Deep learning intelligent grid temperature forecasting platform based on AI
Through the deep learning intelligent grid temperature forecasting platform based on AI, the problem of insufficient temperature prediction accuracy in traditional numerical weather forecasting models at specific locations is solved, and higher temperature prediction accuracy and adaptability are achieved.
Patent Information
- Application Number
- CN202510157435.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-13
- Publication Date
- 2025-06-03
AI Technical Summary
The traditional numerical weather forecast model has insufficient accuracy in temperature prediction at specific locations, making it difficult to meet the needs of agriculture, energy management and public safety.
Using the deep learning intelligent grid temperature forecasting platform based on AI, temperature prediction maps are built through data acquisition, preprocessing, deep learning model prediction and correction modules to improve the accuracy of temperature prediction at specific locations.
Improve the accuracy of temperature predictions in specific locations, make up for the shortcomings of traditional methods in specific location predictions, and adapt to the challenges brought by climate change.
Smart Images
Figure CN120085392A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of meteorological forecasting, and more specifically, to an AI-based deep learning intelligent grid temperature forecasting platform. Background Art
[0002] With the increasingly significant impact of climate change, accurate short-term temperature forecasting is crucial for fields such as agriculture, energy management, and public safety. Traditional temperature forecasting methods rely on numerical weather prediction models. Although these models can provide meteorological trend predictions within a certain spatial range, there are deficiencies in the accuracy of temperature prediction at specific locations.
[0003] Therefore, how to provide an AI-based deep learning intelligent grid temperature forecasting platform is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention
[0004] In view of this, the present invention provides an AI-based deep learning intelligent grid temperature forecasting platform, which improves the accuracy of temperature prediction at specific locations.
[0005] In order to achieve the above object, the present invention adopts the following technical solutions:
[0006] An AI-based deep learning intelligent grid temperature forecasting platform, comprising:
[0007] A data acquisition module, configured to obtain numerical weather prediction data of the area to be predicted;
[0008] A data preprocessing module, configured to preprocess the numerical weather prediction data;
[0009] A deep learning model prediction module, configured to input the preprocessed numerical weather prediction data into a deep learning model to obtain a temperature prediction map, where the temperature prediction map includes temperature prediction values at each position within the entire area; wherein, the deep learning model includes the corresponding relationship between the numerical weather prediction data and the temperature prediction values; the numerical weather prediction data forms a grid according to positions, each grid point represents a longitude and latitude, and each grid point in the grid contains multiple weather parameters;
[0010] A correction module, configured to correct the temperature prediction values based on a correction coefficient and output intelligent grid temperature data.
[0011] Preferably, the data acquisition module includes:
[0012] A grid unit, configured to construct a grid according to the longitude and latitude of the area to be predicted;
[0013] A parameter superposition unit, configured to respectively obtain multiple weather parameters in the numerical weather prediction data corresponding to the longitude and latitude of each grid point in the grid, and superpose the multiple weather parameters on the grid point.
[0014] Preferably, the deep learning model includes: a convolutional layer, a pooling layer, an LSTM unit group, a fully connected layer, and a ResNet residual network;
[0015] The numerical weather prediction data after preprocessing is subjected to feature extraction through the convolutional layer to output an intermediate feature map;
[0016] The pooling layer performs a pooling operation on the intermediate feature map to generate a feature map with reduced dimensions;
[0017] The LSTM unit group captures the change trend in the time dimension of the feature map with reduced dimensions and outputs time series features in the form of a two-dimensional array;
[0018] The fully connected layer converts the time series features in the form of a two-dimensional array into a one-dimensional array and outputs a temperature prediction value;
[0019] The ResNet residual network performs spatial expansion on the preliminary temperature prediction value through ResNet to fill the data blank area and generate a complete temperature prediction map.
[0020] Preferably, the LSTM unit group includes two layers of LSTM units with 256 channels.
[0021] Preferably, it further includes a model training module for training the deep learning model; the model training module includes:
[0022] A sample set unit, configured to obtain historical numerical weather prediction data and corresponding measured temperatures in the area to be predicted, and perform gridification according to longitude and latitude as a sample set;
[0023] A training unit, configured to use the historical numerical weather prediction data in the sample set as input data and the corresponding measured temperature as output data to train the deep learning model.
[0024] Preferably, the correction module includes:
[0025] A historical error analysis unit, configured to analyze historical prediction errors and generate correction coefficients for different geographical locations or meteorological conditions based on the error distribution characteristics;
[0026] A real-time correction unit, configured to compare and calculate the difference between the temperature prediction value and the latest observation data, and perform adjustment in combination with the correction coefficient to obtain a preliminary prediction value. The calculation formula is as follows:
[0027] y adjusted =yinitial +α·(observed - y initial )
[0028] Wherein, y adjusted represents the preliminary predicted value, α represents the correction coefficient, observed represents the latest observed data, and y initial represents the temperature predicted value;
[0029] A spatial interpolation unit, which is used to perform spatial interpolation processing on the preliminary predicted value and output the intelligent grid temperature data.
[0030] Preferably, the data preprocessing module includes:
[0031] A data cleaning unit, which is used to clean the numerical weather prediction data, including removing outliers and filling in missing values;
[0032] A feature engineering unit, which is used to convert the non - normally distributed meteorological data after data cleaning into an approximately normal distribution, z - score standardization, and time marking and periodic encoding.
[0033] Preferably, the time marking and periodic encoding specifically include:
[0034] Calculate which day of the current year the current date is, and divide the current date by the total number of days in the current year to obtain a value in the range [0, 1];
[0035] Calculate which hour of the current day the current time is, and divide it by 24 to obtain a value in the range [0, 1];
[0036] Map the annual progress and hourly progress to the range [0, 2π] respectively;
[0037] Use sine and cosine functions for periodic encoding.
[0038] Preferably, the weather parameters include: wind speed, wind direction, humidity, and pressure at each altitude.
[0039] Through the above - mentioned technical solutions, compared with the prior art, the present invention discloses a deep - learning intelligent grid temperature prediction platform based on AI, which has the following advantages:
[0040] 1) Improve the accuracy of temperature prediction at specific locations
[0041] Fine - grained grid processing: The numerical weather prediction data is used to form a grid according to the location, and it is ensured that each grid point represents a specific longitude and latitude, so that the temperature prediction can reach a higher spatial resolution. This means that for a specific location, a more accurate temperature predicted value can be obtained, making up for the deficiency of the traditional numerical weather prediction model in predicting at specific locations.
[0042] 2) Application of the deep learning model: By using components such as convolutional layers, pooling layers, LSTM cell groups, fully connected layers, and ResNet residual networks, the deep learning model can capture complex patterns and spatio-temporal dependencies in meteorological data, thereby providing more accurate temperature predictions.
[0043] 3) Adaptation to the challenges brought about by climate change
[0044] Flexible response to dynamic changes: As the climate changes, the frequency of extreme weather events increases. Through continuous adjustment of the correction coefficient by the historical error analysis unit in the present invention, it can adapt to new meteorological patterns in real time, improving the response speed and prediction accuracy for abnormal weather events.
[0045] 4) Improved time tagging and periodic encoding: Using sine and cosine functions for time tagging and periodic encoding not only takes into account the annual and daily progress, but also effectively captures the periodic variation patterns of meteorological parameters over time, enhancing the model's ability to understand the changing trends of future meteorological conditions. Description of the Drawings
[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the provided drawings.
[0047] Figure 1 It is a schematic structural diagram of an AI-based deep learning intelligent grid temperature forecasting platform provided by the present invention.
[0048] Figure 2 It is a schematic structural diagram of the data acquisition module provided by the present invention.
[0049] Figure 3 It is a schematic structural diagram of the data preprocessing module of the present invention.
[0050] Figure 4 It is a schematic structural diagram of the deep learning model provided by the present invention.
[0051] Figure 5 It is a schematic structural diagram of the correction module provided by the present invention. Detailed Embodiments
[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0053] An AI-based deep learning intelligent grid temperature forecasting platform is disclosed in an embodiment of the present invention. As Figure 1 shown, it includes:
[0054] A data acquisition module for obtaining numerical weather forecast data of the area to be predicted;
[0055] A data preprocessing module for preprocessing the numerical weather forecast data;
[0056] A deep learning model prediction module for inputting the preprocessed numerical weather forecast data into the deep learning model to obtain a temperature prediction map, where the temperature prediction map includes temperature prediction values at each position in the entire area; among them, the deep learning model includes the correspondence between the numerical weather forecast data and the temperature prediction values; the numerical weather forecast data forms a grid according to positions, each grid point represents a longitude and latitude, and each grid point in the grid contains multiple weather parameters;
[0057] A correction module for correcting the temperature prediction values based on correction coefficients and outputting intelligent grid temperature data.
[0058] In this embodiment, as Figure 2 shown, the data acquisition module includes:
[0059] A grid unit for constructing a grid according to the longitude and latitude of the area to be predicted; specifically, based on the selected longitude and latitude range and resolution, a two-dimensional grid is created to cover the entire prediction area. Each grid point represents a specific geographical location, defined by its longitude and latitude coordinates. For example, at a resolution of 0.125°, a point is set every 0.125° from north to south in latitude and every 0.125° from west to east in longitude.
[0060] A parameter superposition unit is used to obtain multiple weather parameters in the numerical weather prediction data corresponding to the longitude and latitude of each grid point in the grid and superpose the multiple weather parameters on the grid point. Specifically, for each grid point (i.e., a specific longitude and latitude position), it is necessary to obtain the numerical weather prediction data at that position. These data can usually be obtained through the API interface provided by meteorological service providers. The numerical weather prediction data contains various weather parameters, including wind speed: wind speeds at different altitude levels; wind direction: wind directions at different altitude levels; humidity: relative humidity in the air; pressure: atmospheric pressure on the ground and at different altitude levels. The obtained numerical weather prediction data is "superposed" onto the corresponding grid points according to their corresponding longitude and latitude positions. Each grid point not only has its own geographical location information but also contains various meteorological parameter values.
[0061] In this embodiment, as Figure 3 shown, the data preprocessing module includes:
[0062] A data cleaning unit is used to clean the numerical weather prediction data, including removing outliers and filling in missing values;
[0063] A feature engineering unit is used to convert the non-normally distributed meteorological data after data cleaning into an approximately normal distribution, z-score standardization, and time marking and periodic encoding.
[0064] The time marking and periodic encoding specifically include:
[0065] Calculate which day of the current year the current date is, and divide the current date by the total number of days in the current year to obtain a value in the range [0, 1];
[0066] Calculate which hour of the current day the current time is, and divide it by 24 to obtain a value in the range [0, 1];
[0067] Map the year progress and hour progress to the range [0, 2π] respectively;
[0068] Use sine and cosine functions for periodic encoding.
[0069] The specific expression form of the year progress can be:
[0070]
[0071] year_angle = year_progress_ratio × 2π,
[0072] sin year = sin(year_angle),
[0073] cos year= cos(year_angle).
[0074] where day_of_year represents the current date, days_in_year represents the total number of days in the current year, year_progress_ratio represents the ratio of the current date in a year, year_angle represents the position of the current date in a year, sin year and cos year respectively represent the sine and cosine values of the current date in a year. These two values can be used to capture the periodic characteristics of seasonal changes.
[0075] The expression form of hour progress can be:
[0076]
[0077] hour_angle = hour_progress_ratio × 2π,
[0078] sin hour = sin(hour_angle),
[0079] cos hour = cos(hour_angle).
[0080] where hour_of_day represents the current time, hour_progress_ratio represents the position of the current hour in a day, sin hour and cos hour respectively represent the sine and cosine values of the current hour in a day. These two values can be used to capture the periodic characteristics of diurnal changes.
[0081] The present invention converts the original time information into a periodic representation of year progress and hour progress. This method not only retains the periodicity of the original time information, but also makes it easier for the deep learning model to learn the patterns in the data, which is particularly important when dealing with meteorological data with obvious periodicity. This encoding method helps to improve the accuracy of model prediction because it can better capture the changing trends in the time dimension.
[0082] In this embodiment, as Figure 4 shown, the deep learning model includes: a convolutional layer, a pooling layer, a group of LSTM units, a fully connected layer, and a ResNet residual network;
[0083] The preprocessed numerical weather prediction data is subjected to feature extraction through a convolutional layer to output an intermediate feature map. Specifically, one-dimensional convolution operations are performed using convolutional kernels of different sizes (such as 1×1, 1×5, 1×7) to obtain intermediate features with different receptive fields. After splicing these intermediate features, two one-dimensional convolution calculations are performed using a 1×3 convolutional kernel to generate an intermediate feature map.
[0084] The pooling layer performs a pooling operation on the intermediate feature map to generate a feature map with reduced dimensions, thereby reducing the amount of computation.
[0085] The LSTM cell group captures the changing trend in the time dimension of the feature map with reduced dimensions and outputs time series features in the form of a two-dimensional array. The LSTM cell group includes two layers of LSTM cells with 256 channels.
[0086] The fully connected layer converts the time series features in the form of a two-dimensional array into a one-dimensional array and outputs the temperature prediction value.
[0087] The ResNet residual network spatially expands the preliminary temperature prediction value through ResNet to fill the data blank area and generate a complete temperature prediction map. Specifically, the ResNet residual network of the present invention includes three residual blocks. The preliminary temperature prediction map is subjected to feature extraction and non-linear transformation through the first residual block, and the output of the first layer is continuously sent to the second residual block to further enhance the model's ability to capture spatial dependence. The result of the second layer is passed to the third residual block to continue deepening the feature extraction.
[0088] The present invention utilizes the powerful representation ability of ResNet to map the feature map of only meteorological observation stations in the region into a meteorological data feature map including the entire regional grid, thereby realizing the gridification of regional meteorological data.
[0089] It further includes a model training module for training the deep learning model; the model training module includes:
[0090] A sample set unit for obtaining the historical numerical weather prediction data and the corresponding measured temperature of the area to be predicted, and performing gridification according to longitude and latitude as the sample set.
[0091] A training unit for using the historical numerical weather prediction data in the sample set as input data and the corresponding measured temperature as output data to train the deep learning model.
[0092] In this embodiment, as Figure 5 shown, the correction module includes:
[0093] A historical error analysis unit is used to analyze historical prediction errors and generate correction factors for different geographical locations or meteorological conditions based on the characteristics of error distribution. Specifically, it collects in real time the predicted values and actual observed values of the model over a period of time in the past, calculates the errors of each group of data (such as root mean square error, mean absolute error, etc.), and generates correction factors for different geographical locations or meteorological conditions based on the characteristics of error distribution. By continuously collecting the predicted values of the model and the corresponding actual observed values, the present invention ensures the freshness and relevance of the data and achieves the technical effect of continuously adjusting the correction factors.
[0094] A real-time correction unit compares the temperature predicted value with the latest observed data to calculate the difference, and combines the correction factor for adjustment to obtain a preliminary predicted value. The calculation formula is as follows:
[0095] y adjusted =y initial +α·(observed - y initial )
[0096] Wherein, y adjusted represents the preliminary predicted value, α represents the correction factor, observed represents the latest observed data, and y initial represents the temperature predicted value;
[0097] A spatial interpolation unit is used to perform spatial interpolation processing on the preliminary predicted value and output intelligent grid temperature data, making the prediction result smoother and more continuous and reducing the prediction error caused by sparse data. Spatial interpolation processing can adopt techniques such as inverse distance weighted method (IDW), Kriging method, etc.
[0098] Wherein, the latest observed value refers to the measured value of the actual meteorological parameters at the current or the most recent time point obtained from the real-time data source. By comparing the temperature predicted value with the latest observed value, calculating the difference, and dynamically adjusting using the correction factor, the prediction result can be made closer to the actual situation. This real-time correction mechanism helps to improve the accuracy and reliability of the prediction, especially under complex and changeable meteorological conditions.
[0099] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and reference can be made to the description in the method part for the relevant parts.
[0100] The foregoing description of the disclosed embodiments enables those skilled in the art to practice or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A deep learning intelligent grid temperature forecasting platform based on AI, characterized in that: include: A data acquisition module is used to obtain numerical weather forecast data for the area to be predicted; A data preprocessing module, used for preprocessing the numerical weather forecast data; A deep learning model prediction module, used for inputting the preprocessed numerical weather forecast data into the deep learning model to obtain a temperature prediction map, wherein the temperature prediction map includes the temperature prediction value of each location in the entire area; wherein the deep learning model includes the corresponding relationship between the numerical weather forecast data and the temperature prediction value; the numerical weather forecast data is gridded according to the location, each grid point represents a longitude and latitude, and each grid point in the grid contains multiple weather parameters; The correction module is used to correct the temperature prediction value based on the correction coefficient and output the intelligent grid temperature data.
2. According to claim 1, a deep learning intelligent grid temperature forecasting platform based on AI is characterized in that: The data acquisition module comprises: Gridding unit, used to construct a grid according to the longitude and latitude of the area to be predicted; The parameter superposition unit is used to obtain multiple weather parameters in the numerical weather forecast data of the longitude and latitude corresponding to each grid point in the grid, and superimpose the multiple weather parameters on the grid point.
3. According to claim 1, a deep learning intelligent grid temperature forecasting platform based on AI is characterized in that: The deep learning model includes: a convolutional layer, a pooling layer, an LSTM unit group, a fully connected layer and a ResNet residual network; The numerical weather forecast data after preprocessing is subjected to feature extraction through the convolution layer to output an intermediate feature map; The pooling layer performs a pooling operation on the intermediate feature map to generate a feature map after dimensionality reduction; The LSTM unit group captures the change trend of the feature graph after dimension reduction in the time dimension, and outputs the time series features in the form of a two-dimensional array; The fully connected layer converts the time series features in the form of a two-dimensional array into a one-dimensional array and outputs a temperature prediction value; The ResNet residual network spatially expands the preliminary temperature prediction value through ResNet, fills in the data blank area, and generates a complete temperature prediction map.
4. According to claim 3, a deep learning intelligent grid temperature forecasting platform based on AI is characterized in that: The LSTM unit group includes two layers of LSTM units with 256 channels.
5. According to claim 1, a deep learning intelligent grid temperature forecasting platform based on AI is characterized in that: Also included is a model training module for training a deep learning model; the model training module includes: The sample set unit is used to obtain the historical numerical weather forecast data and the corresponding measured temperature of the area to be predicted, and grid them according to the longitude and latitude as a sample set; The training unit is used to train the deep learning model by taking the historical numerical weather forecast data in the sample set as input data and the corresponding measured temperature as output data.
6. The AI-based deep learning intelligent grid temperature forecasting platform according to claim 1, characterized in that: The correction module comprises: A historical error analysis unit, used to analyze historical forecast errors and generate correction coefficients for different geographical locations or meteorological conditions based on the error distribution characteristics; The real-time correction unit compares the temperature prediction value with the latest observation data to calculate the difference, and adjusts it with the correction coefficient to obtain the preliminary prediction value. The calculation formula is as follows: and adjusted =and initial +α·(observed-y initial ) Among them, y adjusted represents the preliminary forecast value, α represents the correction coefficient, observed represents the latest observation data, and y initial represents the predicted temperature value; The spatial interpolation unit is used to perform spatial interpolation processing on the preliminary prediction value and output intelligent grid temperature data.
7. The AI-based deep learning intelligent grid temperature forecasting platform according to claim 1, characterized in that: The data preprocessing module comprises: Data cleaning unit, used to clean numerical weather forecast data, including removing outliers and filling missing values; The feature engineering unit is used to convert the non-normally distributed meteorological data into approximately normal distribution after data cleaning, z-score standardization, and time labeling and periodicity encoding.
8. The AI-based deep learning intelligent grid temperature forecasting platform according to claim 7, characterized in that: Time stamping and periodic coding specifically include: Calculate the day of the year that the current date is, and divide the current date by the total number of days in the year to get a value in the range [0,1]; Calculate the hour of the day at the current time and divide it by 24 to get a value in the range [0,1]; Map the annual progress and hourly progress to the range of [0,2π] respectively; Periodic encoding using sine and cosine functions.
9. The AI-based deep learning intelligent grid temperature forecasting platform according to claim 1, characterized in that: The weather parameters include: wind speed, wind direction, humidity and pressure at each altitude.