Dew intensity prediction method, system and equipment based on deep learning and medium

Abstract

The invention discloses a dew strength prediction method, system and equipment based on deep learning and a medium, and relates to the technical field of atmospheric environment monitoring. The method comprises the following steps: acquiring current meteorological factor data; the meteorological factor data is an arithmetic mean value of monitoring data per hour which is 1.5 m away from the earth surface and is the same as the dew condensation time period; inputting the meteorological factor data into the dew intensity prediction model for calculation to obtain a plurality of prediction values, and adding the plurality of prediction values to obtain a final dew intensity prediction value; the dew strength prediction model is a hybrid network composed of a long-short term memory network, a bidirectional gating circulation unit and a convolutional neural network. According to the invention, the accuracy of dew intensity prediction can be improved.

Description

Technical Field

[0001] The present invention relates to the field of atmospheric environment monitoring technology, and in particular to a dew intensity prediction method, system, equipment and medium based on deep learning. Background Art

[0002] Dew appears frequently in ecosystems such as cities, farmlands, and deserts, and the amount of condensation is considerable. It is an essential meteorological parameter for each ecosystem. Taking the farmland ecosystem as an example, an appropriate amount of dew can promote the further dissolution of fertilizers or pesticides and increase the rapid absorption of crops. However, too much dew will drip from the leaves, resulting in the loss of some foliar fertilizers and pesticides. Some nutrients in the dew can be used by certain fungi or pests, which will have an adverse effect on crop growth. During the flowering period of crops, heavy dew will be detrimental to the spread of pollen. Due to high humidity, it can lead to a high rate of empty rice shells, and a decrease in fertilization and fruiting rates. It can be seen that predicting the intensity of dew condensation in advance can provide a basis for specific operations such as the application of foliar fertilizers and pesticides in agricultural production, and the regulation of field water holding capacity. The research results are of great significance for the preliminary judgment of indicators such as leaf humidity, the preparation concentration and spraying time of foliar fertilizers and pesticides, and the field water holding capacity during the flowering period of rice. At the same time, it can effectively avoid the loss of fertilizers, etc., which is conducive to the efficient use of fertilizers, etc. and promotes grain production.

[0003] Due to the complexity of factors affecting the dew formation process, there is no systematic research on the prediction of dew intensity. Only in a few arid areas, statistical models are used to predict dew intensity. However, due to the nonlinear relationship between dew and local meteorological factors, the prediction accuracy of existing models is low, and they can only predict the general trend of dew intensity, but cannot accurately predict dew intensity. Summary of the invention

[0004] The purpose of the present invention is to provide a dew intensity prediction method, system, device and medium based on deep learning, which can improve the accuracy of dew intensity prediction.

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

[0006] A dew intensity prediction method based on deep learning, comprising:

[0007] Obtain current meteorological factor data; the meteorological factor data is the arithmetic average of hourly monitoring data at a distance of 1.5 meters from the ground surface and during the dew condensation period;

[0008] The data of various meteorological factors are input into the dew intensity prediction model for calculation to obtain multiple prediction values, and the multiple prediction values ​​are added together to obtain the final dew intensity prediction value; the dew intensity prediction model is a hybrid network composed of a long short-term memory network, a bidirectional gated recurrent unit and a convolutional neural network.

[0009] Optionally, the meteorological factor data include but are not limited to air temperature, relative humidity, wind speed, dew point temperature and net radiation.

[0010] Optionally, the construction and training process of the dew intensity prediction model is as follows:

[0011] Extracting training data from a database; the training data includes meteorological factor training data and corresponding dew intensity prediction data;

[0012] Build the initial training model;

[0013] The training data is input into the initial training model to set the training parameters and the zebra optimization algorithm to adjust the model parameters, and the trained model is determined as the dew intensity prediction model; the adjusted parameters include the number of iterations, the learning rate and the number of neurons in each layer.

[0014] Optionally, the database is constructed by:

[0015] Acquire multiple sets of relevant data; the relevant data include daily dew intensity and corresponding meteorological factors;

[0016] The correlation analysis in SPSS was used to determine the meteorological factors that were positively or negatively correlated with the dew intensity, and the selected meteorological factors were determined as independent variables, and the dew intensity was determined as the dependent variable;

[0017] All data are preprocessed: the training set dependent variable data is decomposed using empirical modes based on time-varying filtering to obtain multiple modal function components, and the components are denoised using partial autocorrelation functions to extract important features to obtain the final processed components; the final processed components are used for training the initial training model.

[0018] Optionally, the set training parameters include: 100 training cycles, a learning rate of 1e-2, a batch size of 32, and 64 neurons in all hidden layers.

[0019] The present invention also provides a dew intensity prediction system based on deep learning, comprising:

[0020] A data acquisition module is used to obtain current meteorological factor data; the meteorological factor data is the arithmetic average of hourly monitoring data at a distance of 1.5 meters from the ground surface and during the dew condensation period;

[0021] The dew intensity prediction module is used to input the data of various meteorological factors into the dew intensity prediction model for calculation to obtain multiple prediction values, and add the multiple prediction values ​​to obtain the final dew intensity prediction value; the dew intensity prediction model is a hybrid network composed of a long short-term memory network, a bidirectional gated recurrent unit and a convolutional neural network.

[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 perform the above-mentioned dew intensity prediction method based on deep learning.

[0023] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the dew intensity 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 dew intensity prediction method, system, device and medium based on deep learning, the method comprising obtaining current meteorological factor data; the meteorological factor data is the arithmetic mean of hourly monitoring data at a distance of 1.5 meters from the ground surface and during the same period of dew condensation; inputting each meteorological factor data into a dew intensity prediction model for calculation to obtain multiple prediction values, adding the multiple prediction values ​​to obtain a final dew intensity prediction value; the dew intensity prediction model is a hybrid network composed of a long short-term memory network, a bidirectional gated recurrent unit and a convolutional neural network. The present invention can improve the accuracy of dew intensity prediction. 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 flowchart of the dew intensity prediction method based on deep learning of the present invention;

[0028] Figure 2 This is a schematic diagram of the LSTM model in this embodiment;

[0029] Figure 3 Schematic diagram of the BiGRU model in this embodiment;

[0030] Figure 4 Schematic diagram of the dew intensity prediction model structure in this embodiment;

[0031] Figure 5 is a logic diagram of the dew intensity prediction method in this embodiment;

[0032] Figure 6 This is a schematic diagram of the comparison between the predicted value and the true value of the linear model applied in this embodiment;

[0033] Figure 7 A schematic diagram showing the comparison between the predicted value and the true value of the prediction model of this embodiment; DETAILED DESCRIPTION

[0034] 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.

[0035] The purpose of the present invention is to provide a dew intensity prediction method, system, device and medium based on deep learning, which can improve the accuracy of dew intensity prediction.

[0036] 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.

[0037] like Figure 1 As shown, the present invention provides a dew intensity prediction method based on deep learning, comprising:

[0038] Step 100: Obtain current meteorological factor data; the meteorological factor data is the arithmetic mean of hourly monitoring data at a distance of 1.5 meters from the ground and during the dew condensation period; wherein the meteorological factor data includes but is not limited to air temperature, relative humidity, wind speed, dew point temperature and net radiation.

[0039] Step 200: Input the data of various meteorological factors into a dew intensity prediction model for calculation to obtain multiple prediction values, and add the multiple prediction values ​​to obtain a final dew intensity prediction value; the dew intensity prediction model is a hybrid network composed of a long short-term memory network, a bidirectional gated recurrent unit and a convolutional neural network.

[0040] As a further implementation method, the construction and training process of the dew intensity prediction model is as follows:

[0041] Extract training data from a database; the training data includes meteorological factor training data and corresponding dew intensity prediction data; input the training data into the constructed initial training model to set training parameters and the zebra optimization algorithm to adjust the model parameters, and determine the trained model as a dew intensity prediction model; the adjusted parameters include the number of iterations, the learning rate and the number of neurons in each layer.

[0042] Among them, the method of database construction and data preprocessing is:

[0043] (1) Collecting dew intensity and meteorological factor data

[0044] Meteorological factors include but are not limited to air temperature, relative humidity, wind speed, dew point temperature, net radiation, etc. All meteorological factors are the arithmetic average of hourly monitoring data at a distance of 1.5 meters from the ground surface and during the dew condensation period. Among them, the monitoring method of dew intensity and dew condensation period can be implemented using existing technology.

[0045] (2) Establish a database of dew intensity and its corresponding meteorological factors

[0046] Each set of data includes daily dew intensity and its corresponding meteorological factors, and no less than 300 sets of data should be established. Correlation analysis in SPSS software was used to summarize meteorological factors that were positively correlated (P<0.05) or negatively correlated (P<0.05) with dew intensity. The selected related factors were independent variables, and dew intensity was the dependent variable.

[0047] (3) Data preprocessing of independent and dependent variables

[0048] The original data of independent variables and dependent variables are divided into training set, test set and validation set according to 70%, 15% and 15% respectively. All data are preprocessed, and the training set dependent variable data are decomposed using the time-varying filter-based empirical mode (TVFEMD) to obtain multiple modal function (IMF) components to reduce the model calculation complexity. The partial autocorrelation function (PACF) is used to reduce the noise of the components, extract important features, and input the final processed components into the prediction model.

[0049] As a further implementation method, the Hilbert transform (HT) is first proposed. HT is a mathematical transformation that converts a real-domain signal into a complex-domain signal. This signal contains information such as the amplitude and phase of the original signal. For a real-number signal x(t), its HT transform can be expressed as:

[0050]

[0051] Find the instantaneous amplitude A inst (t) and instantaneous phase

[0052]

[0053] Find the maximum and minimum values ​​of the instantaneous amplitude, and obtain two curves l1(t) and l2(t) based on interpolation estimation, and calculate θ1(t) and θ2(t) by the following formula:

[0054]

[0055] Interpolate A inst (t max ) 2 μ′(t max ) and A inst (t min ) 2 μ′(t min ) to obtain δ1(t) and δ2(t), and then continue to calculate μ′1(t) and μ′2(t):

[0056]

[0057] Calculate the local cutoff frequency:

[0058]

[0059] When the end condition is met, the loop ends, otherwise the above steps are executed repeatedly. The judgment criteria are recorded as:

[0060]

[0061] In the formula, B Loug (t) represents the instantaneous bandwidth, which represents the spectral characteristics of the signal at that time point. It represents the weighted average of instantaneous frequency, which can be defined according to specific requirements and applications.

[0062]

[0063] Where A represents the autocorrelation function (ACF).

[0064] The specific process of model construction is as follows:

[0065] The model is a hybrid of long short-term memory network (LSTM), bidirectional gated recurrent unit (BiGRU), and convolutional neural network (CNN). For the prediction model with manually set parameters, 100 training cycles, a learning rate of 1e-2, batchsize=32, and 64 neurons in all hidden layers are used.

[0066] like Figure 2 As shown, in the long short-term memory network (LSTM), there are:

[0067]

[0068] f t =σ·(W f ·[h t-1 ,x t ]+b f )

[0069] i t =σ·(W i ·[h t-1 ,x t ]+b i )

[0070] o t =σ·(W o ·[h t-1 ,x t ]+b o )

[0071]

[0072] h t =o t ·tanh(c t )

[0073] Among them, W f , W i , W o , W c is the weight. f 、b i 、b o 、b c For bias.

[0074] In a convolutional neural network (CNN), we have:

[0075]

[0076] Where RF i Indicates the size of the receptive field of the i-th layer, stride i is the stride of the convolution, is the size of the convolution kernel in the i-th layer.

[0077] like Figure 3 As shown, in the bidirectional gated recurrent unit (BiGRU), we have:

[0078] r t =σ(W r x t +U r h t-1 +b r )

[0079] z t =σ(W z x t +U z h t-1 +b z )

[0080]

[0081] Then adjust the parameters:

[0082] The dependent variable data and intelligent group algorithm are used to adjust the parameters of the model. The parameters include the number of iterations, learning rate, number of neurons in each layer, etc. The present invention uses the zebra optimization algorithm (IZOA) to adjust the parameters. The parameters stored in the individual output after each iteration are set to the deep learning model, and the prediction model is trained using the training group data. After the number of iterations is reached, the model training is completed. The present invention sets the population number to 20, the number of algorithm iterations to 15, and the parameter adjustment range of the prediction model is: learning rate [1e-3, e-1], batchsize [16, 256], number of training times [10, 100], and the number of neurons in other hidden layers [1, 100].

[0083] The specific process is:

[0084] H(i)=[ω j (i),ω j (i),…,ω j (i)]

[0085]

[0086] Among them, p is a prime number, and ω(n) represents a reverse sequence.

[0087]

[0088] Then perform model verification:

[0089] The trained model is applied to the test set and validation set data, and the mean square error of the predicted dew intensity value (predicted value) and the true value is calculated. The same weight is given to the data of the training set and the validation set to ensure that both can reach the optimal value.

[0090] Finally, the result prediction is:

[0091] Use a trained model (such as Figure 4 The predicted values ​​are added together to obtain the final predicted dew intensity value.

[0092] Based on the above technical solution, the following is provided: Figure 5 The embodiment shown.

[0093] Take the dryland (corn) system in Lishu City, Jilin Province from 2005 to 2022 as an example.

[0094] (1) Collecting dew intensity and meteorological factor data

[0095] The daily dew intensity of corn fields in Lishu City, Jilin Province (43.3081°N, 124.3374°E) was monitored from May to November from 2005 to 2022. The temperature, relative humidity, wind speed and dew point temperature were monitored during the same period. All meteorological factors were the arithmetic mean of the hourly monitoring data at a distance of 1.5 meters from the ground surface during the same period of dew condensation.

[0096] (2) Establish a database of dew intensity and its corresponding meteorological factors

[0097] Each set of data includes daily dew intensity and its corresponding meteorological factors, and a total of 986 sets of data were obtained. Through correlation analysis in SPSS software, air temperature, relative humidity and dew point temperature were positively correlated with dew intensity (P<0.05), and wind speed was negatively correlated with dew intensity (P<0.05). Air temperature, relative humidity, wind speed and dew point temperature were independent variables, and dew intensity was the dependent variable.

[0098] (3) Data preprocessing of independent and dependent variables

[0099] The original data of independent variables and dependent variables are divided into training set, test set and validation set according to 70%, 15% and 15% respectively. All data are preprocessed, and the training set dependent variable data are decomposed using the time-varying filter-based empirical mode (TVFEMD) to obtain 5 modal function (IMF) components. The partial autocorrelation function (PACF) is used to reduce the noise of the components, extract important features, and input the final processed components into the prediction model.

[0100] (4) Model construction

[0101] The prediction model uses a hybrid form of long short-term memory network (LSTM), bidirectional gated recurrent unit (BiGRU), and convolutional neural network (CNN). For the prediction model with manually set parameters, 100 training cycles, a learning rate of 1e-2, batch size = 32, and the number of neurons in all hidden layers is 64.

[0102] (5) Parameter adjustment

[0103] The zebra optimization algorithm (IZOA) is used to adjust the parameters. The parameters stored in the individuals output after each iteration are set to the deep learning model, and the prediction model is trained using the training group data. After reaching the number of iterations, the model training is completed. The population number is set to 20, the number of algorithm iterations is set to 15, and the parameter adjustment range of the prediction model is: learning rate [1e-3, e-1], batch size [16, 256], training times [10, 100], and the number of neurons in other hidden layers [1, 100].

[0104] (6) Model verification

[0105] The trained model is applied to the test set and validation set data, and the mean square error of the predicted dew intensity value (predicted value) and the true value is calculated. The same weight is given to the data of the training set and the validation set to ensure that both can reach the optimal value.

[0106] (7) Result prediction

[0107] like Figure 6 and Figure 7 As shown, the degree of fit between the model prediction value and the measured value using the linear fitting is lower than the degree of fit using the depth model of this embodiment. It can be seen that the depth model proposed by the present invention has a higher accuracy in predicting dew intensity.

[0108] Therefore, compared with the prior art, the beneficial technical effects achieved by the present invention are: overcoming the disadvantage of low accuracy of the previous model, the deep learning model used in the present invention uses back propagation to explore the intrinsic structure in large-scale data, pre-process all data, including data cleaning and feature selection, delete outliers and fill missing values, and select feature data groups with high correlation to reduce the dimension of the mixed model. The present invention can accurately predict dew intensity through the deep nonlinear relationship between dew intensity and meteorological factors. At the same time, it can process data in large-scale parallel and has good fault tolerance and adaptability.

[0109] In addition, the present invention also provides a dew intensity prediction system based on deep learning, comprising:

[0110] A data acquisition module is used to obtain current meteorological factor data; the meteorological factor data is the arithmetic average of hourly monitoring data at a distance of 1.5 meters from the ground surface and during the dew condensation period;

[0111] The dew intensity prediction module is used to input the data of various meteorological factors into the dew intensity prediction model for calculation to obtain multiple prediction values, and add the multiple prediction values ​​to obtain the final dew intensity prediction value; the dew intensity prediction model is a hybrid network composed of a long short-term memory network, a bidirectional gated recurrent unit and a convolutional neural network.

[0112] 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 perform the above-mentioned dew intensity prediction method based on deep learning.

[0113] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the dew intensity prediction method based on deep learning as described above.

[0114] 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.

[0115] 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 dew intensity prediction method based on deep learning, characterized in that: include: Get current meteorological factor data; The meteorological factor data is the arithmetic mean of hourly monitoring data at a distance of 1.5 meters from the ground surface during the dew condensation period; The data of various meteorological factors are input into the dew intensity prediction model for calculation to obtain multiple prediction values, and the multiple prediction values ​​are added together to obtain the final dew intensity prediction value; the dew intensity prediction model is a hybrid network composed of a long short-term memory network, a bidirectional gated recurrent unit and a convolutional neural network.

2. The dew intensity prediction method based on deep learning according to claim 1, characterized in that: The meteorological factor data include but are not limited to air temperature, relative humidity, wind speed, dew point temperature and net radiation.

3. The dew intensity prediction method based on deep learning according to claim 1, characterized in that: The construction and training process of the dew intensity prediction model is as follows: Extracting training data from a database; the training data includes meteorological factor training data and corresponding dew intensity prediction data; Build the initial training model; The training data is input into the initial training model to set the training parameters and the zebra optimization algorithm to adjust the model parameters, and the trained model is determined as the dew intensity prediction model; the adjusted parameters include the number of iterations, the learning rate and the number of neurons in each layer.

4. The dew intensity prediction method based on deep learning according to claim 3 is characterized in that: The database is constructed as follows: Acquire multiple sets of relevant data; the relevant data include daily dew intensity and corresponding meteorological factors; The correlation analysis in SPSS was used to determine the meteorological factors that were positively or negatively correlated with the dew intensity, and the selected meteorological factors were determined as independent variables, and the dew intensity was determined as the dependent variable; Preprocess all data: decompose the dependent variable data of the training set using empirical modes based on time-varying filtering to obtain multiple modal function components, and use partial autocorrelation functions to perform noise reduction on the components, extract important features, and obtain the final processed components; The component that is finally processed is used for training the initial training model.

5. The dew intensity prediction method based on deep learning according to claim 3, characterized in that: The training parameters are set as follows: 100 training cycles, a learning rate of 1e-2, a batch size of 32, and 64 neurons in all hidden layers.

6. A dew intensity prediction system based on deep learning, characterized in that: include: Data acquisition module, used to obtain current meteorological factor data; The meteorological factor data is the arithmetic mean of hourly monitoring data at a distance of 1.5 meters from the ground surface during the dew condensation period; The dew intensity prediction module is used to input the data of various meteorological factors into the dew intensity prediction model for calculation to obtain multiple prediction values, and add the multiple prediction values ​​to obtain the final dew intensity prediction value; the dew intensity prediction model is a hybrid network composed of a long short-term memory network, a bidirectional gated recurrent unit and a convolutional neural network.

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 perform the dew intensity prediction method based on deep learning according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that: It stores a computer program, which, when executed by a processor, implements the dew intensity prediction method based on deep learning as described in any one of claims 1 to 5.

Images