Small-scale heavy fog weather prediction method, system and device based on deep learning and medium

Through a deep learning-based method, the model parameters are updated using LSTM regression network and real-time data, the problem of traditional mathematical models decreasing prediction accuracy when data distribution changes is solved, and high real-time and accuracy of foggy weather prediction is achieved.

CN119065030AInactive Publication Date: 2024-12-03HUNAN UNIV
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411105678.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-13
Publication Date
2024-12-03
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Traditional mathematical models lead to a sharp decline in prediction accuracy of heavy fog weather when data distribution changes.

Method used

A small-scale foggy weather prediction method based on deep learning is adopted. By obtaining meteorological data of the source area and the target area, a visibility prediction model is constructed using the LSTM regression network for pre-training, and the model parameters are updated online in real time to adapt to the changes in data distribution.

Benefits of technology

Improve the real-time and accuracy of foggy weather forecasts, ensuring that the prediction results remain high accuracy when data distribution changes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119065030A_ABST
    Figure CN119065030A_ABST
Patent Text Reader

Abstract

The invention discloses a small-scale heavy fog weather prediction method, system and device based on deep learning, and a medium, and relates to the technical field of weather prediction. The method comprises the following steps: acquiring meteorological data in a source area and meteorological data in a target area; pre-training a visibility prediction model constructed based on an LSTM regression network by using the meteorological data in the source region, inputting the meteorological data in the target region into the trained source region model for prediction and updating, and obtaining a visibility prediction result in the current time period; and performing judgment according to the visibility prediction result in the current time period and a preset threshold interval to obtain a foggy weather early warning level. The method can solve the problem that data distribution and fitting functions are fixed due to traditional mathematical model modeling, and then prediction precision is sharply reduced when data distribution changes.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of weather forecasting technology, and in particular to a method, system, device and medium for predicting small-scale heavy fog weather based on deep learning. Background Art

[0002] Heavy fog is a natural weather phenomenon, which refers to the phenomenon that under certain weather conditions, water vapor in the air condenses into a large number of tiny water droplets or ice crystals, which are suspended in the atmosphere near the ground, causing horizontal visibility to be significantly reduced. Heavy fog can occur in autumn, winter, and spring, and often has a huge impact on a large area of ​​China's territory. It is often accompanied by disastrous weather phenomena such as a sharp drop in visibility and slippery roads, and causes secondary disasters such as traffic accidents, flight delays, and shipwrecks, which have a huge and far-reaching impact on people's lives and property and even social stability. Every accurate forecast of heavy fog often directly serves hundreds of millions of people in many aspects of work and life. The direct influencing factor of heavy fog is visibility, but since visibility cannot be directly collected, meteorological parameters such as wind speed, air pressure, temperature, and humidity are used to predict the occurrence and development of heavy fog.

[0003] At present, weather forecasts rely extensively on mathematical models, which use historical meteorological data to construct mathematical equations for prediction. Specifically, a historical data set is first needed, which records the data of wind speed, air pressure, temperature, humidity and ten-minute average visibility. Scatter plots of ten-minute average visibility are made by wind speed, air pressure, temperature and humidity, and the relevant functions are fitted. The final mathematical equation is obtained through the weight determination method for prediction. However, this approach will preset a fixed data distribution, which will cause the prediction accuracy to drop significantly when the data distribution changes in the future. Summary of the invention

[0004] The purpose of the present invention is to provide a small-scale heavy fog weather prediction method, system, device and medium based on deep learning, which can solve the problem that traditional mathematical model modeling leads to fixed data distribution and fitting function, which in turn causes a sharp drop in prediction accuracy when the data distribution changes.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A small-scale foggy weather prediction method based on deep learning, including:

[0007] Acquire meteorological data in the source area and meteorological data in the target area; the meteorological data includes wind speed data, air pressure data, temperature data and humidity data within a set time period;

[0008] Using the meteorological data in the source area, pre-training a visibility prediction model built based on an LSTM regression network, and determining the pre-trained model as a source domain model;

[0009] Inputting the meteorological data in the target area into the source domain model for prediction to obtain the visibility prediction result in the current time period;

[0010] Using the visibility prediction result and the corresponding data, the model parameters in the previous time period are updated online in real time, and the updated model is used as the visibility prediction model in the next time period;

[0011] A judgment is made based on the visibility prediction result in the current time period and the preset threshold range to obtain the heavy fog weather warning level.

[0012] Optionally, the set time period is 12 hours.

[0013] Optionally, before inputting the meteorological data into the visibility prediction model, the method further includes: concatenating four one-dimensional tensors respectively representing wind speed data, air pressure data, temperature data and humidity data into a four-dimensional tensor.

[0014] The present invention also provides a small-scale foggy weather prediction system based on deep learning, comprising:

[0015] A collection unit, used to obtain meteorological data in a source area and meteorological data in a target area; the meteorological data includes wind speed data, air pressure data, temperature data and humidity data within a set time period;

[0016] A training unit, used to pre-train a visibility prediction model built based on an LSTM regression network using meteorological data in the source area, and determine the pre-trained model as a source domain model;

[0017] A prediction unit, configured to input the meteorological data in the target area into the source domain model for prediction, and obtain a visibility prediction result in a current time period;

[0018] An updating unit, configured to update the model parameters in the previous time period in real time online using the visibility prediction result and the corresponding data, and use the updated model as the visibility prediction model in the next time period;

[0019] The warning unit is used to make a judgment based on the visibility prediction result in the current time period and a preset threshold interval to obtain a heavy fog weather warning level.

[0020] Optionally, the acquisition unit includes a wind speed sensor, an air pressure sensor, a temperature sensor and a humidity sensor.

[0021] Optionally, it also includes: a communication module; the communication module is used to send the heavy fog weather warning level to relevant departments or local vehicles.

[0022] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned small-scale heavy fog weather prediction method based on deep learning.

[0023] The present invention also provides a computer-readable storage medium, characterized in that it stores a computer program, and when the computer program is executed by a processor, it implements the small-scale heavy fog weather prediction method based on deep learning as described above.

[0024] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0025] The present invention discloses a method, system, device and medium for predicting small-scale foggy weather based on deep learning, the method comprising obtaining meteorological data in a source area and meteorological data in a target area; using the meteorological data in the source area to pre-train a visibility prediction model constructed based on an LSTM regression network, inputting the meteorological data in the target area into the trained source domain model for prediction and updating, and obtaining a visibility prediction result in the current time period; judging according to the visibility prediction result in the current time period and a preset threshold interval, and obtaining a foggy weather warning level. The present invention can solve the problem that the traditional mathematical model modeling leads to fixed data distribution and fitting function, which leads to a sharp drop in prediction accuracy when the data distribution changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0027] Figure 1 It is a schematic diagram of the process of the small-scale heavy fog weather prediction method based on deep learning of the present invention;

[0028] Figure 2 is a data flow chart in this embodiment;

[0029] Figure 3 is a block diagram of the edge intelligent device in this embodiment. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] The purpose of the present invention is to provide a small-scale heavy fog weather prediction method, system, device and medium based on deep learning, which can solve the problem that traditional mathematical model modeling leads to fixed data distribution and fitting function, which in turn causes a sharp drop in prediction accuracy when the data distribution changes.

[0032] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] like Figure 1 As shown, the present invention provides a small-scale foggy weather prediction method based on deep learning, comprising:

[0034] Step 100: Acquire meteorological data in the source area and meteorological data in the target area; the meteorological data includes wind speed data, air pressure data, temperature data and humidity data within a set time period.

[0035] Step 200: Using the meteorological data in the source area, pre-train the visibility prediction model built based on the LSTM regression network, and determine the pre-trained model as the source domain model.

[0036] Step 300: input the meteorological data in the target area into the source domain model for prediction to obtain the visibility prediction result in the current time period.

[0037] Step 400: Use the visibility prediction results and corresponding data to perform real-time online updates on the model parameters in the previous time period, and use the updated model as the visibility prediction model in the next time period.

[0038] Step 500: Determine the level of foggy weather warning based on the visibility forecast result in the current time period and the preset threshold range.

[0039] As a specific implementation process, the following steps are included:

[0040] In the forecast area, the wind speed, air pressure, temperature, and humidity weather data collected in real time by edge intelligent devices are used to update the model parameters in real time, capture the relationship between the four variables of wind speed, air pressure, temperature, and humidity and visibility, ensure the real-time distribution of data, and improve the accuracy of foggy weather forecasts.

[0041] The mapping obtained by training with the historical dataset is where a t ,b t ,c t ,d t Represent wind speed, air pressure, temperature, and humidity weather data, respectively, while θ represents the parameters of the network model and a t ,b t ,c t ,d t The relationship parameters between them.

[0042] The sensors are used to continuously collect data, input them into the network and update the above θ. The four one-dimensional tensors representing the weather data of wind speed, air pressure, temperature and humidity are spliced ​​into a four-dimensional tensor, which is input into the visibility prediction model for visibility prediction. When the model converges, the output is obtained, and the collected real-time data is continuously sent to the model for prediction, and the model parameters are continuously changed to ensure the real-time distribution of data and improve the prediction accuracy.

[0043] Then, according to the visibility value obtained by the visibility prediction model, it is determined within which threshold range of the heavy fog weather forecast the current visibility is, and then the heavy fog level existing in the prediction area is determined.

[0044] On the other hand, a system for prediction is proposed, which collects wind speed, air pressure, temperature and humidity in real time through wind speed sensors, air pressure sensors, temperature sensors, humidity sensors and interfaces; sets a GPU module to use historical data sets to train models and uses real-time collected data to update model parameters; sets a 5G communication module to send prediction results to relevant departments or broadcast them to vehicles on the road in real time; sets a CPU module to execute program instructions; sets a readable medium module to store program codes.

[0045] As another embodiment, after investigating meteorological knowledge and the judgment of the China Meteorological Administration on heavy fog weather, the following two conclusions were obtained: First, the direct influencing factor of heavy fog weather is visibility, and the visibility indicator is often given as an average value over a period of time, and it is impossible to obtain a real-time indicator; second, the visibility indicator is affected by four main secondary factors, namely wind speed, air pressure, temperature, and humidity.

[0046] Therefore, after the deep learning model is trained in advance with the historical data set, it is migrated to the target domain, and the distribution of the data is updated online in real time according to the real-time data of wind speed, air pressure, temperature, and humidity collected by the intelligent edge device as described above, so as to obtain the foggy weather prediction result that is better than the traditional mathematical model. The content of this application is elaborated in detail below with the help of the specific implementation process.

[0047] like Figure 2 As shown in the figure, first, four data items of wind speed, air pressure, temperature and humidity are obtained in the prediction target area, and further formed into four one-dimensional tensors. Then, the model parameters are trained with the LSTM-based neural network model to obtain the data distribution and model weights, and capture the complex relationship between the four secondary factors of wind speed, air pressure, temperature, humidity and visibility.

[0048] In the LSTM-based neural network model, the specific process is:

[0049] The model processes the four-dimensional tensor data as described above, trains its own model, and continuously obtains new data from real-time data to update its own parameters. It should be noted that in meteorology, the impact of temperature and humidity on visibility can be seen in a short time, while the impact of wind speed and air pressure takes longer to show. Therefore, in LSTM, the role of the forget gate and update gate is to determine the retention of long-term data of wind speed and air pressure, while the data of temperature and humidity can be retained for a short time and forgotten for a long time.

[0050] Finally, a fully connected layer is set up to obtain the final visibility result through the final weighted summation, and the result is compared with the foggy weather threshold to obtain the foggy weather level result.

[0051] For example:

[0052] As a specific application, the wind speed, air pressure, temperature and humidity of the target area can be obtained. As mentioned above, assuming that a t ,b t ,c t ,d t Represent wind speed, air pressure, temperature, and humidity weather data respectively, and the four one-dimensional tensors obtained are a t ={a1,a2,…,a 12}, b t ={b1,b2,…,b 12}, c t ={c1,c2,…,c 12}, d t ={d1,d2,…,d 12}, assuming that data with 12 timestamps is used for training. The concatenated four-dimensional tensor is X = {a t ,b t ,c t ,d t}, after being sent to the prediction model, it passes through the fully connected layer to obtain a final visibility result Y, which is compared with the threshold to determine the fog threshold interval. After obtaining the result in a certain interval, the fog level weather is reported to the relevant departments or vehicles on the actual driving section.

[0053] like Figure 3 As shown, an edge intelligent device is constructed, including a communication module, a GPU module, a CPU module, and wind speed, air pressure, temperature, and humidity sensor interfaces, that is, four USB3.0 interfaces.

[0054] The communication module can transmit the results generated by the GPU module and the CPU module to relevant departments or broadcast them to vehicles on the road in real time through 5G. The four interfaces in the GPU module and the CPU module can support the access of various devices such as wind speed, air pressure, temperature, humidity sensors, etc., and collect data in real time.

[0055] Among them, the GPU module will train the collected real-time data. The training model is a LSTM-based neural network model, updating the model parameters and the corresponding data distribution. The CPU module will operate the program instructions and hand over the results to the communication module, which will distribute them.

[0056] In the GPU module and the CPU module, the CPU module can also compare the foggy weather determination threshold interval with the prediction result output by the GPU module to obtain the fog level prediction result.

[0057] As a specific implementation, the CPU module and the GPU module may be one or more. It should be noted that in the foggy weather prediction method provided in the embodiment of the present application, the program instructions are executed by the CPU, while the model training and parameter updating are completed by the GPU.

[0058] For a long time, the prediction of foggy weather has used mathematical models to fit data functions, and the method of determining the weights of each part of wind speed, air pressure, temperature, and humidity has been used to obtain the final fitting function. For example, based on the historical data set of the prediction area, temperature, wind speed, air pressure, humidity, and ten-minute average visibility are extracted to generate a characteristic scatter plot of ten-minute average visibility and temperature, wind speed, air pressure, and humidity, and a fitting mathematical function, including: taking temperature, wind speed, air pressure, and humidity as the horizontal axis and taking ten-minute average visibility as the vertical axis, generating a characteristic scatter plot of temperature, wind speed, air pressure, humidity, and visibility, and selecting a mathematical function according to the shape of each scatter plot for fitting. However, such a fitting function is based on a preset data distribution, but as time goes by, the distribution of wind speed, air pressure, temperature, and humidity data will change, which causes the original fitted function to lose accuracy, thereby causing a decrease in prediction accuracy. Furthermore, the heavy fog weather prediction method proposed in this application can update the data distribution of wind speed, air pressure, temperature, and humidity in real time, and update the model parameters based on the LSTM neural network, so that the accuracy of the visibility prediction results remains accurate and the heavy fog weather prediction results are guaranteed to be real-time.

[0059] Therefore, compared with the traditional mathematical model, the present invention has the following advantages:

[0060] (1) Improving the real-time and accuracy of predictions: This method uses the meteorological data collected in real time by the intelligent edge devices in the target area to update the model parameters to obtain more accurate and real-time prediction results.

[0061] (2) Strong adaptability: This method is not only applicable to foggy weather forecasting, but can also be applied to other types of weather forecasting by combining various collected meteorological data according to specific needs.

[0062] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0063] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A small-scale foggy weather prediction method based on deep learning, characterized in that: include: Acquire meteorological data in the source area and meteorological data in the target area; the meteorological data includes wind speed data, air pressure data, temperature data and humidity data within a set time period; Using the meteorological data in the source area, pre-training a visibility prediction model built based on an LSTM regression network, and determining the pre-trained model as a source domain model; Inputting the meteorological data in the target area into the source domain model for prediction to obtain the visibility prediction result in the current time period; Using the visibility prediction result and the corresponding data, the model parameters in the previous time period are updated online in real time, and the updated model is used as the visibility prediction model in the next time period; A judgment is made based on the visibility prediction result in the current time period and the preset threshold range to obtain the heavy fog weather warning level.

2. The small-scale foggy weather prediction method based on deep learning according to claim 1 is characterized in that: The set time period is 12 hours.

3. The small-scale foggy weather prediction method based on deep learning according to claim 1 is characterized in that: Before inputting the meteorological data into the visibility prediction model, the method further includes: concatenating four one-dimensional tensors representing wind speed data, air pressure data, temperature data and humidity data into a four-dimensional tensor.

4. A small-scale fog weather prediction system based on deep learning, characterized in that: include: A collection unit, used to obtain meteorological data in a source area and meteorological data in a target area; the meteorological data includes wind speed data, air pressure data, temperature data and humidity data within a set time period; A training unit, used to pre-train a visibility prediction model built based on an LSTM regression network using meteorological data in the source area, and determine the pre-trained model as a source domain model; A prediction unit, configured to input the meteorological data in the target area into the source domain model for prediction, and obtain a visibility prediction result in a current time period; An updating unit, configured to update the model parameters in the previous time period in real time online using the visibility prediction result and the corresponding data, and use the updated model as the visibility prediction model in the next time period; The warning unit is used to make a judgment based on the visibility prediction result in the current time period and a preset threshold interval to obtain a heavy fog weather warning level.

5. The small-scale fog weather prediction system based on deep learning according to claim 4 is characterized in that: The acquisition unit includes a wind speed sensor, an air pressure sensor, a temperature sensor and a humidity sensor.

6. The small-scale fog weather prediction system based on deep learning according to claim 4 is characterized in that: Also includes: Communication module: The communication module is used to send the heavy fog weather warning level to relevant departments or local vehicles.

7. An electronic device, characterized in that: It includes a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the small-scale heavy fog weather prediction method based on deep learning according to any one of claims 1-3.

8. A computer-readable storage medium, characterized in that: It stores a computer program, which, when executed by a processor, implements the small-scale heavy fog weather prediction method based on deep learning as described in any one of claims 1 to 3.

Citation Information

Patent Citations

  • Airport fog forecasting and early warning method based on long-term and short-term memory algorithm

    CN111175852A

  • Sea fog classification early warning method and device based on intelligent prediction, equipment and medium

    CN115097548A

  • Visibility prediction method and system, electronic equipment and storage medium

    CN118468099A