model to perform winter abnormal snowfall prediction based on meteorological data and statistical models and combined with hybrid learning to obtain winter abnormal snowfall prediction data.

[0052] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0053] (1) In order to solve the technical problem that abnormal snowfall in winter involves complex meteorological data, artificial feature selection or traditional optimization methods may lead to model overfitting or underfitting, thus affecting the accuracy of subsequent predictions, this solution creatively adopts the tuna optimization algorithm for feature selection. By efficiently searching and selecting the best feature combination, the basic performance of the prediction model is improved.

[0054] (2) In the process of predicting abnormal snowfall in winter, there is a technical problem that a single model cannot fully handle the diverse characteristics of abnormal snowfall prediction. The complexity of meteorological data determines that abnormal snowfall prediction requires both adaptive design for time series data and optimization of computing efficiency and training time, which leads to poor prediction results of traditional methods. This scheme creatively adopts the method of hybrid learning model to design the prediction model. By designing a convolutional long short-term memory machine learning model and a gated recurrent unit, the convolutional long short-term memory machine learning model optimizes the model's prediction ability for time series data, and also improves the training efficiency by designing a gated recurrent unit.

[0055] (3) In the process of predicting abnormal snowfall in winter, the complex conditions of meteorological data lead to low model interpretability when using machine learning model methods alone, which increases the difficulty of understanding model decisions, and further leads to the technical problem that the practicality of existing methods decreases with the complexity and performance of machine learning models. This solution creatively adopts a method of combining random forest statistical models and further designing hybrid learning models. While supplementing the feature extraction performance, it also improves the interpretability of the model and improves the overall usability of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A schematic diagram of a flow chart of a method for predicting abnormal snowfall in winter based on meteorological data and statistical models provided by the present invention;

[0057] Figure 2 is a schematic flow chart of step S3;

[0058] Figure 3 is a schematic flow chart of step S4;

[0059] Figure 4 A data flow diagram of the method for predicting abnormal snowfall in winter based on meteorological data and statistical models provided by the present invention.

[0060] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0062] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0063] Example 1, see Figure 1 The present invention provides a method for predicting abnormal snowfall in winter based on meteorological data and statistical models, the method comprising the following steps:

[0064] Step S1: weather station monitoring;

[0065] Step S2: data preprocessing;

[0066] Step S3: feature engineering;

[0067] Step S4: prediction model design;

[0068] Step S5: Prediction of abnormal snowfall in winter.

[0069] Example 2, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S1, the weather station monitoring is used to collect the original data required for the prediction of abnormal snowfall in winter. Specifically, historical meteorological data is collected through weather station monitoring to obtain a meteorological data set. The meteorological data set includes precipitation, temperature, humidity, wind speed and atmospheric pressure. The precipitation includes rainfall and snowfall.

[0070] Example 3, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S2, the data preprocessing is used to optimize the data set, specifically, by performing data denoising, outlier deletion and missing value filtering operations on the meteorological data set to obtain a meteorological optimized data set.

[0071] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the feature engineering is used to select the optimal feature from the data set, specifically using the tuna optimization algorithm to perform feature selection to obtain the optimal feature of abnormal snowfall in winter, including the following steps:

[0072] Step S31: Initializing a tuna group, specifically, randomly initializing a group of tunas in the feature space of the meteorological optimization data set, each tuna representing a feature subset;

[0073] Step S32: defining a fitness function for evaluating the quality of the feature subset represented by each tuna, specifically taking the accuracy of the winter abnormal snowfall prediction model as the fitness function;

[0074] Step S33: simulating the foraging behavior of tuna, specifically updating the position of tuna by simulating the foraging behavior of tuna, including the following steps:

[0075] Step S331: Design a spiral foraging strategy to simulate tuna chasing prey. Specifically, the tuna explores the feature subset through a spiral search path. The calculation formula is:

[0076] ;

[0077] Where j is the tuna index, t is the iteration index, is the position of the jth tuna in the next iteration, is the first weight coefficient, is the optimal individual position, specifically the position of the optimal tuna in the current iteration, is a constant used to adjust individual position updates, is the second weight coefficient, is the current individual position, specifically the position of the jth tuna in the current iteration, is the position of the j-1th tuna in the current iteration, N is the number of tunas, rand is a random number in the range (0, 1), is the random individual position, specifically the position of the tuna randomly generated in the current iteration, t max is the maximum number of iterations;

[0078] Step S332: Design a parabolic foraging strategy to simulate the activity of tuna groups. Specifically, tuna explores feature subsets through a parabolic search path. The calculation formula is:

[0079] ;

[0080] Where TF is a random number in the range of {-1, 1}, and p is a parabola adjustment parameter used to adjust the amplitude of the parabola during its development. The calculation formula of the parabola adjustment parameter is: ;

[0081] Step S333: updating the position of the tuna according to the spiral foraging strategy and the parabolic foraging strategy;

[0082] Step S34: Select the optimal feature subset by repeatedly performing step S34 operation max Perform iterative calculations and select the tuna with the highest fitness value as the optimal feature subset;

[0083] Step S35: feature selection, specifically, performing feature selection by initializing the tuna group, defining the fitness function, simulating the tuna foraging behavior and selecting the optimal feature subset to obtain the optimal feature of abnormal snowfall in winter, wherein the optimal feature of abnormal snowfall in winter specifically refers to the optimal feature subset;

[0084] By performing the above operations, in order to solve the technical problem that in the process of abnormal snowfall prediction in winter, abnormal snowfall involves complex meteorological data, and manual feature selection or traditional optimization methods may lead to model overfitting or underfitting, thereby affecting the accuracy of subsequent predictions, this solution creatively uses the tuna optimization algorithm for feature selection, and improves the basic performance of the prediction model by efficiently searching and selecting the best feature combination.

[0085] Example 5, see Figure 1 , Figure 3 and Figure 4 , this embodiment is based on the above embodiment. In step S4, the prediction model design is used to establish a winter abnormal snowfall prediction model. Specifically, based on the meteorological optimization data set and the optimal features of winter abnormal snowfall, a hybrid learning model combining a convolutional long short-term memory machine learning model, a gated recurrent unit and a decision tree statistical model is used to design the prediction model to obtain a winter abnormal snowfall prediction model;

[0086] The hybrid learning model combining the convolutional long short-term memory machine learning model, the gated recurrent unit and the decision tree statistical model comprises a convolutional long short-term memory sub-model, a gated recurrent unit sub-model and a decision tree statistical sub-model;

[0087] The convolutional long short-term memory sub-model is used to extract hidden features from time series data;

[0088] The gated recurrent unit sub-model is used to optimize training parameters and training time;

[0089] The decision tree statistical sub-model is used to supplement the feature extraction capability and enhance the interpretability of the model through the tree structure;

[0090] The prediction model design comprises the following steps:

[0091] Step S41: constructing a convolutional long short-term memory sub-model, including the following steps:

[0092] Step S411: construct a convolutional subnet, specifically, setting a convolutional layer and a maximum pooling layer in the convolutional subnet, and the calculation formula of the convolutional layer is:

[0093] ;

[0094] Where k is the feature map index, is the kth feature map output by the convolutional layer, μ(·) is the nonlinear activation function, and W k is the convolution kernel weight, specifically the convolution kernel weight connected to the kth feature map, * is the convolution operator, X is the input of the convolution layer, specifically refers to the optimal feature of abnormal snowfall in winter, g k is the convolutional layer bias;

[0095] Step S412: constructing a long short-term memory subnet, specifically receiving the output of the maximum pooling layer as the input of the long short-term memory subnet, and setting the input gate, forget gate, output gate and cell state in the long short-term memory subnet. The calculation formula is:

[0096] ;

[0097] Where T is the time step index, which is used to represent the current moment, i T is the output vector of the input gate, is the S-type activation function, W xi is the input weight of the input gate, which is used to control the influence of the input vector at the current moment, x T is the input of the LSTM subnet at the current moment, W hi is the recurrent weight of the input gate, which is used to control the influence of the long short-term memory hidden state at the previous moment, h T-1 is the long short-term memory hidden state of the previous moment, b i is the input gate bias term, f T is the output vector of the forget gate, W xf is the input weight of the forget gate, W hf is the recurrent weight of the forget gate, b f is the forget gate bias term, o T is the output vector of the output gate, W xo is the input weight of the output gate, W ho is the recurrent weight of the output gate, b o is the output gate bias term, c T is the current cell state, c T-1 is the cell state at the previous time step, is a candidate cell state;

[0098] Step S413: setting a fully connected layer, specifically connecting the output of the long short-term memory sub-network to the fully connected layer, and calculating the prediction result of the convolutional long short-term memory sub-model through the fully connected layer;

[0099] Step S414: constructing a convolutional long short-term memory sub-model by constructing the convolutional sub-network, constructing the long short-term memory sub-network, and setting the fully connected layer, and predicting the convolutional long short-term memory sub-model prediction output P by constructing the convolutional long short-term memory sub-model. CL ;

[0100] Step S42: construct a gated recurrent unit sub-model and predict the gated recurrent unit sub-model prediction output P GR , the calculation formula for constructing the gated recurrent unit sub-model is:

[0101] ;

[0102] In the formula, z T is the update gate, W z is the update gate weight, H T-1 is the gated recurrent hidden state at the previous moment, X T is the input of the gated recurrent unit sub-model at the current moment, b z is the update gate bias term, r T is the reset gate, W r is to reset the gate weight, b r is the reset gate bias term, is the candidate hidden state at the current moment, φ(·) is the hyperbolic tangent function, W H is the weight of the gated recurrent hidden state, b H is the bias term of the gated recurrent hidden state, H T is the gated recurrent hidden state at the current moment;

[0103] Step S43: construct a random forest statistical sub-model and calculate the random forest statistical sub-model prediction output P TE , the calculation formula for constructing the random forest statistical sub-model is:

[0104] ;

[0105] Where P TE is the prediction output of the random forest statistical sub-model, B is the number of decision trees, b is the decision tree index, u is the input of the random forest statistical sub-model, and t b (·) is the calculation function of the bth decision tree in the random forest;

[0106] Step S44: hybrid learning model construction, the calculation formula is:

[0107] ;

[0108] Where P F is the prediction output of the hybrid learning model, which is used to represent the hybrid fusion prediction output of all sub-models. is the convolutional long short-term memory sub-model fusion weight, P CL is the predicted output of the convolutional long short-term memory sub-model, is the gated recurrent unit sub-model fusion weight, P GR is the predicted output of the gated recurrent unit sub-model, is the fusion weight of the random forest statistical sub-model, P TE is the prediction output of the random forest statistical sub-model;

[0109] Step S45: training a winter abnormal snowfall prediction model, specifically, training the model through the winter abnormal snowfall prediction model, constructing the gated recurrent unit sub-model, constructing the random forest statistical sub-model and the hybrid learning model, to obtain the winter abnormal snowfall prediction model;

[0110] By performing the above operations, in view of the fact that a single model cannot fully handle the diverse characteristics of abnormal snowfall prediction in winter, and the complexity of meteorological data determines that abnormal snowfall prediction requires both adaptive design for time series data and optimization of computing efficiency and training time, which leads to the technical problem that the traditional method has poor prediction effect, this solution creatively adopts the hybrid learning model method to design the prediction model, and optimizes the prediction ability of the model for time series data through the convolutional long short-term memory machine learning model and the gated recurrent unit through the convolutional long short-term memory machine learning model, and also improves the training efficiency by designing the gated recurrent unit;

[0111] In the process of predicting abnormal snowfall in winter, the complex conditions of meteorological data lead to low model interpretability when using machine learning model methods alone, which increases the difficulty of understanding model decisions, and further leads to the technical problem that the practicality of existing methods decreases with the complexity and performance of machine learning models. This solution creatively adopts a method of combining random forest statistical models and further designing hybrid learning models. While supplementing the feature extraction performance, it improves the interpretability of the model and improves the overall usability of the method.

[0112] Example 6, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S5, the abnormal snowfall in winter prediction is used to use a prediction model to predict abnormal snowfall in winter. Specifically, the abnormal snowfall in winter prediction model is used to perform abnormal snowfall in winter prediction based on meteorological data and statistical models, and combined with hybrid learning to obtain abnormal snowfall in winter prediction data. The abnormal snowfall in winter prediction data includes abnormal snowfall probability, predicted snowfall amount and predicted snowfall level.

[0113] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0114] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0115] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for predicting abnormal snowfall in winter based on meteorological data and statistical models, characterized by: The method comprises the following steps: Step S1: Meteorological station monitoring is used to collect raw data required for winter abnormal snowfall prediction, specifically, collecting historical meteorological data through meteorological station monitoring to obtain a meteorological data set, wherein the meteorological data set includes precipitation, temperature, humidity, wind speed and atmospheric pressure, and the precipitation includes rainfall and snowfall; Step S2: data preprocessing; Step S3: feature engineering, specifically, using the tuna optimization algorithm to select features and obtain the optimal features of abnormal snowfall in winter; Step S4: prediction model design, specifically, based on the meteorological optimization data set and the optimal features of abnormal snowfall in winter, a hybrid learning model combining a convolutional long short-term memory machine learning model, a gated recurrent unit and a decision tree statistical model is used to design a prediction model to obtain a prediction model for abnormal snowfall in winter; The hybrid learning model combining the convolutional long short-term memory machine learning model, the gated recurrent unit and the decision tree statistical model comprises a convolutional long short-term memory sub-model, a gated recurrent unit sub-model and a decision tree statistical sub-model; The convolutional long short-term memory sub-model is used to extract hidden features from time series data; The gated recurrent unit sub-model is used to optimize training parameters and training time; The decision tree statistical sub-model is used to supplement the feature extraction capability and enhance the interpretability of the model through the tree structure; The prediction model design is used to establish a winter abnormal snowfall prediction model, including the following steps: Step S41: constructing a convolutional long short-term memory sub-model, including the following steps: Step S411: construct a convolutional subnet, specifically, setting a convolutional layer and a maximum pooling layer in the convolutional subnet, and the calculation formula of the convolutional layer is: ； Where k is the feature map index, is the kth feature map output by the convolutional layer, μ(·) is the nonlinear activation function, and W k is the convolution kernel weight, specifically the convolution kernel weight connected to the kth feature map, * is the convolution operator, X is the input of the convolution layer, specifically refers to the optimal feature of abnormal snowfall in winter, g k is the convolutional layer bias; Step S412: constructing a long short-term memory subnet, specifically receiving the output of the maximum pooling layer as the input of the long short-term memory subnet, and setting the input gate, forget gate, output gate and cell state in the long short-term memory subnet. The calculation formula is: ； Where T is the time step index, which is used to represent the current moment, i T is the output vector of the input gate, is the S-type activation function, W xi is the input weight of the input gate, which is used to control the influence of the input vector at the current moment, x T is the input of the LSTM subnet at the current moment, W hi is the recurrent weight of the input gate, which is used to control the influence of the long short-term memory hidden state at the previous moment, h T-1 is the long short-term memory hidden state of the previous moment, b i is the input gate bias term, f T is the output vector of the forget gate, W xf is the input weight of the forget gate, W hf is the recurrent weight of the forget gate, b f is the forget gate bias term, o T is the output vector of the output gate, W xo is the input weight of the output gate, W ho is the recurrent weight of the output gate, b o is the output gate bias term, c T is the current cell state, c T-1 is the cell state at the previous time step, is a candidate cell state; Step S413: setting a fully connected layer, specifically connecting the output of the long short-term memory sub-network to the fully connected layer, and calculating the prediction result of the convolutional long short-term memory sub-model through the fully connected layer; Step S414: constructing a convolutional long short-term memory sub-model by constructing the convolutional sub-network, constructing the long short-term memory sub-network, and setting the fully connected layer, and predicting the convolutional long short-term memory sub-model prediction output P by constructing the convolutional long short-term memory sub-model. CL ; Step S42: construct a gated recurrent unit sub-model and predict the gated recurrent unit sub-model prediction output P GR , the calculation formula for constructing the gated recurrent unit sub-model is: ； In the formula, z T is the update gate, W z is the update gate weight, H T-1 is the gated recurrent hidden state at the previous moment, X T is the input of the gated recurrent unit sub-model at the current moment, b z is the update gate bias term, r T is the reset gate, W r is to reset the gate weight, b r is the reset gate bias term, is the candidate hidden state at the current moment, φ(·) is the hyperbolic tangent function, W H is the weight of the gated recurrent hidden state, b H is the bias term of the gated recurrent hidden state, H T is the gated recurrent hidden state at the current moment; Step S43: construct a random forest statistical sub-model and calculate the random forest statistical sub-model prediction output P TE , the calculation formula for constructing the random forest statistical sub-model is: ； Where P TE is the prediction output of the random forest statistical sub-model, B is the number of decision trees, b is the decision tree index, u is the input of the random forest statistical sub-model, and t b (·) is the calculation function of the bth decision tree in the random forest; Step S44: hybrid learning model construction, the calculation formula is: ； Where P F is the prediction output of the hybrid learning model, which is used to represent the hybrid fusion prediction output of all sub-models. is the convolutional long short-term memory sub-model fusion weight, P CL is the convolutional long short-term memory sub-model prediction output, is the gated recurrent unit sub-model fusion weight, P GR is the predicted output of the gated recurrent unit sub-model, is the fusion weight of the random forest statistical sub-model, P TE is the prediction output of the random forest statistical sub-model; Step S45: training a winter abnormal snowfall prediction model, specifically, training the model through the winter abnormal snowfall prediction model, constructing the gated recurrent unit sub-model, constructing the random forest statistical sub-model and the hybrid learning model, to obtain the winter abnormal snowfall prediction model; Step S5: Prediction of abnormal snowfall in winter.

2. The method for predicting abnormal snowfall in winter based on meteorological data and statistical models according to claim 1, characterized in that: In step S3, the feature engineering is used to select the best features from the data set, including the following steps: Step S31: Initializing a tuna group, specifically, randomly initializing a group of tunas in the feature space of the meteorological optimization data set, each tuna representing a feature subset; Step S32: defining a fitness function for evaluating the quality of the feature subset represented by each tuna, specifically taking the accuracy of the winter abnormal snowfall prediction model as the fitness function; Step S33: simulating the foraging behavior of tuna, specifically updating the position of tuna by simulating the foraging behavior of tuna; Step S34: Select the optimal feature subset by repeatedly performing step S33 operation max Perform iterative calculations and select the tuna with the highest fitness value as the optimal feature subset; Step S35: feature selection, specifically, performing feature selection by initializing the tuna population, defining the fitness function, simulating the tuna foraging behavior and selecting the optimal feature subset to obtain the optimal feature of abnormal snowfall in winter, wherein the optimal feature of abnormal snowfall in winter specifically refers to the optimal feature subset.

3. The method for predicting abnormal snowfall in winter based on meteorological data and statistical models according to claim 2 is characterized in that: In step S33, the simulating tuna foraging behavior comprises the following steps: Step S331: Design a spiral foraging strategy to simulate tuna chasing prey. Specifically, the tuna explores the feature subset through a spiral search path. The calculation formula is: ； Where j is the tuna index, t is the iteration index, is the position of the jth tuna in the next iteration, is the first weight coefficient, is the optimal individual position, specifically the position of the optimal tuna in the current iteration, is a constant used to adjust individual position updates, is the second weight coefficient, is the current individual position, specifically the position of the jth tuna in the current iteration, is the position of the j-1th tuna in the current iteration, N is the number of tunas, rand is a random number in the range (0, 1), is the random individual position, specifically the position of the tuna randomly generated in the current iteration, t max is the maximum number of iterations; Step S332: Design a parabolic foraging strategy to simulate the activity of tuna groups. Specifically, tuna explores feature subsets through a parabolic search path. The calculation formula is: ； Where TF is a random number in the range of {-1, 1}, and p is the parabola adjustment parameter, which is used to adjust the amplitude of the parabola during its development; Step S333: updating the position of the tuna according to the spiral foraging strategy and the parabolic foraging strategy.

4. The method for predicting abnormal snowfall in winter based on meteorological data and statistical models according to claim 3 is characterized in that: In step S2, the data preprocessing is used to optimize the data set, specifically, by performing data denoising, outlier deletion and missing value filtering operations on the meteorological data set to obtain a meteorological optimized data set.

5. The method for predicting abnormal snowfall in winter based on meteorological data and statistical models according to claim 4 is characterized in that: In step S5, the abnormal snowfall in winter prediction is used to adopt a prediction model to predict abnormal snowfall in winter, specifically, to use the abnormal snowfall in winter prediction model to perform abnormal snowfall in winter prediction based on meteorological data and statistical models, and combined with hybrid learning to obtain abnormal snowfall in winter prediction data.

Images