Layered progressive meteorological element data reconstruction method based on BP neural network

Abstract

The invention discloses a hierarchical progressive meteorological element data reconstruction method based on a BP neural network. The method comprises the following steps: S1, collecting meteorological element data of meteorological stations in a research area and around the research area; comprising average air temperature, average air pressure, average relative humidity and sunshine duration observed day by day; s2, dividing the weather stations into verification weather stations and training weather stations based on the geographical location information of the weather stations in the research area and around the research area; s3, according to the data reconstruction target, determining observation data of corresponding characteristic parameters and target variables in the training meteorological station, and according to the observation data, constructing and optimizing a corresponding target model based on a BP neural network; and S4, according to the data reconstruction target, calling the optimized target model with the optimal model parameters to carry out gridding meteorological element data reconstruction. The high-quality and multi-element meteorological data reconstruction method provided by the invention provides a precious technical scheme for research and practice in a plurality of research fields such as water circulation, climate change, drought and flood disasters and the like.

Description

Technical Field

[0001] The invention belongs to the technical field of meteorological element reconstruction in hydrology, and in particular relates to a hierarchical and progressive meteorological element data reconstruction method based on a BP neural network. Background Art

[0002] In hydrological research and practice, high-quality and high-resolution meteorological data form the basis for understanding climate change and its impact on the water cycle in a watershed. Meteorological elements, including temperature, air pressure, relative humidity and sunshine hours, directly affect the evapotranspiration and water balance of a watershed, and play a decisive role in regulating the surface water and groundwater cycles. These elements are particularly critical for managing flood risks and optimizing water resource use, especially in the current environment where extreme weather events caused by climate change are becoming more frequent. However, the spatial distribution of meteorological observation sites is very uneven, especially in high-altitude areas, which makes it impossible to effectively capture the impact of complex terrain on meteorological parameters. Existing spatial interpolation methods often ignore the impact of terrain complexity when performing gridding processing, such as local changes caused by terrain on temperature, air pressure and humidity. In addition, traditional methods often fail to fully utilize the nonlinear relationship between these data when dealing with the interactions between multiple meteorological elements, thereby limiting the accuracy and reliability of hydrological model predictions. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the hierarchical and progressive meteorological element data reconstruction method based on BP neural network provided by the present invention can achieve high-resolution and high-quality reconstruction of various meteorological elements by adopting machine learning technology to deeply study the spatial distribution and interdependence of meteorological elements taking into account terrain factors.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a hierarchical progressive meteorological element data reconstruction method based on BP neural network, comprising the following steps:

[0005] S1. Collect meteorological element data from meteorological stations in and around the study area;

[0006] The meteorological element data include the average temperature, average air pressure, average relative humidity and sunshine hours observed daily;

[0007] S2. Based on the geographical location information of meteorological stations in and around the study area, the meteorological stations are divided into verification meteorological stations and training meteorological stations;

[0008] S3, reconstructing the target according to the data, determining the observation data of the corresponding characteristic parameters and target variables in the training weather station, and constructing and optimizing the corresponding target model based on the BP neural network according to the data;

[0009] S4, according to the data reconstruction target, calling the optimized target model with the optimal model parameters to reconstruct the gridded meteorological element data;

[0010] The target model includes a hierarchical and progressive average temperature model, an average air pressure model, an average relative humidity model and a sunshine duration model; wherein the characteristic parameters of the average temperature model include the geographical location information of the training meteorological station, and the target variable is the corresponding average temperature; the characteristic parameters of the average air pressure model include the geographical location information and the average temperature of the training meteorological station, and the target variable is the corresponding average air pressure; the characteristic parameters of the average relative humidity model include the geographical location information, the average temperature and the average air pressure of the training meteorological station, and the target variable is the corresponding average relative humidity; the characteristic parameters of the sunshine duration model include the geographical location information, the average temperature, the average air pressure and the average relative humidity of the training meteorological station, and the target variable is the corresponding sunshine duration.

[0011] The beneficial effects of the present invention are:

[0012] (1) The present invention can effectively reconstruct high-resolution meteorological element products by adopting BP neural network technology, significantly enhancing the spatial coverage of data, especially in areas where meteorological observation stations are scarce;

[0013] (2) Based on the hierarchical progressive strategy, the present invention gradually introduces temperature, air pressure, relative humidity and sunshine hours as the characteristic parameters and target variables of the model, which not only enhances the model's ability to learn the nonlinear relationship between the elements, but also improves the accuracy and reliability of the prediction results;

[0014] (3) The present invention uses meteorological station observation data that is not involved in model training for verification, thereby ensuring the accuracy and reliability of the reconstructed product.

[0015] (4) The high-quality, multi-factor meteorological data reconstruction method provided by the present invention provides a valuable technical solution for research and practice in multiple research fields such as water circulation, climate change, drought and flood disasters.

[0016] Further, in step S2, the verified meteorological stations include meteorological stations in the study area and its demarcated surrounding areas, which fully cover the upstream, midstream and downstream areas of the basin and its surrounding areas, as well as meteorological stations within different longitudes and latitudes and altitudes, and meteorological stations in areas where the density of meteorological station distribution is greater than a set threshold;

[0017] The training meteorological stations include the remaining meteorological stations within the study area and its designated surrounding areas except for the meteorological stations determined as verification meteorological stations.

[0018] The beneficial effects of the above further scheme are: by covering the observation data of comprehensive meteorological verification stations, it helps to verify the accuracy of reconstructed products in different geographical ranges; the balanced meteorological station allocation strategy aims to ensure full utilization of available observation data as training data, enhance the reliability of model training, and improve the reconstruction quality of meteorological products in areas where observations are missing.

[0019] Furthermore, step S3 is specifically as follows:

[0020] S1. Determine the characteristic parameters of the target model and their corresponding target variables;

[0021] S2. Input the daily observation data of characteristic parameters and target variables into the constructed BP neural network, and use the Adam optimizer and mean square error loss function to optimize the BP neural network, save the model parameters with the best performance every day, and obtain the corresponding target model.

[0022] The beneficial effects of the above further scheme are: the target model obtained by training and optimizing based on the above method fully considers the influence of different characteristic parameters on the corresponding target variables under the reconstruction targets based on different data;

[0023] Further, step S4 is specifically as follows:

[0024] S41, determining the current data reconstruction target, and extracting the geographical location information in the high-resolution DEM;

[0025] S42, the geographic location information is combined with the gridded meteorological element data reconstructed by the target model in all levels before the level where the current data reconstruction target is located, and the characteristic parameter standardization instance corresponding to the current data reconstruction target is used to perform standardization processing on it;

[0026] S43, according to the current data reconstruction target, calling the target model corresponding to the model parameters with the best performance on that day to process the standardized data, and generating target reconstruction data consistent with the DEM resolution;

[0027] S44, repeating steps S42 to S43, using the target model to independently process the daily standardized data, and reconstructing the gridded meteorological element data containing a long time series corresponding to the current data reconstruction target.

[0028] Furthermore, the construction method of the characteristic parameter standardization instance is:

[0029] According to the data reconstruction target, the observation data of the corresponding characteristic parameters in the training meteorological station are read, merged into a two-dimensional characteristic matrix, and then standardized and saved as a characteristic parameter standardized instance.

[0030] The beneficial effect of the above further scheme is that the meteorological element data is reconstructed through the target model of the optimal model parameters on a daily basis, which reduces the impact of short-term fluctuations on the overall accuracy of the long time series data set and ensures the consistency and traceability of the model over the entire time series.

[0031] Furthermore, the method also includes estimating and filling in the observed data, and the specific method is as follows:

[0032] When the daily observation data corresponding to the target variable of a meteorological station is missing, the characteristic parameter standardization instance is used to standardize the geographical location information of the meteorological station and the daily observation data corresponding to the characteristic parameters in all levels before the level where the target variable is located, and the target model with the optimal model parameters on the corresponding date is called to estimate and fill in the observation data corresponding to the target variable.

[0033] The beneficial effect of the above further scheme is that it can provide more accurate meteorological element information for areas lacking measurement.

[0034] Furthermore, the method further comprises:

[0035] The long sequence of meteorological element observation data observed by the verification meteorological station is used to evaluate the accuracy of the reconstructed meteorological element data of the target model.

[0036] Furthermore, the evaluation indicators for the accuracy evaluation of the reconstructed meteorological element data include the average error and correlation coefficient between the long-sequence meteorological element observation data and the corresponding reconstructed meteorological element data.

[0037] The beneficial effect of the above further scheme is that the reconstructed meteorological element data is evaluated by covering the observation data of a wide range of verification meteorological stations, ensuring a comprehensive evaluation of the quality of the reconstructed data.

[0038] Furthermore, the BP neural network includes an input layer, a first hidden layer, a second hidden layer and an output layer;

[0039] The number of neurons in the input layer is the number of characteristic parameters corresponding to the data reconstruction target;

[0040] The first hidden layer contains 64 neurons, uses the ReLU activation function, and applies an L2 regularization coefficient of 0.001;

[0041] The second hidden layer contains 32 neurons, uses the ReLU activation function, and applies an L2 regularization coefficient of 0.001;

[0042] The output layer is a single neuron that outputs the target variable.

[0043] The beneficial effects of the above further scheme are:

[0044] Through the refined design of multiple hidden layers, the BP neural network has enhanced its ability to learn the nonlinear relationship between complex meteorological elements and improved its generalization performance; the application of L2 regularization helps alleviate the overfitting problem and enhances the predictive stability and reliability of the model when processing unseen data; the precise configuration of the output layer enables the model to make detailed predictions and outputs of specific meteorological elements to meet diverse data reconstruction needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flow chart of the hierarchical and progressive meteorological element data reconstruction method based on BP neural network provided by the present invention.

[0046] Figure 2 This is a diagram of an application example of the data reconstruction method provided by the present invention. DETAILED DESCRIPTION

[0047] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0048] The embodiment of the present invention provides a hierarchical and progressive meteorological element data reconstruction method based on BP neural network, such as Figure 1 As shown, the following steps are included:

[0049] S1. Collect meteorological element data from meteorological stations in and around the study area;

[0050] The meteorological element data include the average temperature, average air pressure, average relative humidity and sunshine hours observed daily;

[0051] S2. Based on the geographical location information of meteorological stations in and around the study area, the meteorological stations are divided into verification meteorological stations and training meteorological stations;

[0052] S3, reconstructing the target according to the data, determining the observation data of the corresponding characteristic parameters and target variables in the training weather station, and constructing and optimizing the corresponding target model based on the BP neural network according to the data;

[0053] S4, according to the data reconstruction target, calling the optimized target model with the optimal model parameters to reconstruct the gridded meteorological element data;

[0054] The target model includes a hierarchical and progressive average temperature model, an average air pressure model, an average relative humidity model and a sunshine duration model; wherein the characteristic parameters of the average temperature model include the geographical location information of the training meteorological station, and the target variable is the corresponding average temperature; the characteristic parameters of the average air pressure model include the geographical location information and the average temperature of the training meteorological station, and the target variable is the corresponding average air pressure; the characteristic parameters of the average relative humidity model include the geographical location information, the average temperature and the average air pressure of the training meteorological station, and the target variable is the corresponding average relative humidity; the characteristic parameters of the sunshine duration model include the geographical location information, the average temperature, the average air pressure and the average relative humidity of the training meteorological station, and the target variable is the corresponding sunshine duration.

[0055] In step S2 of the embodiment of the present invention, the verified meteorological stations include meteorological stations in the study area and its demarcated surrounding areas, which fully cover the upstream, midstream and downstream areas of the basin and its surrounding areas, as well as meteorological stations within different longitudes and latitudes and altitudes, and meteorological stations in areas where the density of meteorological station distribution is greater than a set threshold;

[0056] The training meteorological stations include the remaining meteorological stations within the study area and its designated surrounding areas except for the meteorological stations determined as verification meteorological stations.

[0057] In this embodiment, the above-mentioned balanced allocation strategy of selecting verification meteorological stations according to the density of meteorological station distribution ensures the sufficiency and diversity of subsequent training data, enhances the reliability of model training, and improves the reconstruction quality of missing meteorological element data.

[0058] In step S2 of this embodiment, illustratively, the geographical location information of the weather station includes longitude, latitude, altitude, etc.

[0059] Step S3 of the embodiment of the present invention is specifically:

[0060] S1. Determine the characteristic parameters of the target model and their corresponding target variables;

[0061] S2. Input the observed data of feature parameters and target variables into the constructed BP neural network, and use the Adam optimizer and mean square error loss function to optimize the BP neural network, save the model parameters with the best performance every day, and obtain the corresponding target model.

[0062] Step S4 of the embodiment of the present invention is specifically as follows:

[0063] S41, determining the current data reconstruction target, and extracting the geographical location information in the high-resolution DEM;

[0064] S42, the geographic location information is combined with the gridded meteorological element data reconstructed by the target model in all levels before the level where the current data reconstruction target is located, and the characteristic parameter standardization instance corresponding to the current data reconstruction target is used to perform standardization processing on it;

[0065] S43, according to the current data reconstruction target, calling the target model corresponding to the model parameters with the best performance on that day to process the standardized data, and generating target reconstruction data consistent with the DEM resolution;

[0066] S44, repeating steps S42 to S43, using the target model to independently process the daily standardized data, and reconstructing the gridded meteorological element data containing a long time series corresponding to the current data reconstruction target.

[0067] In this embodiment, based on the hierarchical progressive structure of the above-mentioned target model, in step S42, exemplarily, when the data reconstruction target is to reconstruct gridded average air pressure data, the data that need to be standardized using the characteristic parameter standardization instance include the geographic location information of the meteorological station, and the gridded average temperature data reconstructed according to the geographic location information using the average temperature model.

[0068] Exemplarily, when the data reconstruction target is to reconstruct gridded sunshine duration data, the data that need to be standardized using the characteristic parameter standardization instance include the geographic location information of the weather station, the average relative humidity data reconstructed using the average relative humidity model, the average air pressure data reconstructed using the average air pressure model, and the average temperature data reconstructed using the average temperature model.

[0069] In this embodiment, in the above step S42, the method for constructing the characteristic parameter standardization instance is:

[0070] According to the data reconstruction target, the observation data of the corresponding characteristic parameters in the training meteorological station are read, merged into a two-dimensional characteristic matrix, and then standardized and saved as a characteristic parameter standardized instance.

[0071] Exemplarily, the two-dimensional feature matrix is ​​standardized using a Z-score standardization method.

[0072] In the embodiment of the present invention, the above-mentioned meteorological data reconstruction method further includes estimating and filling the observed data using the target model, and the method is specifically as follows:

[0073] When the daily observation data corresponding to the target variable of a meteorological station is missing, the characteristic parameter standardization instance is used to standardize the geographical location information of the meteorological station and the daily observation data corresponding to the characteristic parameters in all levels before the level where the target variable is located, and the target model with the optimal model parameters on the corresponding date is called to estimate and fill in the observation data corresponding to the target variable.

[0074] For example, for weather stations with missing average temperatures in the observation data for the day, the geographic location information of the missing weather stations is standardized using the characteristic parameter standardization instance, and then the optimal model parameters of the corresponding date are called to estimate and fill in the average temperature.

[0075] For example, for weather stations whose average air pressure is missing in the observation data for the day, the geographic location information and average temperature of the missing weather stations are standardized using the characteristic parameter standardization instance, and then the optimal model parameters of the corresponding date are called to estimate and fill in the average air pressure.

[0076] For example, for meteorological stations whose average relative humidity is missing in the observation data for the day, the geographic location information, average temperature, and average air pressure of the missing meteorological stations are standardized using pre-stored characteristic parameter standardization instances, and then the optimal model parameters of the corresponding date are called to estimate and fill in the average relative humidity.

[0077] For example, for meteorological stations with missing sunshine hours in the observation data for the day, the geographic location information, average temperature, average air pressure, and average relative humidity of the missing meteorological stations are standardized using pre-stored standardized instances, and then the optimal model parameters of the corresponding date are called to estimate and fill in the sunshine hours.

[0078] In an embodiment of the present invention, the above-mentioned meteorological data reconstruction method further includes:

[0079] The long sequence of meteorological element observation data observed by the verification meteorological station is used to evaluate the accuracy of the reconstructed meteorological element data of the target model.

[0080] In this embodiment, the evaluation index for the accuracy evaluation of the reconstructed meteorological element data includes the average error and correlation coefficient between the long-sequence meteorological element observation data and the corresponding reconstructed meteorological element data.

[0081] In the embodiment of the present invention, as the basic model of the above-mentioned target model, the BP neural network includes an input layer, a first hidden layer, a second hidden layer and an output layer;

[0082] The number of neurons in the input layer is the number of characteristic parameters corresponding to the data reconstruction target;

[0083] The first hidden layer contains 64 neurons, uses the ReLU activation function, and applies an L2 regularization coefficient of 0.001;

[0084] The second hidden layer contains 32 neurons, uses the ReLU activation function, and applies an L2 regularization coefficient of 0.001;

[0085] The output layer is a single neuron that outputs the target variable (average temperature, average air pressure, average relative humidity or sunshine hours).

[0086] In the embodiment of the present invention, Figure 2 As shown, a specific application example of the above method is provided.

[0087] in, Figure 2 (a) is a spatial distribution diagram of the study area and the training and verification meteorological stations of an embodiment of the present invention; Figure 2 (b) The spatial distribution map of the gridded multi-year mean temperature, multi-year mean air pressure, multi-year mean relative humidity and multi-year mean sunshine duration reconstructed from 1971 to 2018; Figure 2 (c) Box plot of the average error and correlation coefficient between the long series observation data and the gridded data of 35 validation meteorological stations.

[0088] This example integrates observation data from 1,040 meteorological stations in the Yellow River Basin and its surrounding areas since 1970, including the latitude and longitude of the stations, altitude, and daily meteorological data (such as average temperature, atmospheric pressure, relative humidity, and sunshine hours). Among them, data from 1,005 meteorological stations are used to train the model, and data from the remaining 35 meteorological stations are used to evaluate the accuracy of the reconstructed products. The reconstructed products cover the time range from 1971 to 2018, with daily temporal resolution and 1 km spatial resolution. Figure 2 The experimental results shown indicate that the median average errors of the gridded products of average temperature, atmospheric pressure, average relative humidity and sunshine duration were -0.09°C, -0.29hPa, 0.31% and 0.03h, respectively, and the median correlation coefficients with the measured data were 1, 0.98, 0.94 and 0.93, respectively, indicating that the reconstructed products generated by the method of the present invention have significant advantages in quality and accuracy.

[0089] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. 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 a limitation on the present invention.

[0090] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A hierarchical and progressive meteorological element data reconstruction method based on BP neural network, It is characterized in that The following steps are involved: S1. Collect meteorological element data from meteorological stations in and around the study area; The meteorological element data include the average temperature, average air pressure, average relative humidity and sunshine hours observed daily; S2. Based on the geographical location information of meteorological stations in and around the study area, the meteorological stations are divided into verification meteorological stations and training meteorological stations; S3, reconstructing the target according to the data, determining the observation data of the corresponding characteristic parameters and target variables in the training weather station, and constructing and optimizing the corresponding target model based on the BP neural network according to the data; S4, according to the data reconstruction target, calling the optimized target model with the optimal model parameters to reconstruct the gridded meteorological element data; The target model includes a hierarchical and progressive average temperature model, an average air pressure model, an average relative humidity model and a sunshine duration model; wherein the characteristic parameters of the average temperature model include the geographical location information of the training meteorological station, and the target variable is the corresponding average temperature; the characteristic parameters of the average air pressure model include the geographical location information and the average temperature of the training meteorological station, and the target variable is the corresponding average air pressure; the characteristic parameters of the average relative humidity model include the geographical location information, the average temperature and the average air pressure of the training meteorological station, and the target variable is the corresponding average relative humidity; the characteristic parameters of the sunshine duration model include the geographical location information, the average temperature, the average air pressure and the average relative humidity of the training meteorological station, and the target variable is the corresponding sunshine duration.

2. The hierarchical and progressive meteorological element data reconstruction method based on BP neural network according to claim 1, It is characterized in that In step S2, the verified meteorological stations include meteorological stations in the study area and its demarcated surrounding areas, which fully cover the upstream, midstream and downstream areas of the basin and its surrounding areas, as well as meteorological stations within different longitudes and latitudes and altitudes, and meteorological stations in areas where the density of meteorological station distribution is greater than a set threshold; The training meteorological stations include the remaining meteorological stations within the study area and its designated surrounding areas except for the meteorological stations determined as verification meteorological stations.

3. The hierarchical and progressive meteorological element data reconstruction method based on BP neural network according to claim 1, It is characterized in that The step S3 is specifically as follows: S1. Determine the characteristic parameters of the target model and their corresponding target variables; S2. Input the observed data of feature parameters and target variables into the constructed BP neural network, and use the Adam optimizer and mean square error loss function to optimize the BP neural network, save the model parameters with the best performance every day, and obtain the corresponding target model.

4. The hierarchical and progressive meteorological element data reconstruction method based on BP neural network according to claim 3, It is characterized in that The step S4 is specifically as follows: S41, determining the current data reconstruction target, and extracting the geographical location information in the high-resolution DEM; S42, the geographic location information is combined with the gridded meteorological element data reconstructed by the target model in all levels before the level where the current data reconstruction target is located, and the characteristic parameter standardization instance corresponding to the current data reconstruction target is used to perform standardization processing on it; S43, according to the current data reconstruction target, calling the target model corresponding to the model parameters with the best performance on that day to process the standardized data, and generating target reconstruction data consistent with the DEM resolution; S44, repeating steps S42 to S43, using the target model to independently process the daily standardized data, and reconstructing the gridded meteorological element data containing a long time series corresponding to the current data reconstruction target.

5. The hierarchical and progressive meteorological element data reconstruction method based on BP neural network according to claim 4, It is characterized in that The construction method of the characteristic parameter standardization instance is: According to the data reconstruction target, the observation data of the corresponding characteristic parameters in the training meteorological station are read, merged into a two-dimensional characteristic matrix, and then standardized and saved as a characteristic parameter standardized instance.

6. The hierarchical and progressive meteorological element data reconstruction method based on BP neural network according to claim 5, It is characterized in that The method also includes estimating and filling in the observed data, and the specific method is as follows: When the daily observation data corresponding to the target variable of a meteorological station is missing, the characteristic parameter standardization instance is used to standardize the geographical location information of the meteorological station and the daily observation data corresponding to the characteristic parameters in all levels before the level where the target variable is located, and the target model with the optimal model parameters on the corresponding date is called to estimate and fill in the observation data corresponding to the target variable.

7. The hierarchical and progressive meteorological element data reconstruction method based on BP neural network according to claim 1, It is characterized in that The method further comprises: The long sequence of meteorological element observation data observed by the verification meteorological station is used to evaluate the accuracy of the reconstructed meteorological element data of the target model.

8. The hierarchical and progressive meteorological element data reconstruction method based on BP neural network according to claim 7, It is characterized in that The evaluation indicators for the accuracy assessment of the reconstructed meteorological element data include the average error and correlation coefficient between the long-sequence meteorological element observation data and the corresponding reconstructed meteorological element data.

9. The hierarchical and progressive meteorological element data reconstruction method based on BP neural network according to claim 1, It is characterized in that The BP neural network includes an input layer, a first hidden layer, a second hidden layer and an output layer; The number of neurons in the input layer is the number of characteristic parameters corresponding to the data reconstruction target; The first hidden layer contains 64 neurons, uses the ReLU activation function, and applies an L2 regularization coefficient of 0.001; The second hidden layer contains 32 neurons, uses the ReLU activation function, and applies an L2 regularization coefficient of 0.001; The output layer is a single neuron that outputs the target variable.

Images