Meteorological temperature prediction method and system, storage medium and equipment
By combining bidirectional LSTM and attention mechanism meteorological temperature prediction methods, the problem of difficult traditional methods to deal with complex nonlinearity and long-term dependencies is solved, achieving higher prediction accuracy and robustness.
Patent Information
- Application Number
- CN202411610574.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-12
- Publication Date
- 2025-03-04
AI Technical Summary
Traditional meteorological temperature prediction methods, such as linear regression models and autoregressive moving average models, are difficult to effectively deal with complex nonlinear and long-term dependencies, resulting in insufficient prediction accuracy and reliability.
The bidirectional long and short-term memory network (LSTM) and attention mechanism are used to process time series data through forward and reverse LSTM networks, generate a bidirectional hidden state sequence, and weighted sum through attention mechanism to generate context vectors, and finally generate a temperature prediction value through the full connection layer.
It significantly improves the accuracy and robustness of meteorological temperature prediction, can better capture long-term and short-term dependencies in complex temperature time series data, and enhances the processing power and prediction performance of the model.
Smart Images
Figure CN119557849A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of meteorological forecasting, and in particular to a meteorological temperature forecasting method, system, storage medium and equipment. Background Art
[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.
[0003] Temperature prediction is required in the field of meteorology. Traditional methods generally use linear regression models and autoregressive moving average models. Although they can obtain predicted values, they often perform poorly when processing temperature time series data, faced with nonlinear, complex time series patterns, and long-term dependent temperature data.
[0004] The reason is that the linear regression model is too simple and cannot capture the nonlinear relationship in temperature change; while the autoregressive moving average (ARMA) model can handle a certain amount of autocorrelation, but its ability to handle data dependencies over a long time span is limited. As the complexity of meteorological data increases, the prediction accuracy and reliability of these traditional methods will be limited.
[0005] Deep learning technology has relatively superior sequence modeling capabilities, such as the long short-term memory network (LSTM), which can handle long-term dependencies in long time series and significantly improve the accuracy of meteorological temperature forecasts. However, when processing complex time series data, LSTM still has deficiencies in computational efficiency and model interpretability. Summary of the invention
[0006] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a meteorological temperature prediction method, system, storage medium and device, so that the temperature prediction process is combined with a bidirectional LSTM and an attention mechanism, which not only enhances the model's processing capability for complex temperature time series data, but also improves the accuracy and robustness of the prediction, thereby addressing the shortcomings of traditional methods.
[0007] In order to achieve the above object, the present invention adopts the following technical solution:
[0008] A first aspect of the present invention provides a meteorological temperature prediction method, comprising the following steps:
[0009] Obtain and pre-process historical temperature data and meteorological data that affect temperature changes;
[0010] The preprocessed data is processed by the forward and reverse LSTM networks respectively to obtain the forward and backward hidden state sequences of the time series, and the bidirectional hidden state sequence is obtained by concatenation or element-by-element addition;
[0011] The bidirectional hidden state is mapped to the attention score through the weight matrix, and the attention weight of each time step is obtained after normalization. The obtained attention weight is used to weight the sum of the bidirectional hidden state sequence, and the obtained context vector is processed by the fully connected layer to generate the temperature prediction value.
[0012] Furthermore, preprocessing includes normalizing and time window segmentation of the original data to generate input sequences and target sequences, using the data of the past time window as input and the future temperature as the prediction target.
[0013] Furthermore, the preprocessed data is converted into time series data, and the forward LSTM network is used to process the forward information of the time series data to generate the forward hidden state sequence H f , use the reverse LST M network to process the reverse information of the time series data and generate the reverse hidden state sequence H b ; The forward and reverse hidden state sequences are concatenated or element-by-element summed in the feature dimension to generate a bidirectional hidden state sequence H bi , where H bi =[H f ,H b ].
[0014] Furthermore, the bidirectional hidden state is mapped to the attention score through the weight matrix, as shown in the following formula:
[0015] e i =tanh(H bi ,i·W att +b att );
[0016] Among them, e i is the score at time step i, W att is the attention weight matrix, b att is the bias term.
[0017] Furthermore, the attention weight of each time step is obtained by normalization, specifically: normalization processing e by Softmax function i , get the attention weight α for each time step i i .
[0018] Furthermore, the obtained attention weights are used to weight the sum of the bidirectional hidden state sequence to obtain the context vector, as shown in the following formula:
[0019]
[0020] Among them, C is the context vector, H bi is a bidirectional hidden state sequence, α i is the attention weight for each time step i.
[0021] Furthermore, the obtained context vector is processed by the fully connected layer to generate a temperature prediction value, as shown in the following formula:
[0022]
[0023] Among them, C is the context vector, W out and b out are the weights and biases of the fully connected layers.
[0024] A second aspect of the present invention provides a meteorological temperature prediction system, comprising.
[0025] The preprocessing module is configured to: obtain and preprocess historical temperature data and meteorological data affecting temperature changes;
[0026] The bidirectional LSTM module is configured as follows: the preprocessed data is processed by the forward and reverse LSTM networks respectively to obtain the forward and backward hidden state sequences of the time series, and the bidirectional hidden state sequence is obtained by concatenation or element-by-element addition;
[0027] The attention module is configured to: map the bidirectional hidden state to the attention score through the weight matrix, obtain the attention weight of each time step through normalization, and use the obtained attention weight to weighted sum the bidirectional hidden state sequence to obtain the context vector;
[0028] The prediction output module is configured as follows: the obtained context vector is processed by a fully connected layer to generate a temperature prediction value.
[0029] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps in the above-mentioned meteorological temperature prediction method when executed by a processor.
[0030] A fourth aspect of the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in the above-mentioned meteorological temperature prediction method when executing the program.
[0031] Compared with the prior art, one or more of the above technical solutions have the following beneficial effects:
[0032] 1. The prediction process combines bidirectional LSTM and attention mechanism. Bidirectional LSTM captures the forward and backward dependencies of each time point in the sequence at the same time, outputs the hidden state sequence for further processing, and uses the attention mechanism to determine which time steps are more important for prediction. For example, when predicting the temperature of a certain day, the model will pay more attention to the temperature changes in the previous few days, the current humidity and wind speed, and the seasonal characteristics of the date. This dynamic focusing capability enables the model to make more accurate predictions when processing complex meteorological data, which not only enhances the model's ability to handle complex temperature time series data, but also improves the accuracy and robustness of the prediction, thereby addressing the shortcomings of traditional methods. This approach enables the trained model to fully understand the long-term and short-term dependencies in the temperature data, thereby improving the prediction accuracy.
[0033] 2. The introduced attention score can help the model automatically focus on the most important part of the input sequence for the current prediction task, so that the model can dynamically adjust the degree of attention to different time steps when processing long sequences, thereby better capturing key time dependencies and improving prediction performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.
[0035] Figure 1 is a schematic diagram of the architecture of a meteorological temperature prediction model provided by one or more embodiments of the present invention;
[0036] Figure 2 It is a schematic diagram of the training process of the meteorological temperature prediction model provided by one or more embodiments of the present invention. DETAILED DESCRIPTION
[0037] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0038] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.
[0039] Terminology explanation:
[0040] Interpretability refers to the ability to understand how a model makes decisions. In computer science, interpretability means that the model needs to provide clear and explicit information about its decision-making process so that the model's users (humans) can understand and explain the model's output.
[0041] Autocorrelation, also known as serial correlation or spatial correlation, is a statistic that describes the dependence between the values of a signal or data sequence at different time points (or spatial points). In simple terms, it is the correlation between the data and itself, that is, measuring the correlation between a random variable and itself at different time points (or spatial points).
[0042] As introduced in the background technology, deep learning technology has relatively superior sequence modeling capabilities. For example, the long short-term memory network (LSTM) can handle long-term dependencies in long time series and significantly improve the accuracy of meteorological temperature forecasts. However, when processing complex time series data, LSTM still has deficiencies in computational efficiency and model interpretability.
[0043] Therefore, the following embodiments propose a meteorological temperature prediction method, system, storage medium and device, so that the temperature prediction process combines the bidirectional LSTM and attention mechanism, which not only enhances the model's processing ability for complex temperature time series data, but also improves the accuracy and robustness of the prediction, thereby addressing the shortcomings of traditional methods.
[0044] Embodiment 1:
[0045] The meteorological temperature prediction method comprises the following steps:
[0046] Obtain and pre-process historical temperature data and meteorological data that affect temperature changes;
[0047] The preprocessed data is processed by the forward and reverse LSTM networks respectively to obtain the forward and backward hidden state sequences of the time series, and the bidirectional hidden state sequence is obtained by concatenation or element-by-element addition;
[0048] The bidirectional hidden state is mapped to the attention score through the weight matrix, and the attention weight of each time step is obtained after normalization. The obtained attention weight is used to weight the sum of the bidirectional hidden state sequence, and the obtained context vector is processed by the fully connected layer to generate the temperature prediction value.
[0049] The meteorological temperature prediction method given in this embodiment is based on bidirectional LSTM and attention mechanism. The key lies in obtaining bidirectional information of the time series through the bidirectional LSTM layer and dynamically weighting the influence of each time step through the attention mechanism to improve the prediction accuracy.
[0050] like Figure 1 As shown, the raw data is obtained and preprocessed, which occurs in the input layer.
[0051] Obtain historical temperature data and related meteorological variables that affect temperature changes, such as humidity, wind speed, and air pressure, from meteorological sensors, weather stations, or online meteorological databases. In order to ensure the comprehensiveness of the data, obtain data for at least several months to cover the temperature changes in different seasons and climate conditions. The acquired data is saved in groups, that is, the temperature data at each moment corresponds to the relevant meteorological variables that affect temperature changes at the current moment, for example, the temperature at moment t is x, the corresponding humidity is a, the wind speed is b, the air pressure is c, etc.
[0052] Preprocess the raw data, specifically:
[0053] Min-Max normalization is performed on temperature data and other meteorological variables to scale the data to the range of [0,1];
[0054] Split the data into fixed-length time windows, for example, using the data from the past 24 hours to predict the temperature in the next hour;
[0055] Generate input sequence and target sequence, use data from past time windows as model input, and future temperature as prediction target;
[0056] The data is divided into training set (70%), validation set (15%) and test set (15%) to ensure the generalization ability of the model on unseen data.
[0057] like Figure 1 As shown in Figure 1, the preprocessed data is simultaneously input into the bidirectional LSTM layer. Among them, the forward LSTM processes the forward information of the time series data and generates a forward hidden state sequence H f The reverse LSTM processes the reverse information of the time series data and generates a reverse hidden state sequence H b The forward and reverse hidden state sequences are concatenated or element-wise summed in the feature dimension to generate a bidirectional hidden state sequence H bi , where H bi =[H f ,H b ].
[0058] In this embodiment, the hidden state sequence is the complex pattern and trend of the input data, including rich information such as season and time trend.
[0059] In the bidirectional LSTM layer, the data will be processed by both the forward and backward LSTM networks to capture the forward and backward context information of the time series. Bidirectional LSTM can capture long-term dependencies and improve the modeling ability of time series.
[0060] Introduce the attention mechanism. Figure 1 As shown, the attention score is calculated by introducing the weight matrix W attThe hidden state of the bidirectional LSTM is mapped to the attention score as shown in the following formula:
[0061] e i =tanh(H bi ,i·W att +b att );
[0062] where e i is the score at time step i, W att is the attention weight matrix, b att is the bias term.
[0063] The bidirectional LSTM simultaneously captures the forward and backward dependencies at each time point in the sequence, outputs the hidden state sequence for further processing, and uses the attention mechanism to determine which time steps are more important for prediction. For example, when predicting the temperature of a certain day, the model may pay more attention to the temperature changes in the previous few days, the current humidity and wind speed, and the seasonal characteristics of that date. This dynamic focusing ability enables the model to make more accurate predictions when processing complex meteorological data.
[0064] The attention weights for each time step are obtained by normalizing with the Softmax function. The Softmax function ensures that the sum of all weights is 1, so that it can be used as a weight to weight the input information, as shown in the following formula:
[0065] α i =softmax(e i );
[0066] where α i is the attention weight at time step i.
[0067] Use these weights to perform a weighted summation of the hidden states of the bidirectional LSTM to generate a context vector C, which represents a weighted synthesis of the information that the model focuses on, as shown in the following formula:
[0068]
[0069] In this embodiment, the information that the model focuses on includes, but is not limited to, historical temperature data, influencing factors, time characteristics, and geographic information.
[0070] like Figure 1 As shown, the obtained context vector is input into the fully connected layer for processing to generate the final temperature prediction value, as shown in the following formula:
[0071]
[0072] Where W out and b out are the weights and biases of the fully connected layers.
[0073] The above process introduces an attention mechanism to calculate the importance weight of each time step based on the output of the bidirectional LSTM. The attention weight is applied to the hidden state of the bidirectional LSTM through weighted summation to generate a context vector. The context vector is input into the fully connected layer to generate the temperature prediction value at the future time point. This method enables the trained model to fully understand the long-term and short-term dependencies in the temperature data, thereby improving the prediction accuracy. The main function of the attention score is to help the model automatically focus on the most important part of the input sequence for the current prediction task, so that when processing long sequences, the model can dynamically adjust the degree of attention to different time steps, so as to better capture the key time dependencies and improve the prediction performance.
[0074] There is dependency in the process of meteorological temperature prediction. Since temperature data has a strong time series characteristic, there is a correlation between temperature values at similar time points. At the same time, temperature changes are also affected by other meteorological factors, such as humidity, wind speed, air pressure, etc. There is a complex interdependence between these factors and temperature. Temperature changes in different geographical locations may be affected by the surrounding environment, such as terrain, vegetation cover, and urbanization level.
[0075] The use of the attention mechanism further enhances the model's ability to capture key information, making the model more robust in the face of complex and changing meteorological conditions. The model combining bidirectional LSTM and attention mechanism can better adapt to different meteorological conditions and data characteristics. The attention mechanism can dynamically adjust the degree of attention to different time steps through attention weights.
[0076] The diversity of the dataset can also be increased by adding noise, interpolating time series, or synthesizing new data points, which can help the model better generalize and capture the complexity of temperature changes.
[0077] Figure 2 The model training process is shown, including the calculation of the loss function, the update of the optimizer, and the monitoring of the training process. During the model training process, the mean square error (MSE) is used as the loss function to measure the difference between the predicted value and the actual value, as shown in the following formula:
[0078]
[0079] in is the predicted value, y iis the actual value, and N is the number of samples. In terms of optimizer, Adam optimizer is used to optimize model parameters to improve training efficiency. An early stopping strategy is used during training. Training is stopped when the loss on the validation set no longer decreases significantly to prevent overfitting. Hyperparameters such as learning rate, number of LSTM units, and parameters of the attention mechanism are adjusted through the validation set.
[0080] In model evaluation, indicators such as mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE) are used to evaluate the prediction performance of the model on the test set.
[0081] Apply the trained model to new time series data to generate temperature forecasts for future time points. It can be applied to practical scenarios such as weather forecasting systems, climate monitoring, and environmental protection to provide accurate temperature forecast information for decision-making and planning. The combination of bidirectional LSTM and attention mechanism significantly improves the accuracy and robustness of temperature forecasting.
[0082] Consider that the model may perform well on the training data, but lack generalization ability in actual applications. Prevent overfitting by introducing regularization techniques such as L1 or L2 regularization; use cross-validation and early stopping strategies.
[0083] A model that performs well on the training data but does not generalize well in real applications is usually because the model overfits the training data. Overfitting means that the model learns the noise and details in the training data, not just the underlying data distribution. This causes the model to perform poorly on new, unseen data.
[0084] Specific implementation steps:
[0085] Data preprocessing stage: Use K-fold cross validation to divide the data set.
[0086] Model definition stage: Add regularization parameters when defining the LSTM layer and fully connected layer of the model.
[0087] Model training phase: Use the EarlyStopping callback function to monitor the loss of the validation set and stop training when the validation set loss no longer improves.
[0088] In addition, considering that the linear regression model of the existing technology is too simple to capture the nonlinear relationship in temperature changes. Although the autoregressive moving average (ARMA) model can handle certain autocorrelations, it has limited ability to handle data dependencies over long time spans. Although the LSTM in the deep model can handle long-term dependencies in long time series, it is insufficient in computational efficiency and model interpretability when processing complex time series data.
[0089] In addition to historical temperature data and basic meteorological data (such as humidity, wind speed, and air pressure), this solution can consider adding more types of data, such as satellite remote sensing data, geographic information system (GIS) data, etc., to provide more comprehensive input features, but the impact of abnormal data needs to be considered. By adding outlier detection and processing steps in the data preprocessing stage, the impact of abnormal data on model prediction can be reduced. Cross-regional model calibration also needs to be considered: the model should be calibrated for different geographical locations and climate conditions to improve the prediction accuracy of different regions.
[0090] The above process combines the bidirectional LSTM and attention mechanism in the temperature prediction process, which not only enhances the model's ability to handle complex temperature time series data, but also improves the accuracy and robustness of predictions, thereby addressing the shortcomings of traditional methods. Bidirectional LSTM can capture both the positive and negative dependencies of time series data, thereby indirectly improving computational efficiency; the attention mechanism enables the model to focus on the most important time steps for the prediction task, which can reduce the amount of information that the model needs to process, thereby improving computational efficiency. The attention weight matrix provides the information that the model pays attention to when making predictions, which helps to improve the interpretability of the model.
[0091] Embodiment 2:
[0092] Meteorological temperature forecast system, including:
[0093] The preprocessing module is configured to: obtain and preprocess historical temperature data and meteorological data affecting temperature changes;
[0094] The bidirectional LSTM module is configured as follows: the preprocessed data is processed by the forward and reverse LSTM networks respectively to obtain the forward and backward hidden state sequences of the time series, and the bidirectional hidden state sequence is obtained by concatenation or element-by-element addition;
[0095] The attention module is configured to: map the bidirectional hidden state to the attention score through the weight matrix, obtain the attention weight of each time step through normalization, and use the obtained attention weight to weighted sum the bidirectional hidden state sequence to obtain the context vector;
[0096] The prediction output module is configured as follows: the obtained context vector is processed by a fully connected layer to generate a temperature prediction value.
[0097] The temperature prediction process combines bidirectional LSTM and attention mechanism, which not only enhances the model's ability to process complex temperature time series data, but also improves the accuracy and robustness of the prediction, thereby addressing the shortcomings of traditional methods.
[0098] Embodiment three:
[0099] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the meteorological temperature prediction method described in the above-mentioned embodiment 2 are implemented.
[0100] Embodiment 4:
[0101] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the meteorological temperature prediction method described in the above-mentioned embodiment 2 are implemented.
[0102] The steps involved in the above embodiments 2 to 4 correspond to those in embodiment 1. For the specific implementation, please refer to the relevant description of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.
[0103] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A meteorological temperature prediction method, characterized in that: The following steps are involved: Obtain and pre-process historical temperature data and meteorological data that affect temperature changes; The preprocessed data is processed by the forward and reverse LSTM networks respectively to obtain the forward and backward hidden state sequences of the time series, and the bidirectional hidden state sequence is obtained by concatenation or element-by-element addition; The bidirectional hidden state is mapped to the attention score through the weight matrix, and the attention weight of each time step is obtained after normalization. The obtained attention weight is used to weight the sum of the bidirectional hidden state sequence, and the obtained context vector is processed by the fully connected layer to generate the temperature prediction value.
2. The meteorological temperature prediction method according to claim 1, characterized in that: Preprocessing, specifically: the original data is normalized and time-windowed to generate input sequences and target sequences, with the data in the past time window as input and the future temperature as the prediction target.
3. The meteorological temperature prediction method according to claim 1, characterized in that: The preprocessed data forms time series data, and the forward LSTM network is used to process the forward information of the time series data to generate the forward hidden state sequence H f , use the reverse LSTM network to process the reverse information of the time series data and generate the reverse hidden state sequence H b ; The forward and reverse hidden state sequences are concatenated or element-by-element summed in the feature dimension to generate a bidirectional hidden state sequence H bi , where H bi =[H f ,H b ].
4. The meteorological temperature prediction method according to claim 1, characterized in that: The bidirectional hidden state is mapped to the attention score through the weight matrix as shown below: e i =tanh(H bi ,i·W att +b att ); Among them, e i is the score at time step i, W att is the attention weight matrix, b att is the bias term.
5. The meteorological temperature prediction method according to claim 1, characterized in that: The attention weight of each time step is obtained by normalization, specifically: normalization through the Softmax function e i , get the attention weight α for each time step i i .
6. The meteorological temperature prediction method according to claim 1, characterized in that: The obtained attention weights are used to weight the bidirectional hidden state sequence to obtain the context vector, as shown in the following formula: Among them, C is the context vector, H bi is a bidirectional hidden state sequence, α i is the attention weight for each time step i.
7. The meteorological temperature prediction method according to claim 1, characterized in that: The obtained context vector is processed by the fully connected layer to generate the temperature prediction value, as shown in the following formula: Among them, C is the context vector, W out and b out are the weights and biases of the fully connected layers.
8. A meteorological temperature prediction system, characterized in that: include: The preprocessing module is configured to: obtain and preprocess historical temperature data and meteorological data affecting temperature changes; The bidirectional LSTM module is configured as follows: the preprocessed data is processed by the forward and reverse LSTM networks respectively to obtain the forward and backward hidden state sequences of the time series, and the bidirectional hidden state sequence is obtained by concatenation or element-by-element addition; The attention module is configured to: map the bidirectional hidden state to the attention score through the weight matrix, obtain the attention weight of each time step through normalization, and use the obtained attention weight to weighted sum the bidirectional hidden state sequence to obtain the context vector; The prediction output module is configured as follows: the obtained context vector is processed by a fully connected layer to generate a temperature prediction value.
9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the steps in the meteorological temperature prediction method as described in any one of claims 1 to 7 are implemented.
10. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the meteorological temperature prediction method as described in any one of claims 1 to 7 are implemented.