Meteorological disaster prediction method based on AI

Abstract

The invention relates to an AI-based meteorological disaster prediction method, which belongs to the field of marine meteorological data analysis and prediction, and comprises the following steps: (1) data acquisition and preprocessing; (2) constructing a time sequence data set; (3) designing and constructing a model; (4) model training; (5) model verification; and (6) model application: applying the trained model to an actual marine meteorological disaster prediction task. The method has the advantages that (1) the prediction precision is improved, and the prediction precision of the marine meteorological data is effectively improved by combining the Bi-LSTM and the MLP model; and (2) the robustness and stability of the model are enhanced, and the robustness of the model in different data environments is enhanced through the synergistic effect of Bi-LSTM and MLP. And (3) the flexibility and application universality of the model are improved, and the flexible structure enables the model to be widely applied to a plurality of fields such as marine meteorology, environmental monitoring, meteorological disaster early warning and the like, and has good adaptability.

Description

Technical Field

[0001] The present invention relates to a meteorological disaster prediction method based on AI, and belongs to the field of analysis and prediction of marine meteorological data. Background Art

[0002] With the intensification of global climate change, the importance of marine meteorological disaster prediction has become increasingly prominent. Marine meteorological data is not only of great significance to the protection of marine ecosystems, the study of climate change, and marine transportation and fishery management, but also directly affects human daily life and economic activities. Therefore, the development of accurate and efficient marine meteorological disaster prediction methods has a far-reaching impact on ensuring the safety and sustainable development of human activities. Zhang Feifei studied the regression and prediction of marine meteorological elements, including wind speed, temperature, air pressure, relative humidity, rainfall and visibility data, and proposed a marine meteorological prediction model based on spatiotemporal data model; Wang Shengzheng et al. proposed an alternating sparse autoencoder (ASAE) network model based on the sparse autoencoder (SAE) network model, and constructed a prediction framework for the impact of marine meteorology on ship speed; Li Tianbao et al. used deep learning algorithms to make important progress in the prediction of extreme marine weather phenomena. Taking El Niño, typhoons and short-term rainfall as examples, they introduced prediction algorithms based on artificial intelligence and analyzed the advantages and disadvantages of these algorithms. The existing marine meteorological prediction methods are shown below.

[0003] Current weather forecasting techniques are mainly divided into empirical statistics, dynamic analysis and model output statistics, which are based on traditional meteorology and statistics.

[0004] (1) The empirical statistical method is an empirical statistical method based on mathematical statistics. It makes a comprehensive and systematic statistical analysis of the historical changes of meteorological elements, combines the influence of relevant factors, summarizes certain change rules, and determines the prediction equation of local meteorological elements. The characteristics of this method are long calculation process, detailed statistical analysis, and simple operation. However, this method has obvious limitations, that is, it lacks specific analysis of the evolution process within the meteorological system, only makes rough estimates, and cannot provide detailed change processes and results. Therefore, it has a large subjective judgment component and is very prone to large errors.

[0005] (2) Dynamic analysis uses physical and mathematical tools to study the dynamic and thermal changes involved in the evolution of atmospheric movement and meteorological systems. This analysis usually includes two main directions: low-temperature thermodynamics and lunar dynamic extension method. Low-temperature thermodynamics introduces water and heat balance equations to construct mathematical models to describe the state changes of meteorological systems and the basic physical laws they follow. Research in this field has been relatively mature and has achieved extensive results in engineering applications. The lunar dynamic extension law combines dynamic and statistical methods to establish a statistical model by analyzing the correlation between circulation characteristics and meteorological elements. This method is proposed for my country's specific geographical environment. Although it is relatively new, many numerical experiments have been carried out in the field of meteorological forecasting. However, dynamic analysis also has certain limitations. For example, it is subject to the limitations of model prediction accuracy and sample size, and it still needs to be further improved and refined in practical applications.

[0006] (3) Model output statistics method uses probability statistics methods to study the historical observation data of a large number of meteorological elements, analyze the statistical laws of climate change, establish a reasonable prediction model, and predict future meteorological conditions through statistical mathematical models. This method originated in my country and has become one of the most widely used technologies in domestic meteorological research. Its core is to build a prediction model based on time series analysis through historical meteorological observation data. In the modeling process, in addition to mastering the basic characteristics and properties of the data, it is also necessary to select prediction factors that are relevant to the prediction and have physical significance from a large number of meteorological elements. Since meteorological data belongs to nonlinear time series, the randomness caused by internal changes in the data must also be taken into account, so the analysis method is different from traditional data. The characteristic of nonlinear time series is that the correlation of observation data changes with the time dimension and the order of data delay. Meteorological prediction technology based on time series models uses these characteristics to predict future meteorological conditions by analyzing historical data.

[0007] In recent years, the rapid development of deep learning technology has provided new ideas for marine meteorological disaster prediction. Traditional meteorological disaster prediction methods usually rely on physical models and statistical methods, which require a large amount of observational data and computing resources. However, deep learning methods can efficiently make predictions without relying on complex physical models by learning massive historical data. Therefore, deep learning technology has shown great potential in solving complex nonlinear problems in marine meteorological disaster prediction.

[0008] (1) Application of time series prediction models: Recurrent neural networks (RNNs), long short-term memory networks (LSTMs) and their improved versions, such as bidirectional LSTMs (Bi-LSTMs) and models with attention mechanisms, in deep learning have shown good results in processing marine meteorological data. They are able to capture long-term dependencies, especially in the prediction of time series features such as wave height, sea surface temperature, and wind speed, with high accuracy.

[0009] (2) Multimodal data fusion. Marine meteorological disaster prediction usually requires the combination of multiple data sources, such as ocean observation data, satellite remote sensing data, and meteorological forecast data. Deep learning models, especially convolutional neural networks (CNNs) and multi-layer perceptrons (MLPs), are used to integrate these multi-source data to improve the accuracy of predictions. These models can better capture complex marine meteorological patterns through feature extraction and data fusion.

[0010] (3) Exploration of space-time modeling. In recent years, space-time models based on deep learning, such as space-time convolutional networks (ST-CNN) and space-time graph convolutional networks (ST-GCN), have been widely used in the field of marine meteorological forecasting. These models can simultaneously consider the spatial and temporal dependencies of marine meteorological data, thereby improving the timeliness and accuracy of forecasts. Summary of the invention

[0011] In order to overcome the defects of the prior art, the present invention provides a meteorological disaster prediction method based on AI. The technical solution of the present invention is:

[0012] A meteorological disaster prediction method based on AI, comprising the following steps:

[0013] (1) Data collection and preprocessing: Obtain historical marine meteorological data, clean and normalize the data, and perform time stamp correction;

[0014] (2) Construction of time series data set: set a fixed time step and use the multi-dimensional feature data of the previous n days to predict the target value of the n+1 day;

[0015] (3) Model design and construction: Design and construct a deep learning model that combines a bidirectional long short-term memory network (Bi-LSTM) and a multi-layer perceptron (MLP). Bi-LSTM is used to extract dynamic features of time series, and MLP is used to perform nonlinear modeling of the relationship between features. The final prediction value is obtained by weighted fusion of the results of the two models.

[0016] (4) Model training, using error back propagation to update model parameters and adjust hyperparameters to optimize model performance;

[0017] (5) Model validation: Use the validation set to evaluate the model and adjust the model hyperparameters and weighting coefficients to optimize the prediction performance;

[0018] (6) Model application: applying the trained model to actual marine meteorological disaster prediction tasks.

[0019] The step (1) is specifically as follows:

[0020] 1.1 Handle missing values, outliers and noisy data; for missing values, use interpolation, mean filling or time series-based prediction methods to fill them; for outliers, use statistical analysis and threshold setting to identify and remove data points that deviate from the normal range;

[0021] 1.2 Normalize or standardize each feature data;

[0022] 1.3 Correct and unify the timestamps to ensure the consistency of each feature data in the time dimension;

[0023] 1.4 Smooth the data or extract the periodic components.

[0024] The step (2) is specifically as follows:

[0025] The time series data set construction steps include:

[0026] 2.1 Traverse the entire data set and generate several input and output pairs, ensuring that the time correspondence between the input sequence and the output sequence is accurate to avoid prediction errors caused by data misalignment;

[0027] 2.2 Divide the data set into training set and test set. The division can be done in chronological order to avoid future data leakage into model training;

[0028] 2.3 Convert data into tensor format.

[0029] The step (3) is specifically as follows:

[0030] 3.1 The bidirectional characteristics of the Bi-LSTM network model are used to obtain the forward and backward correlation information of the time series data; the Bi-LSTM network model captures the forward and backward features of the time series data at the same time, and uses the past and future information to mine the deep patterns in the data. Specifically, the input data is a multidimensional feature sequence divided by time steps. The Bi-LSTM network model encodes the sequence data through the forward and backward LSTM layers to generate hidden states for each time step. These hidden states contain the dynamic information of the time series at different time points; 3.2 The MLP model is used to perform horizontal prediction of features and deeply mine the nonlinear relationship between features; the MLP model takes the multidimensional features of the current time step as input, and learns the complex nonlinear relationship between features through multiple layers of nonlinear transformation; the model includes an input layer, several hidden layers and an output layer, and the hidden layer uses an activation function to enhance the expressive power of the model.

[0031] The step (4) is specifically as follows: the model training includes two parts: Bi-LSTM model and MLP model; Bi-LSTM is used to process time series data, and MLP is used to process the nonlinear relationship of the current time step feature; the training steps are as follows:

[0032] 4.1 In Bi-LSTM, time series data is input into the forward and backward LSTM layers to capture the bidirectional dependencies in the time series; the output of Bi-LSTM passes through the fully connected layer to generate longitudinal prediction results;

[0033] (2) MLP processes the features of the current time step. Through a multi-layer fully connected network, MLP extracts nonlinear features layer by layer and finally generates a lateral correction result.

[0034] (3) Error back propagation to update each weight;

[0035] (4) Adjust the model hyperparameters and repeat the above steps, compare the verification value and error value, select the optimal model, and complete the model training.

[0036] In the above step 3.1, the LSTM network has three gates: forget gate, input gate and output gate. There are two activation functions in LSTM: σ and Tanh. Among them, σ is the Sigmoid activation function, which will be used in the forget gate, input gate and output gate, and σ is used to output [0, 1]. When generating candidate memories, the Tanh activation function is used to output [-1, 1]. The forget gate f t : Used to determine the cell state C from the previous moment t-1 The amount of information that should be discarded and retained until the current time C t Information, f t =σ(W f [h t-1, x t ]+b f );W f is the weight matrix of the forget gate; h t-1 is the output of the hidden layer at the previous moment; x t b is the input data at time t; f is the bias vector of the forget gate; input gate i t : Used to determine the input information x t Save to cell state C t The degree of t =σ(W i [h t-1 , x t ]+b i );W i is the weight matrix of the input gate; b i is the bias vector of the input gate; the candidate state created by the tanh layer for: Where: w c is the weight matrix of the current state; b is the bias vector of the current state;

[0037] Unit information update: used to calculate the unit state c at the current moment t ,

[0038] Output gate: O t Used to control the status of the unit C t How much information is input to the current hidden layer output value h; o t =σ(W o [h t-1 , x t ]+b o );h t =o t ·tanh(c i );

[0039] Where: b o is the bias term of the output gate; W o is the weight matrix of the output gate; tanh is the hyperbolic tangent activation function; tanh and σ are defined as:

[0040]

[0041] The Bi-LSTM network comprises two LSTM units; one LSTM unit processes input data in the forward order of the time series; the other LSTM unit processes input data in the reverse order of the time series; the forward LSTM and backward LSTM of the Bi-LSTM respectively calculate the hidden state of the time series, and the formula is as follows:

[0042]

[0043] In the formula, h t fw and h t bw Represent the forward and backward hidden states respectively, x t are input features, and is the hidden state of the previous time step and the next time step; the final hidden state is obtained by concatenating the forward and backward hidden states:

[0044] In step 3.3, the MLP input layer receives the multidimensional feature vector x at the current time step. t , whose dimension is d input , that is, the number of features; input feature vector x t =[x1, x2, ..., x dinput ] is passed to the hidden layer through the first fully connected layer; each hidden layer processes the output of the previous layer through linear transformation and activation function; assuming there are N hidden layers, the output of the i-th hidden layer is expressed as:

[0045] h (i) =σW (i) h (i-1) +b (i) ), where W (i) is the weight matrix of the i-th layer, b (i) is the bias vector, σ is the activation function, usually the ReLU function; the number of neurons in the hidden layer is two layers; through this bidirectional data processing method, the Bi-LSTM network uses both historical and future information to enhance its learning ability for long sequence data. Its mathematical expression is:

[0046] σ(x) = max(0, x), in each hidden layer, the activation function maps the linear combination result to a nonlinear space;

[0047] The last fully connected layer outputs the final predicted value y MLP , based on all input features x at the current time step t The comprehensive output of the output layer is a linear layer, and its formula is as follows:

[0048] y MLP =W (n) h (n-1) +b (n) , where W (n) is the weight matrix of the output layer, h (n-1) is the output of the last hidden layer, b (n) is the bias term of the output layer; output yMLP is a single predicted value or a multi-dimensional forecast.

[0049] The Bi-LSTM network model is used for the longitudinal changes of the time series, and the MLP model is used for the lateral relationship of the current time step features. By fusing the outputs of these two models, the final prediction result is obtained by adjusting the longitudinal correction value y LSTM and the lateral correction value y MLP Perform weighted summation, the formula is as follows:

[0050] y fianl =λy LSTM +(1-λ)y LMP ; Among them, λ is the fusion weight, which indicates the relative importance of longitudinal correction and transverse correction in the final prediction; initially, λ is set to 0.5, that is, the longitudinal and transverse predictions each account for 50% weight.

[0051] In the step (5), the optimal fusion weight is found by continuously adjusting λ on the validation set. The specific steps are as follows: Initially, λ=0.5 is set, that is, the vertical and horizontal prediction contributions are equal; the prediction error of the model is calculated on the validation set, and the mean square error is used as the evaluation indicator. The formula is:

[0052] Among them, y i is the true value, is the predicted value after weighted fusion; according to the MSE performance on the validation set, adjust the value of λ and select the λ that minimizes MSE; use grid search or cross-validation to optimize λ; when λ is optimized, the model's prediction on the test set will use the final weighted fusion value y final , the dynamic information of the time series and the nonlinear relationship between the features are integrated to improve the overall accuracy and robustness of the prediction.

[0053] The advantages of the present invention are:

[0054] (1) Improving prediction accuracy: By combining the Bi-LSTM and MLP models, the present invention effectively improves the prediction accuracy of marine meteorological data. Bi-LSTM can make full use of the forward and backward information of the time series and capture the dynamic dependencies in the time series, thereby enhancing the model's understanding of historical and future information. The MLP model focuses on modeling nonlinear relationships between features and mining the complex interactions between features through multi-layer nonlinear transformations. Through the weighted fusion mechanism, the present invention combines the advantages of these two models, utilizing both the dynamic changes of the time series and the feature interactions of the current time step, thereby greatly improving the accuracy of the prediction results. Experimental results show that compared with the traditional single model, the hybrid model of the present invention performs well in complex marine meteorological disaster prediction tasks.

[0055] (2) Enhance the robustness and stability of the model: Through the synergy of Bi-LSTM and MLP, the robustness of the model in different data environments is enhanced. Bi-LSTM can handle abnormal fluctuations that may exist in time series, especially in dealing with instability and noise in marine meteorological data. MLP focuses on lateral correction between features and can perform detailed nonlinear feature optimization of the model, so that the model can find a balance between complex relationships between features at the current time step. Since these two models process data in different dimensions, the final fusion result has higher stability and noise resistance, ensuring the reliability of the model in practical applications.

[0056] (3) Improve the flexibility and wide application of the model: It is highly flexible and can be adaptively adjusted according to specific marine meteorological disaster prediction tasks. The structure of Bi-LSTM can be adjusted according to different data time spans, such as increasing or decreasing the number of LSTM layers or neurons to cope with time series data of different lengths. The MLP model can also adjust the depth of the hidden layer and the number of neurons according to the complexity of the features to ensure that the relationship between nonlinear features is fully captured. In addition, by adjusting the weight coefficients of weighted fusion, the model can find the best balance between longitudinal and lateral corrections, so that the model can achieve better prediction performance on different types of data sets. This flexible structure enables the model to be widely used in multiple fields such as marine meteorology, environmental monitoring, meteorological disaster warning, etc., and has good adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic diagram of the process of the present invention.

[0058] Figure 2 Schematic diagram of the LSTM network of the present invention.

[0059] Figure 3 It is the Bi-LSTM structure diagram of the present invention.

[0060] Figure 4 It is the MLP network structure diagram of the present invention.

[0061] Figure 5 This is the overall structure diagram of the Bi-LSTM and MLP horizontal and vertical correction of the present invention. DETAILED DESCRIPTION

[0062] The present invention will be further described below in conjunction with specific embodiments, and the advantages and features of the present invention will become clearer as the description proceeds. However, these embodiments are exemplary only and do not constitute any limitation to the scope of the present invention. It should be understood by those skilled in the art that the details and forms of the technical solution of the present invention may be modified or replaced without departing from the spirit and scope of the present invention, but these modifications and replacements all fall within the scope of protection of the present invention.

[0063] See also Figures 1 to 4 The present invention relates to a meteorological disaster prediction method based on AI, comprising the following steps:

[0064] (1) Data collection and preprocessing: Obtain historical marine meteorological data. First, obtain historical marine meteorological data to provide sufficient basic data for model training and prediction. These data include but are not limited to multi-dimensional marine meteorological characteristics such as wave height, wave direction, temperature, humidity, wind speed, etc., to ensure the comprehensiveness and representativeness of the data. In order to ensure the quality and reliability of the data, the collected raw data is preprocessed as necessary, and the data is cleaned, normalized, and timestamp corrected;

[0065] (2) Construction of time series data set: set a fixed time step and use the multi-dimensional feature data of the previous n days to predict the target value of the n+1 day;

[0066] (3) Model design and construction: Design and construct a deep learning model that combines a bidirectional long short-term memory network (Bi-LSTM) and a multi-layer perceptron (MLP). Bi-LSTM is used to extract dynamic features of time series, and MLP is used to perform nonlinear modeling of the relationship between features. The final prediction value is obtained by weighted fusion of the results of the two models.

[0067] (4) Model training, using error back propagation to update model parameters and adjust hyperparameters to optimize model performance;

[0068] (5) Model validation: Use the validation set to evaluate the model and adjust the model hyperparameters and weighting coefficients to optimize the prediction performance;

[0069] (6) Model application: applying the trained model to actual marine meteorological disaster prediction tasks.

[0070] The step (1) is specifically as follows:

[0071] 1.1 Handle missing values, outliers and noisy data; for missing values, use interpolation, mean filling or time series-based prediction methods to fill them; for outliers, use statistical analysis and threshold setting to identify and remove data points that deviate from the normal range;

[0072] 1.2 In order to eliminate the dimensional differences between different features and avoid numerical instability during model training, the feature data is normalized or standardized; commonly used methods include minimum-maximum normalization, scaling the data to the [0,1] interval, and Z-score normalization, so that the data has zero mean and unit variance.

[0073] 1.3 At the same time, in order to maintain the temporal continuity and accuracy of the data, the timestamps are corrected and unified to ensure the consistency of each feature data in the time dimension; any possible duplication or omission of timestamps is checked and supplemented to avoid data disorder caused by time dislocation.

[0074] 1.4 In addition, considering the periodic and seasonal characteristics of marine meteorological data, the data can be smoothed or the periodic components can be extracted to provide more representative input for the model.

[0075] After the above data collection and preprocessing steps, a marine meteorological data set with reliable quality and complete structure was obtained, which provided a solid foundation for subsequent model training and prediction, ensuring that the model can fully learn the inherent laws and characteristics in the data and improve the accuracy and stability of the prediction.

[0076] The step (2) is specifically as follows:

[0077] The time series data set construction steps include:

[0078] 2.1 Traverse the entire data set and generate several input and output pairs. The dimension of the input data is (three-dimensional tensor): number of samples × time step × number of features, and the output data is the corresponding target variable sequence. This method can not only capture the short-term change trend of the data, but also learn the long-term time dependency. It ensures that the time correspondence between the input sequence and the output sequence is accurate to avoid prediction errors caused by data misalignment.

[0079] 2.2 In order to enhance the generalization ability of the model, the data set is divided into a training set and a test set. The division can be done in chronological order to prevent future data from leaking into the model training; for example, the first 80% of the data is used as the training set, and the last 20% of the data is used as the test set. In order to prevent the model from overfitting, a part of the training set can be divided as a validation set for tuning the model hyperparameters and formulating the early stopping strategy;

[0080] 2.3 Convert data into tensor format.

[0081] The step (3) is specifically as follows:

[0082] 3.1 In the longitudinal prediction of the model, the preprocessed time series data is used to construct a Bi-LSTM model. The bidirectional characteristics of the Bi-LSTM network model are used to obtain the forward and backward correlation information of the time series data; the Bi-LSTM network model simultaneously captures the forward and backward features of the time series data, and uses the past and future information to mine the deep patterns in the data. Specifically, the input data is a multidimensional feature sequence divided by time steps. The Bi-LSTM network model encodes the sequence data through two LSTM layers, forward and backward, and generates hidden states for each time step. These hidden states contain the dynamic information of the time series at different time points;

[0083] 3.2 In the lateral prediction of the model, in order to capture the lateral relationship between different features, a multi-layer perceptron (MLP) model was constructed. The MLP model was used to perform lateral prediction of features and deeply explore the nonlinear relationship between features. The MLP model takes the multi-dimensional features of the current time step as input, and learns the complex nonlinear relationship between features through multiple layers of nonlinear transformations. The model includes an input layer, several hidden layers and an output layer. The hidden layer uses an activation function (such as ReLU or tanh) to enhance the expressive power of the model. The output of the MLP is the lateral prediction result, which supplements the feature relationship that the Bi-LSTM model may ignore.

[0084] MLP neural network is a type of feedforward neural network, which includes an input layer, at least one intermediate hidden layer and an output layer. It minimizes the prediction error of the model by adjusting the neuron weights, thereby realizing the training of the model and the prediction of the results. In actual use, the model maps a set of input vectors to a set of output vectors, and the different layers are fully connected, that is, any neuron in the previous layer is connected to all neurons in the next layer. Compared with the traditional BP (BackPropagation) neural network that usually uses the sigmoid function as the activation function, MLP can flexibly use different activation functions to better handle multi-parameter nonlinear relationships. This article uses an MLP neural network with a two-hidden layer structure.

[0085] MLP is used to perform horizontal correction on features, deeply explore the nonlinear relationship between features, and enhance the generalization ability of the model. Through vertical time series feature extraction and horizontal feature optimization, high-precision prediction of marine meteorological data is achieved to meet the needs of marine environmental monitoring, shipping safety, fishery management and other fields for accurate weather forecasts, and improve the quality and efficiency of marine meteorological services.

[0086] The step (4) is specifically as follows: the model training includes two parts: Bi-LSTM model and MLP model; Bi-LSTM is used to process time series data, and MLP is used to process the nonlinear relationship of the current time step feature; the training steps are as follows:

[0087] 4.1 In Bi-LSTM, time series data is input into the forward and backward LSTM layers to capture the bidirectional dependencies in the time series; the output of Bi-LSTM passes through the fully connected layer to generate longitudinal prediction results;

[0088] (2) MLP processes the features of the current time step. Through a multi-layer fully connected network, MLP extracts nonlinear features layer by layer and finally generates a lateral correction result.

[0089] (3) Error back propagation to update each weight;

[0090] (4) Adjust the model hyperparameters and repeat the above steps, compare the verification value and error value, select the optimal model, and complete the model training.

[0091] In the above step 3.1, the LSTM network has three gates: forget gate, input gate and output gate. There are two activation functions in LSTM: σ and Tanh. Among them, σ is the Sigmoid activation function, which will be used in the forget gate, input gate and output gate, and σ is used to output [0, 1]. When generating candidate memories, the Tanh activation function is used to output [-1, 1]. The forget gate f t : Used to determine the cell state C from the previous moment t-1 The amount of information that should be discarded and retained until the current time C t Information, f t =σ(W f [h t-1 , x t ]+b f );W f is the weight matrix of the forget gate; h t-1 is the output of the hidden layer at the previous moment; x t b is the input data at time t; f is the bias vector of the forget gate; input gate i t : Used to determine the input information x t Save to cell state C t The degree of t =σ(W i [h t-1 , x t ]+b i );W i is the weight matrix of the input gate; b iis the bias vector of the input gate; the candidate state created by the tanh layer for: Where: w c is the weight matrix of the current state; b is the bias vector of the current state;

[0092] Unit information update: used to calculate the unit state c at the current moment t ,

[0093]

[0094] Output gate: O t Used to control the status of the unit C t How much information is input to the current hidden layer output value h; o t =σ(W o [h t-1 , x t ]+b o );h t =o t ·tanh(c t );

[0095] Where: b o is the bias term of the output gate; W o is the weight matrix of the output gate; tanh is the hyperbolic tangent activation function; tanh and σ are defined as:

[0096]

[0097] The Bi-LSTM network contains two LSTM units: Bi-LSTM can use both forward and backward time information, and can better model time series dependencies. One LSTM unit processes input data in the forward order of the time series; the other LSTM unit processes input data in the reverse order of the time series. The BiLSTM network structure is as follows: Figure 2 As shown. The forward LSTM and backward LSTM of Bi-LSTM calculate the hidden state of the time series respectively, and the formula is as follows:

[0098]

[0099] In the formula, h t fw and h t bw Represent the forward and backward hidden states respectively, x t are input features, and is the hidden state of the previous time step and the next time step; the final hidden state is obtained by concatenating the forward and backward hidden states:

[0100] In step 3.3, the MLP input layer receives the multidimensional feature vector x at the current time step. t , whose dimension is d input , which is the number of features. For example, for marine meteorological data, the input features may include wave height (Hs), maximum wave height (Hmax), period (Tz), sea surface temperature (SST), etc. Input feature vector x t =[x1, x2, ..., x dinput ] is passed to the hidden layer through the first fully connected layer; each hidden layer processes the output of the previous layer through linear transformation and activation function; assuming there are N hidden layers, the output of the i-th hidden layer is expressed as:

[0101] h (i) =σ(W (i) h (i-1) +b (i) ), where W (i) is the weight matrix of the i-th layer, b (i) is the bias vector, σ is the activation function, usually the ReLU function; the number of neurons in the hidden layer is two layers; through this bidirectional data processing method, the Bi-LSTM network uses both historical and future information to enhance its learning ability for long sequence data. Its mathematical expression is:

[0102] σ(x) = max(0, x). In each hidden layer, the activation function maps the linear combination results to a nonlinear space, allowing the model to learn more complex feature relationships.

[0103] The last fully connected layer outputs the final predicted value y MLP , based on all input features x at the current time step t The comprehensive output of the output layer is usually a linear layer, and its formula is as follows:

[0104] y MLP =W (n) h (n-1) +b (n) , where W (m) is the weight matrix of the output layer, h (n-1) is the output of the last hidden layer, b (n) is the bias term of the output layer; output y MLP It is a single prediction value or a multi-dimensional prediction result, depending on the target task.

[0105] MLP plays a role of horizontal correction, focusing on the nonlinear relationship between the features at the current time step. Compared with the longitudinal time dependency modeling of the Bi-LSTM model, MLP starts more from the interaction between features, supplementing the horizontal relationship that Bi-LSTM may miss. Through the combination of Bi-LSTM and MLP, the model can simultaneously capture the dynamic characteristics of the time dimension and the complex relationship between features, thereby improving the overall accuracy of the prediction.

[0106] The Bi-LSTM network model is used for the longitudinal changes of the time series, and the MLP model is used for the lateral relationship of the current time step features. By fusing the outputs of these two models, the advantages of both can be comprehensively utilized. The final prediction result is obtained by LSTM and the lateral correction value y MLP Perform weighted summation, the formula is as follows:

[0107] y final =λy LSTM +(1-λ)y MLP ; Among them, λ is the fusion weight, which indicates the relative importance of longitudinal correction and transverse correction in the final prediction; initially, λ is set to 0.5, that is, the longitudinal and transverse predictions each account for 50% weight.

[0108] In the step (5), the optimal fusion weight is found by continuously adjusting λ on the validation set. The specific steps are as follows: Initially, λ=0.5 is set, that is, the vertical and horizontal prediction contributions are equal; the prediction error of the model is calculated on the validation set, and the mean square error (MSE) is used as the evaluation indicator. The formula is:

[0109] Among them, y i is the true value, is the predicted value after weighted fusion; according to the MSE performance on the validation set, adjust the value of λ and select the λ that minimizes MSE; use grid search or cross-validation to optimize λ; when λ is optimized, the model's prediction on the test set will use the final weighted fusion value y final ,This fusion strategy can effectively integrate the dynamic information of time series,and the nonlinear relationship between features, and improve the overall,accuracy and robustness of the prediction.

[0110] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A meteorological disaster prediction method based on AI, characterized in that: The following steps are involved: (1) Data collection and preprocessing: Obtain historical marine meteorological data, clean and normalize the data, and perform time stamp correction; (2) Construction of time series data set: set a fixed time step and use the multi-dimensional feature data of the previous n days to predict the target value of the n+1 day; (3) Model design and construction: Design and construct a deep learning model that combines a bidirectional long short-term memory network (Bi-LSTM) and a multi-layer perceptron (MLP). Bi-LSTM is used to extract dynamic features of time series, and MLP is used to perform nonlinear modeling of the relationship between features. The final prediction value is obtained by weighted fusion of the results of the two models. (4) Model training, using error back propagation to update model parameters and adjust hyperparameters to optimize model performance; (5) Model validation: Use the validation set to evaluate the model and adjust the model hyperparameters and weighting coefficients to optimize the prediction performance; (6) Model application: applying the trained model to actual marine meteorological disaster prediction tasks.

2. The AI-based meteorological disaster prediction method according to claim 1, characterized in that: The step (1) is specifically as follows: 1.1 Handle missing values, outliers and noisy data; for missing values, use interpolation, mean filling or time series-based prediction methods to fill them; for outliers, use statistical analysis and threshold setting to identify and remove data points that deviate from the normal range; 1.2 Normalize or standardize each feature data; 1.3 Correct and unify the timestamps to ensure the consistency of each feature data in the time dimension; 1.4 Smooth the data or extract the periodic components.

3. The AI-based meteorological disaster prediction method according to claim 1 or 2, characterized in that: The step (2) is specifically as follows: The time series data set construction steps include: 2.1 Traverse the entire data set and generate several input and output pairs, ensuring that the time correspondence between the input sequence and the output sequence is accurate to avoid prediction errors caused by data misalignment; 2.2 Divide the data set into training set and test set. The division can be done in chronological order to avoid future data leakage into model training; 2.3 Convert data into tensor format.

4. The AI-based meteorological disaster prediction method according to claim 1 or 2, characterized in that: The step (3) is specifically as follows: 3.1 The bidirectional characteristics of the Bi-LSTM network model are used to obtain the forward and backward correlation information of the time series data; the Bi-LSTM network model captures the forward and backward features of the time series data at the same time, and uses the past and future information to mine the deep patterns in the data. Specifically, the input data is a multidimensional feature sequence divided by time steps. The Bi-LSTM network model encodes the sequence data through the forward and backward LSTM layers to generate the hidden state of each time step. These hidden states contain the dynamic information of the time series at different time points; 3.2 The MLP model is used to perform horizontal prediction of features and deeply explore the nonlinear relationship between features; the MLP model takes the multi-dimensional features of the current time step as input, and learns the complex nonlinear relationship between features through multiple layers of nonlinear transformation; the model includes an input layer, several hidden layers and an output layer, and the hidden layer uses an activation function to enhance the expressive power of the model.

5. The AI-based meteorological disaster prediction method according to claim 4, characterized in that: The step (4) is specifically as follows: the model training includes two parts: Bi-LSTM model and MLP model; Bi-LSTM is used to process time series data, and MLP is used to process the nonlinear relationship of the current time step feature; The training steps are as follows: 4.1 In Bi-LSTM, time series data is input into the forward and backward LSTM layers to capture the bidirectional dependencies in the time series; The output of Bi-LSTM passes through the fully connected layer to generate the vertical prediction results; (2) MLP processes the features of the current time step. Through a multi-layer fully connected network, MLP extracts nonlinear features layer by layer and finally generates a lateral correction result. (3) Error back propagation to update each weight; (4) Adjust the model hyperparameters and repeat the above steps, compare the verification value and error value, select the optimal model, and complete the model training.

6. The AI-based meteorological disaster prediction method according to claim 4, characterized in that: In the above step 3.1, the LSTM network has three gates: forget gate, input gate and output gate. There are two activation functions in LSTM: σ and Tanh. Among them, σ is the Sigmoid activation function, which will be used in the forget gate, input gate and output gate, and σ is used to output [0, 1]. When generating candidate memories, the Tanh activation function is used to output [-1, 1]. The forget gate f t : Used to determine the cell state C from the previous moment t-1 The amount of information that should be discarded and retained until the current time C t Information, f t =σ(W f [h t-1 , x t ]+b f );W f is the weight matrix of the forget gate; h t-1 is the output of the hidden layer at the previous moment; x t b is the input data at time t; f is the bias vector of the forget gate; input gate i t : Used to determine the input information x t Save to cell state C t The degree of t =σ(W i [h t-1 , x t ]+b i );W i is the weight matrix of the input gate; b i is the bias vector of the input gate; the candidate state created by the tanh layer for: Where: w c is the weight matrix of the current state; b is the bias vector of the current state; Unit information update: used to calculate the unit state c at the current moment t , Output gate: O t Used to control the status of the unit C t How much information is input to the current hidden layer output value h; o t =σ(W o [h t-1 , x t ]+b o );h t =o t ·tanh(c t ); Where: b o is the bias term of the output gate; W o is the weight matrix of the output gate; tanh is the hyperbolic tangent activation function; tanh and σ are defined as:

7. The AI-based meteorological disaster prediction method according to claim 5, characterized in that: The Bi-LSTM network comprises two LSTM units; one LSTM unit processes input data in the forward order of the time series; the other LSTM unit processes input data in the reverse order of the time series; the forward LSTM and backward LSTM of the Bi-LSTM respectively calculate the hidden state of the time series, and the formula is as follows: h t bw =LSTM bw (x t ,h t+1 bw ); In the formula, h t fw and h t bw Represent the forward and backward hidden states respectively, x t are input features, and is the hidden state of the previous time step and the next time step; the final hidden state is obtained by concatenating the forward and backward hidden states:

8. The AI-based meteorological disaster prediction method according to claim 6, characterized in that: In step 3.3, the MLP input layer receives the multidimensional feature vector x at the current time step. t , whose dimension is d input , that is, the number of features; input feature vector x t =[x1,x2,...,x dinput ] is passed to the hidden layer through the first fully connected layer; each hidden layer processes the output of the previous layer through linear transformation and activation function; assuming there are N hidden layers, the output of the i-th hidden layer is expressed as: h (i) =σ(W (i) h (i-1) +b (i) ), where W (i) is the weight matrix of the i-th layer, b (i) is the bias vector, σ is the activation function, usually the ReLU function; the number of neurons in the hidden layer is two layers; through this bidirectional data processing method, the Bi-LSTM network uses both historical and future information to enhance its learning ability for long sequence data. Its mathematical expression is: σ(x) = max(0,x). In each hidden layer, the activation function maps the linear combination result to a nonlinear space; the last fully connected layer outputs the final predicted value y MLP , based on all input features x at the current time step t The comprehensive output of the output layer is a linear layer, and its formula is as follows: y MLP =W (n) h (n-1) +b (n) , where W (n) is the weight matrix of the output layer, h (n-1) is the output of the last hidden layer, b (n) is the bias term of the output layer; output y MLP is a single predicted value or a multi-dimensional forecast.

9. The AI-based meteorological disaster prediction method according to claim 7, characterized in that: The Bi-LSTM network model is used for the longitudinal changes of the time series, and the MLP model is used for the lateral relationship of the current time step features. By fusing the outputs of these two models, the final prediction result is obtained by adjusting the longitudinal correction value y LSTM and the lateral correction value y MLP Perform weighted summation, the formula is as follows: y final =λy LSTM +(1-λ)y MLP ; Among them, λ is the fusion weight, which indicates the relative importance of longitudinal correction and transverse correction in the final prediction; initially, λ is set to 0.5, that is, the longitudinal and transverse predictions each account for 50% weight.

10. The AI-based meteorological disaster prediction method according to claim 9, characterized in that: In the step (5), the optimal fusion weight is found by continuously adjusting λ on the validation set. The specific steps are as follows: Initially, λ=0.5 is set, that is, the vertical and horizontal prediction contributions are equal; the prediction error of the model is calculated on the validation set, and the mean square error is used as the evaluation indicator. The formula is: Among them, y i is the true value, is the predicted value after weighted fusion; according to the MSE performance on the validation set, adjust the value of λ and select the λ that minimizes MSE; use grid search or cross-validation to optimize λ; when λ is optimized, the model's prediction on the test set will use the final weighted fusion value y final , the dynamic information of the time series and the nonlinear relationship between the features are integrated to improve the overall accuracy and robustness of the prediction.

Images