CN210895535U - Air temperature prediction system based on convolution cyclic neural network - Google Patents
Air temperature prediction system based on convolution cyclic neural network Download PDFInfo
- Publication number
- CN210895535U CN210895535U CN202020203130.6U CN202020203130U CN210895535U CN 210895535 U CN210895535 U CN 210895535U CN 202020203130 U CN202020203130 U CN 202020203130U CN 210895535 U CN210895535 U CN 210895535U
- Authority
- CN
- China
- Prior art keywords
- neural network
- air temperature
- computer
- prediction system
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The utility model provides an air temperature prediction system and method based on convolution circulation Neural network (CRNN), which relates to the technical field of artificial intelligence, the utility model takes a Neural network chip as the core, and a plurality of temperature sensors are utilized to collect the required air temperature data in the target area; using a computer to carry out data preprocessing and data storage; and finally, completing the tasks of training a neural network model, searching an optimal weight value, predicting the future temperature condition and the like by using a neural network chip. The utility model discloses required training data is the historical gas temperature value of multiple spot in the prediction range, and required input data is the multiple spot gas temperature value in a period before the prediction target date. Compare in traditional temperature prediction system based on degree of depth neural network, the utility model discloses need not thoroughly to know the regional atmosphere physical structure of target, also reduced the task volume of required analog computation, have higher practical value.
Description
Technical Field
The utility model belongs to the technical field of the artificial intelligence technique and specifically relates to an air temperature prediction system based on convolution circulation neural network.
Background
With the rapid development of artificial intelligence and deep neural networks in recent years, people obtain great convenience in daily life. The deep neural network has achieved great success in the fields of image recognition, voice translation, intelligent recommendation, automatic driving of automobiles and the like. However, in the field of air temperature prediction, the existing mainstream air temperature prediction method is still simulation prediction based on the meteorological theory, but the traditional simulation prediction based on the meteorological theory has the defects of large amount of computing resources, high computing time consumption and the like. Therefore, the air temperature prediction based on the deep neural network has the necessity, but the prediction precision of the air temperature prediction algorithm based on the deep neural network still needs to be improved.
The more accurate temperature prediction is helpful to the social aspect. For most people, air temperature forecasting helps them choose to dress; for some people with special diseases such as rheumatism, climate prediction can help them prevent pain; for many plants and businesses, air temperature prediction may also help them develop reasonable production and sales strategies.
Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) are two deep Neural network structures that are currently widely used. Convolutional neural networks are particularly suitable for processing two-dimensional data, and typically consist of the following structure: a convolution layer, an activation function layer, a pooling layer and a full-link layer. The recurrent neural network is developed for processing continuous data, and is effective in finding time correlation between continuous data. The convolution cyclic neural network has two advantages and is very suitable for processing two-dimensional continuous data. The method is well applied to the fields of audio classification, character recognition with missing parts, damaged video patching and the like.
Based on the current situation, the patent provides an air temperature prediction system based on a convolution cyclic neural network.
SUMMERY OF THE UTILITY MODEL
The utility model aims at providing an air temperature prediction system based on convolution circulation neural network to solve the problem that proposes among the above-mentioned background art.
In order to realize the purpose, the technical scheme of the utility model is that:
the air temperature prediction system based on the convolution circulation neural network comprises air temperature data collection equipment, a computer memory, a computer processor, a neural network chip and a computer display screen; the output of the air temperature data collection device is connected with the input of the computer processor; the computer memory is bidirectionally connected with the computer processor; the neural network chip is bidirectionally connected with the computer processor; the computer processor is bidirectionally connected with the computer display screen.
Further, the neural network chip can be built in a tablet computer, a notebook computer or a combination of the tablet computer and the notebook computer.
Furthermore, the neural network chip can be made into a computer board card form and inserted into a personal desktop computer, a server, a workstation and a large and medium-sized computer to construct a large-scale neural network.
Compared with the prior art, the beneficial effects of the utility model are that:
the air temperature prediction system based on the convolution circulation neural network comprises a novel convolution circulation neural network structure, and the structure can effectively predict future air temperature values through time series air temperature data. The structure is a deep neural network structure synthesized by a convolutional neural network and a cyclic neural network. The structure can effectively predict the future air temperature value through the time series air temperature data. The convolutional neural network part is used for processing the spatial correlation in each air temperature data map, and the cyclic neural network part is used for processing the time correlation between the continuous air temperature data maps. And finally, generating a predicted value of the future air temperature through the full connection layer.
Compare in traditional meteorology air temperature prediction system, the utility model provides a model can be under the condition of the meteorology physical structure of working environment of not having been solved, come from the time-space correlation of main its air temperature value of study and then carry out future air temperature prediction through local historical air temperature data.
Compared with the traditional air temperature prediction method based on one-place time sequence, the system substitutes a plurality of air temperature values into the deep neural network for training in the form of an air temperature map, and further improves the accuracy of air temperature prediction by finding out the spatial correlation among all air temperature observation points.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present application and should not be construed as unduly limiting the present application.
Fig. 1 is a schematic structural diagram of an air temperature prediction system based on a convolutional recurrent neural network in the present application.
Fig. 2 is a flowchart of a prediction method of the gas temperature prediction system of the present application.
Fig. 3 is a block diagram of a convolutional recurrent neural network in the present application.
Fig. 4 is a network structure diagram of a convolutional neural network part in the convolutional recurrent neural network of the present application.
Fig. 5 is a diagram of a long-term and short-term memory layer in the convolutional recurrent neural network of the present application.
FIG. 6 is a graph comparing the predicted air temperature value with the actual air temperature value in the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some, but not all, embodiments of the present application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The air temperature prediction system provided by the present application will be described with reference to fig. 1 to 6.
Example 1:
an air temperature prediction system based on a convolution circulation neural network comprises air temperature data collection equipment, a computer memory, a computer processor, a neural network chip and a computer display screen; the output of the air temperature data collection device is connected with the input of the computer processor; the computer memory is bidirectionally connected with the computer processor; the neural network chip is bidirectionally connected with the computer processor; the computer processor is bidirectionally connected with the computer display screen. The neural network chip can be arranged in a tablet computer, a notebook computer or a device integrating the tablet computer and the notebook computer. The neural network chip can be made into a computer board card form and inserted into a personal desktop computer, a server, a workstation and a large and medium computer to construct a large-scale neural network.
Example 2:
the prediction method of the air temperature prediction system is shown in a flow chart of fig. 2, and comprises the following steps:
acquiring historical temperature data with time labels of all temperature detection points in a target area; the utility model discloses can be used to many places future temperature prediction on a large scale, like the future average temperature every day that will predict all over the country, need the historical average temperature data every day of all over the country. For further explanation, the daily temperature data set provided by the chinese meteorological data network is selected in this embodiment. The data set comprises daily average temperature data from 1 month 1 day in 1951 to 12 months 31 days in 2018 and longitude and latitude information of all temperature observation stations provided by about seven hundred temperature observation stations all over the country.
Drawing the air temperature values of the same time label into a two-dimensional air temperature map according to the actual geographical position of each temperature detection point in the target area; in the embodiment, a national daily average air temperature map is drawn according to the longitude and latitude information of the air temperature observation station and the corresponding daily average air temperature value.
Step three, sequencing the temperature maps at all times according to a time sequence, intercepting a historical temperature map time sequence tensor with a specific length according to the specific time length, and carrying out normalization processing on temperature values in the tensor; in this example, the length of the time series was selected to be 7 days. Therefore, the daily average temperature maps are sorted according to the dates to generate a plurality of historical temperature map time series tensors with the length of 7, and the temperature values in all the temperature map time series tensors are normalized.
Step four, the historical temperature map time sequence tensor generated in the step three is calculated according to the following formula 3: 1: the ratio of 1 is divided into a training set, a verification set and a test set.
And fifthly, constructing a convolution cyclic neural network structure, wherein the constructed convolution cyclic neural network structure is shown in fig. 3, the convolution cyclic neural network comprises a convolution neural network part and a cyclic neural network part, the convolution neural network part comprises a two-dimensional convolution layer, an activation function layer, a specification layer and a pooling layer, the cyclic neural network part comprises a flattening layer, a long-time memory cycle layer, a specification layer, a full-connection layer and an activation function layer, the output of the convolution neural network part is the input of the cyclic neural network part, and the initial value of each hyper-parameter in the convolution cyclic neural network structure is set at the same time. The hyper-parameters comprise the number of convolution kernels of each two-dimensional convolution layer, the size of the convolution kernel of each two-dimensional convolution layer, the activation function of each activation function layer, the pooling size of each pooling layer, the number of long-term memory neurons in the circulation layer, loss function selection, optimizer selection, initial learning rate setting and the like. There will be several possible values for each hyper-parameter. And finally, the selected hyper-parameter value combination is the hyper-parameter value combination with the best effect after K-fold cross validation evaluation.
Step six, inputting the classified historical temperature chart time sequence tensor in the step four into the convolution cyclic neural network structure constructed in the step five;
step seven, respectively extracting temperature space correlation characteristics embedded in each historical temperature diagram through the convolution neural network part of the convolution circulation neural network structure in the step five, inputting the characteristics to the circulation neural network part according to the time sequence, extracting temperature value time correlation embedded between the historical temperature diagram time sequences by using the circulation neural network part, and finally obtaining a prediction result through a full connection layer;
in this embodiment, the network structure of the convolutional neural network is shown in FIG. 3, in this portion, the temperature map time series tensor x with the size of T × H × W is inputi,t. T represents the time series length, which is 7 in this example, H and W represent the pixel value length and the pixel value width of the two-dimensional temperature map, respectively, and i represents the ordinal number of this tensor in all tensors. The input data is sent to the convolutional neural network part, passes through three rounds of two-dimensional convolutional layers, the activation function layer, the normalization layer and the pooling layer, and the output tensor is zi，t. The functional mapping of the convolutional neural network portion output to input is as follows:
zi,t＝f(xi,t；{wx})
in the above function mapping relation, wxAre the weight values of the convolutional neural network portion. Three rounds of convolution extract and learn the spatial correlation features in each of the temperature maps. Output z of convolutional neural networki，tWhich is also the input to the next recurrent neural network portion.
In the recurrent neural network part, the long-term memory layer is the core structure of the part, the structure diagram of which is shown in fig. 4, and the long-term memory layer can extract and learn the time correlation characteristics in the temperature diagram time series. The function mapping relation between the timing sequence vectors extracted by the long-time and short-time memory layers is as follows:
hi,t＝f(hi,t-1,zi,t；{wz})
in the above function mapping relation, wzThe weight value of the long-time and short-time memory layers. The output hi of the long-time and short-time memory layer becomes the input of the following full-connection layer, and the function mapping relation of the full-connection layer is as follows:
in the above function mapping relation, whThe weight value of the long-time and short-time memory layers. The output of the full connection layer is the predicted value of the future air temperature, and the size of the predicted value is equal to the wholeThe input of each convolution cyclic neural network is T × H × W.
Step eight, comparing the prediction result obtained in the step seven with the actual historical air temperature value, solving an error value and sending the error value back to the convolutional neural network structure through a back propagation method;
step nine, repeating the step seven and the step eight for a plurality of times, and storing a group of neural network weight values with minimum verification loss;
and step ten, repeating the steps from four to nine by adopting a K-fold cross validation method, judging a group of hyper-parameter values which are most suitable for the current target area environment according to the prediction result, and storing the optimal network weight value under the corresponding hyper-parameter value, wherein the hyper-parameter result with the optimal effect is obtained in the embodiment, the number of convolution kernels of three-wheel two-dimensional convolution layers is respectively 64, 128 and 256, the convolution sum of each two-dimensional convolution layer is 3 × 3, the activation function of each activation function layer is a leaked nonlinear rectification function (Leaky-Relu), the pooling size of each pooling layer is 2 × 2, the number of memory neurons in the cycle layer is 1024 when the cycle layer is long, the loss function is a mean square error loss function, the optimizer is an Adam optimizer adopting Nesterov momentum, and the initial learning rate is 0.02.
Step eleven, acquiring air temperature value data of each temperature detection point in the target area through a temperature sensor;
step twelve, calculating the air temperature value data of each temperature detection point in the target area obtained in the step eleven and the optimal neural network weight value stored in the step ten, and then performing inverse normalization processing on the calculation result to further obtain the required air temperature prediction value.
Finally, the total average air temperature error of the prediction result in the embodiment is 0.907 ℃, the root mean square error is 1.697 ℃, the proportion of the air temperature difference value smaller than 1 ℃ is about 68.9%, the proportion of the air temperature difference value smaller than 2 ℃ is about 83.0%, and the proportion of the air temperature difference value smaller than 3 ℃ is about 91.4%. Compared with the traditional air temperature prediction algorithm based on the deep neural network, the overall prediction result is improved. Fig. 6 shows a comparison between the predicted air temperature value and the actual air temperature value, which are predicted using the optimal weight values. The graph is a comparison graph of actual air temperatures measured by about 700 observation stations nationwide on 10.1.2018 and predicted values obtained by the method. The abscissa is the serial number of the air temperature observation station, and the ordinate is centigrade. It can be seen from the figure that the temperature prediction result of the method is close enough to the actual value.
It is to be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The foregoing detailed description of the embodiments of the present application has been presented to illustrate the principles and implementations of the present application, and the above description of the embodiments is only provided to help understand the method and the core concept of the present application; meanwhile, for a person skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.
Claims (3)
1. An air temperature prediction system based on a convolution cyclic neural network is characterized in that: the system comprises air temperature data collection equipment, a computer memory, a computer processor, a neural network chip and a computer display screen;
the output of the air temperature data collection device is connected with the input of the computer processor;
the computer memory is bidirectionally connected with the computer processor;
the neural network chip is bidirectionally connected with the computer processor;
the computer processor is bidirectionally connected with the computer display screen.
2. The air temperature prediction system based on the convolutional recurrent neural network as claimed in claim 1, wherein: the neural network chip can be arranged in a tablet computer, a notebook computer or a device integrating the tablet computer and the notebook computer.
3. The air temperature prediction system based on the convolutional recurrent neural network as claimed in claim 1, wherein: the neural network chip can be made into a computer board card form and inserted into a personal desktop computer, a server, a workstation and a large and medium computer to construct a large-scale neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202020203130.6U CN210895535U (en) | 2020-02-24 | 2020-02-24 | Air temperature prediction system based on convolution cyclic neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202020203130.6U CN210895535U (en) | 2020-02-24 | 2020-02-24 | Air temperature prediction system based on convolution cyclic neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN210895535U true CN210895535U (en) | 2020-06-30 |
Family
ID=71317588
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202020203130.6U Expired - Fee Related CN210895535U (en) | 2020-02-24 | 2020-02-24 | Air temperature prediction system based on convolution cyclic neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN210895535U (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111199283A (en) * | 2020-02-24 | 2020-05-26 | 张早 | Air temperature prediction system and method based on convolution cyclic neural network |
-
2020
- 2020-02-24 CN CN202020203130.6U patent/CN210895535U/en not_active Expired - Fee Related
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111199283A (en) * | 2020-02-24 | 2020-05-26 | 张早 | Air temperature prediction system and method based on convolution cyclic neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110598029B (en) | Fine-grained image classification method based on attention transfer mechanism | |
| CN114067160B (en) | Small sample remote sensing image scene classification method based on embedded smooth graph neural network | |
| CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| US11874429B2 (en) | High-temperature disaster forecast method based on directed graph neural network | |
| Xiao et al. | Classification of ice crystal habits observed from airborne Cloud Particle Imager by deep transfer learning | |
| CN110059716B (en) | Construction of CNN-LSTM-SVM network model and MOOC conjugate prediction method | |
| CN111369535B (en) | Cell detection method | |
| CN112598165B (en) | Urban functional area transfer flow prediction method and device based on private car data | |
| CN114445634A (en) | Sea wave height prediction method and system based on deep learning model | |
| CN114332545B (en) | Image data classification method and device based on low-bit pulse neural network | |
| WO2022039675A1 (en) | Method and apparatus for forecasting weather, electronic device and storage medium thereof | |
| CN111428419A (en) | Suspended sediment concentration prediction method and device, computer equipment and storage medium | |
| Barnes et al. | This looks like that there: Interpretable neural networks for image tasks when location matters | |
| CN116345555A (en) | CNN-ISCA-LSTM model-based short-term photovoltaic power generation power prediction method | |
| CN210895535U (en) | Air temperature prediction system based on convolution cyclic neural network | |
| Barve et al. | Air quality index forecasting using parallel dense neural network and LSTM cell | |
| CN112560633B (en) | Plant key matter waiting period time point prediction method and system based on deep learning | |
| CN113220915B (en) | Remote sensing image retrieval method and device based on residual attention | |
| CN110716998A (en) | Method for spatializing fine-scale population data | |
| CN111199283A (en) | Air temperature prediction system and method based on convolution cyclic neural network | |
| CN113762331A (en) | Relational self-distillation method, apparatus and system, and storage medium | |
| CN117408167A (en) | Debris flow disaster vulnerability prediction method based on deep neural network | |
| CN113095251A (en) | Human body posture estimation method and system | |
| CN117233869A (en) | Site short-term wind speed prediction method based on GRU-BiTCN | |
| Yang et al. | Predictor selection for CNN-based statistical downscaling of monthly precipitation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200630 |