Wind and light resource endowment quantitative evaluation method and system under karst landform

Abstract

The invention discloses a wind and light resource endowment quantitative evaluation method and system under karst landform, and relates to the field of wind and light resource endowment evaluation, and the method comprises the steps: collecting geographical environment data in a karst landform region, and carrying out the data preprocessing of the geographical environment data, the geographical environment data comprising wind and light data and landform data; the number of hillsides is counted according to the landform data, karst weight factors are determined, the Euclidean distance between stations is calculated, an Euclidean adjacency matrix is generated, and weighting and calculation are carried out; constructing a karst space-time diagram convolutional network model, inputting the weighted sum matrix into the karst space-time diagram convolutional network model, and performing network training; and grading the wind and light predicted values, carrying out weighted sum processing according to karst weight factors, and carrying out convolution operation on data after weighted sum processing to obtain a wind and light resource endowment quantitative evaluation result. The problem that meteorological elements are incompatible with the landform environment is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of wind and solar resource endowment assessment, and in particular to a method and system for quantitatively assessing wind and solar resource endowment in karst landforms. Background Art

[0002] Guizhou Province is one of the regions with the most developed karst landforms in China. Its unique geological environment has an important impact on the distribution of wind and solar resources, the amount of reserves converted into electricity, and the protection of the ecological environment. Therefore, it is very necessary to consider the karst weight factor in the resource endowment assessment of Guizhou Province. The karst weight factor is a quantitative indicator obtained by a certain weight calculation method based on the altitude of the karst landform. This indicator can reflect the degree of influence of karst landforms on resource endowment and provide important data support for subsequent deep learning network training.

[0003] Karst Spatiotemporal Graph Convolution (KSTGCN) is a deep learning model that combines graph convolutional networks and spatiotemporal convolutional networks. In the resource endowment assessment of Guizhou Province, the model is able to process data with graph structure and spatiotemporal dependency, such as the spatial distribution of karst landforms and the temporal changes of meteorological data. Specifically, the KSTGCN model first extracts the spatial characteristics of karst landforms through a graph convolutional network, including the spatial relationship between different regions, the complexity of topography, etc., and then captures the temporal variation characteristics of meteorological data through a spatiotemporal convolutional network, such as the changing trends of wind speed, light radiation, etc. over time. Finally, the spatial features and temporal features are integrated to obtain a deep learning model that can fully reflect the resource endowment characteristics of Guizhou Province.

[0004] After deep learning network training, a model that can predict the resource endowment of Guizhou Province can be obtained. The model can predict the resource endowment of different regions in Guizhou Province, including the richness and distribution of wind energy resources and light energy resources, based on the input karst weight factor and meteorological data. Through resource endowment assessment, an important scientific basis can be provided for the sustainable development of Guizhou Province. The government can formulate reasonable resource development plans, ecological and environmental protection strategies and other policy measures based on the assessment results to promote the sustainable development of Guizhou Province's economy and society. At the same time, enterprises can also choose appropriate investment areas and development directions based on the assessment results to achieve a win-win situation of economic and social benefits. Summary of the invention

[0005] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0006] In view of the above existing problems, the present invention is proposed.

[0007] Therefore, the present invention provides a method and system for quantitatively evaluating the endowment of wind and light resources in karst landforms, which can solve the problems mentioned in the background technology.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0009] In a first aspect, the present invention provides a method for quantitatively evaluating the endowment of wind and light resources in karst landforms, which comprises collecting geographical environment data in a karst landform area and performing data preprocessing on the geographical environment data, wherein the geographical environment data comprises wind and light data and landform data;

[0010] Counting the number of hillsides and determining karst weight factors according to the geomorphic data, calculating the Euclidean distances between the sites, generating a Euclidean adjacency matrix, and performing weighted sum calculations to generate a weighted sum matrix;

[0011] Constructing a karst space-time graph convolutional network model, inputting the weighted sum matrix into the karst space-time graph convolutional network model, and performing network training to obtain wind and solar prediction values ​​and calculate various indicators of wind and solar resources;

[0012] The wind and solar power forecast values ​​are graded, weighted and processed according to the karst weight factor, and a convolution operation is performed on the weighted and summed data to obtain a quantitative assessment result of the wind and solar power resource endowment.

[0013] As a preferred solution of the method for quantitatively evaluating the wind and light resource endowment under the karst landform described in the present invention, the wind and light data include u component data of 10m wind, v component data of 10m wind, surface solar radiation data and surface net solar radiation data;

[0014] The landform data includes DEM digital elevation data, longitude and latitude information data, terrain slope data, aspect data, karst distribution data, soil type data, geological structure data and vegetation coverage data;

[0015] The data preprocessing includes missing value processing and Z-Score normalization;

[0016] The missing value processing includes performing missing value processing on the geographic environment data, including determining whether there are missing values, and filling the missing values;

[0017] If the missing values ​​in the data set are missing for a whole day of observations, the geographical environment data are filled with the improved adjacent mean;

[0018] The Z-Score normalization includes performing Z-Score normalization on the geographic environment data.

[0019] As a preferred solution of the method for quantitatively evaluating the wind and light resource endowment under karst landforms described in the present invention, the statistical number of hillsides includes:

[0020] The DEM digital elevation data is processed, and the areas with an altitude higher than 100m are defined as having mountains. On the regular grid of longitude and latitude coordinates, each grid is marked with 1 if there is a mountain and 0 if there is no mountain. The karst weight factor W'=[ω' ij ], the weight factor calculation formula is as follows:

[0021]

[0022] Among them, h′ ij is the relative height of the grid area in row i and column j, h ij is the altitude in the grid area of ​​row i and column j, is the average altitude in the entire grid area;

[0023] The generating of the Euclidean adjacency matrix comprises:

[0024] The Euclidean distance calculation is used to obtain the Euclidean adjacency matrix, based on the Euclidean distance d between coordinates i and j. ij , we get the Euclidean adjacency matrix W = [ω ij ], the calculation formula is as follows:

[0025]

[0026] Among them, r is the scale factor, dmin is the threshold distance when the weight is zero;

[0027] The Euclidean adjacency matrix and the Karst weight factor are weighted and the weighted sum matrix is The calculation formula is as follows:

[0028]

[0029] Among them, λ 1 is the karst weight matrix weight factor, λ 2 is the Euclidean adjacency matrix weight factor.

[0030] As a preferred solution of the method for quantitatively evaluating the wind and light resource endowment under karst landforms described in the present invention, wherein: the construction of the karst spatiotemporal graph convolutional network model includes:

[0031] Construct L identical KSTGCN layers, each of which contains an independent part and a shared part;

[0032] The blocks of the shared parts of different layers share the same set of parameters, and the blocks of the independent parts of different layers retain different sets of parameters. Each shared part and independent part has three different blocks, which are used to process the hidden states in the spatial, temporal and feature dimensions respectively;

[0033] When there are six blocks in a layer, a fusion block is used to integrate the outputs of the six blocks and keep the size of the hidden state consistent;

[0034] For the KSTGCN network model, is the input hidden vector of the l-th layer KSTGCN, and the input data is converted into H through the fully connected layer 0 ,Right now:

[0035] H 0 =W 0 *X t-P+1:t +b 0

[0036] Among them, W and b are adaptive learning parameters;

[0037] The input hidden state is processed by three blocks in separate parts:

[0038] H l,S =f l,S (H l )

[0039] H l,T =f l,T (H l )

[0040] H l,F =f l,F (H l )

[0041] Among them, f l,S 、f l,T and f l,F are the spatial block, temporal block and feature block of the lth layer of the independent part, respectively, and H l,S is the output vector of the lth layer of the independent part space layer, H l,T is the output vector of the lth layer of the independent part time layer, H l,F is the output vector of the lth layer of the independent part feature layer, H l is the input hidden vector of the lth layer;

[0042] The hidden state is also handled by the shared part, namely:

[0043] H l,S′ =f S (H l )

[0044] H l,T′ =f T (H l )

[0045] H l,F′ =f F (H l )

[0046] Among them, f S 、f T and f F are the space block, time block and feature block of the shared part respectively, H l,S is the output vector of the lth layer of the shared spatial layer, H l,T is the output vector of the lth layer of the shared part time layer, H l,F is the output vector of the lth layer of the shared feature layer;

[0047] The hidden states of the six blocks are fused, and the fusion formula is as follows:

[0048] H l′ =W l，f *[H l，S ;H l，T ;H l，F ;H l，S′ ;H l，T′ ;H l，F′ ]+b l，f

[0049] Among them, H l′ is the fusion output hidden vector, W l,f and b l,f is the hyperparameter of the feature layer l, [;] is the cascade operation;

[0050] After each layer, the input of the next layer is obtained through the residual link, that is:

[0051] H l+1 =Batch_Norm(H l +ReLU(H l′ ))

[0052] Among them, H l+1 It is the input of the next layer of the residual link;

[0053] Moreover, the hidden state of each layer is jump-connected to the output layer for final prediction, namely:

[0054]

[0055] in, is the final prediction result, t is the current time, and Q is the predicted future time.

[0056] As a preferred solution of the method for quantitatively evaluating the wind and light resource endowment under karst landforms described in the present invention, the reorganization of the adjacent tensors includes:

[0057] Define a specific tensor and approximate it by multiplying a core tensor and a factor matrix for each modality, namely:

[0058] Adjacency tensor Calculate four matrices And it is automatically updated, the calculation formula is as follows:

[0059] A = Softmax(ReLU(E k × 1 E t × 2 E s × 3 E e ))

[0060] Among them, × 1 、× 2 、× 3 is the tensor-matrix multiplication in the i-th dimension, is the Nt time period, is the information of the starting node, is the parameter of the end node, is the unit matrix;

[0061] Build the temporal relationship between input time slots as a dynamic graph

[0062] For the temporal graph convolution layer, the temporal graph convolution block is constructed according to A φ(t) Aggregate information into time slots, A φ(t) The (i, j) entry of is the impact strength of node j on node i. The temporal convolution on the l-layer independent partial dynamic graph is defined as:

[0063]

[0064] Where P is the input time slot, is a tensor-matrix;

[0065] In the shared part of layer l, the time block is defined as:

[0066]

[0067] in, is the temporal relationship matrix shared by different layers;

[0068] The spatial graph convolution layer consists of two mix-hop layers, which are used to process the net inflow and outflow information through each node respectively;

[0069] The net inflow information is obtained by adding the outputs of the two hybrid hop propagation layers;

[0070] The output of each layer in the L layer will be retained and used as the output result of KTSGCN after jump connection. The hybrid jump includes information propagation and information selection;

[0071] The information propagation includes the process of transforming the input data to the next step, that is, the process of transforming H(1) into H(2), ..., H(k-1) into H(k), and the formula is as follows:

[0072]

[0073] in, is the Karst weight factor, k is the number of iterations within the hybrid jump propagation layer, β is a hyperparameter used to control the ratio of keeping the original state of the root node, H in is the input hidden state of the previous layer output;

[0074] Information selection includes selecting all the outputs in the obtained H layer to obtain the final result. The calculation formula is as follows:

[0075]

[0076] Among them, K is the propagation depth, H out Represents the output hidden state of the current layer; H (k) is the output hidden state of the kth layer;

[0077] A hybrid hop propagation layer represents the difference between two consecutive hops, and the specific formula is as follows:

[0078] H out =Δ(H (2) ,H (1) )=H 2 -H 1

[0079] The final H out The output of the spatial graph convolution layer in the KTSGCN network model will be fused with the output of other layers. The fused output will be used as the output of the first layer of the KTSGCN network model, and so on for each layer below.

[0080] For the feature dimension layer, the feature block of each layer is implemented as a fully connected layer in the feature dimension, and huberloss is used as the loss function, that is,

[0081]

[0082] Among them, δ is the hyperparameter of balanced square error, ι is the average loss, is the error of a single value, N is the spatial dimension, and Q is the time dimension;

[0083] Select input X at different time t by mapping Φ(t) t-P+1:t , select a different time map

[0084] Training the Karst weight matrix weight factor λ 1 and the Euclidean adjacency matrix weight factor λ 2 And output the predicted value of scenery through the KSTGCN network model to obtain the matrix A.

[0085] As a preferred solution of the method for quantitatively evaluating the wind and solar resources endowment under karst landforms described in the present invention, the various indicators for calculating wind and solar resources include:

[0086] The wind and light prediction values ​​include wind speed and light radiation prediction values;

[0087] Based on the wind and solar forecast values, the forecast values ​​of wind shear, wind power density, turbulence intensity, sunshine hours, direct radiation and diffuse radiation are calculated, and a matrix is ​​generated according to each indicator, collectively referred to as Specifically include:

[0088] By calculating the logarithmic vertical gradient of wind speed, the calculation formula is as follows:

[0089]

[0090] Where α is the wind shear coefficient, z 1 and z 2 are two different heights, v 1 and v 2 They are z 1 and z 2 Wind speed at two altitudes;

[0091] Turbulence intensity refers to the degree of change in wind speed fluctuations caused by vortices and turbulent structures in the airflow relative to the average wind speed. The calculation formula is as follows:

[0092]

[0093] Where TI is the turbulence intensity, σ is the standard deviation of wind speed, is the average wind speed;

[0094] Wind power density describes the wind power per unit area and determines the optimal number of wind turbines. The calculation formula is as follows:

[0095]

[0096] Where P is the wind power density, ρ is the air density, is the average wind speed;

[0097] Sunshine hours refers to the time that sunlight directly hits the earth's surface in a day. It is used to describe the sunshine conditions in a certain area or a specific period of time. The calculation formula is as follows:

[0098]

[0099] Among them, S' is the sunshine hours, h is the solar altitude angle;

[0100] The relationship between the attenuation of solar radiation after passing through the atmosphere and the atmospheric transparency coefficient and the mass of the atmosphere passing through is expressed by the Bouguer formula, that is:

[0101] I=I 0 ×p×m

[0102] Among them, I is the solar radiation intensity reaching the ground, I0 is the solar constant, p is the air transparency coefficient, and m is the atmospheric mass number;

[0103] The intensity of scattered radiation is also related to the solar altitude angle and atmospheric transparency. When the solar altitude angle increases, the direct radiation reaching the near-ground layer increases, and the scattered radiation increases.

[0104] Similarly, when the solar altitude angle decreases, the scattered radiation weakens. The scattered radiation is equal to the total solar radiation minus the direct radiation. The calculation formula is as follows:

[0105] Q′=QI

[0106] Among them, Q' is diffuse radiation, Q is total solar radiation, and I is direct radiation.

[0107] As a preferred solution of the method for quantitatively evaluating the wind and light resource endowment under karst landforms described in the present invention, the grading includes:

[0108] Narrow the prediction results as a whole, and reduce the minimum to maximum values ​​of each indicator to the range of 0 to 7;

[0109] And divided into seven levels, the division criteria include: when the prediction result is greater than or equal to 0 and less than 1, it is set to the first level; when the prediction result is greater than or equal to 1 and less than 2, it is set to the second level; when the prediction result is greater than or equal to 2 and less than 3, it is set to the third level; when the prediction result is greater than or equal to 3 and less than 4, it is set to the fourth level; when the prediction result is greater than or equal to 4 and less than 5, it is set to the fifth level; when the prediction result is greater than or equal to 5 and less than 6, it is set to the sixth level; when the prediction result is greater than or equal to 6 and less than or equal to 7, it is set to the seventh level, and converted into a matrix

[0110] The prediction results are weighted and added with the karst weight factor, and a matrix reflecting the impact of the geographical environment on the actual richness of scenery resources is obtained based on the karst weight factor. The calculation formula is as follows:

[0111]

[0112] Where A is the karst weight factor, is the prediction result after the classification standard. is the weight coefficient of A, for The weight coefficient of

[0113] After filling the edge of the weighted sum matrix and performing convolution operations, the matrix is ​​visualized.

[0114] Pair Matrix The Karst matrix filling method is performed to ensure that the output matrix has the same size as the input matrix when performing subsequent convolution operations. The formula is as follows, from top to bottom, it is the center area of ​​the matrix, the upper edge, the lower edge, the left edge, and the right edge:

[0115]

[0116] Among them, 0≤i<n, 0≤j<n, is the filled matrix, n×n is size;

[0117] The values ​​of the top, bottom, left, and right corners are directly obtained from the original matrix The calculation formula is as follows: the upper left corner, the upper right corner, the lower left corner, and the lower right corner:

[0118]

[0119] From the filled matrix Select 9 data of 3×3 and perform convolution operation to generate a new matrix with more endowment meaning

[0120] The convolution kernel slides on the input matrix. Each time it slides, the 9 elements of the convolution kernel are multiplied element by element with the 9 elements at the corresponding positions on the input matrix, and the products are added to obtain an output value.

[0121] exist Repeat above, Each data point is represented by The nine adjacent data points in are calculated using specific weights, and the calculation formula is as follows:

[0122]

[0123] Among them, i and j are the rows and columns corresponding to the matrix, m and n are the indices of the convolution kernel, and K is the convolution kernel.

[0124] In a second aspect, the present invention provides a quantified evaluation system for wind and solar resource endowment under karst landform, which comprises: an acquisition preprocessing module, a matrix generation and calculation module, a model building and training module, and a result evaluation convolution module;

[0125] The acquisition and preprocessing module is used to acquire geographical environment data in the karst landform area and perform data preprocessing on the geographical environment data, wherein the geographical environment data includes scenery data and landform data;

[0126] The matrix generation and calculation module is used to count the number of hillsides and determine the karst weight factor according to the geomorphic data, calculate the Euclidean distance between each site, generate a Euclidean adjacency matrix, and perform weighted sum calculation to generate a weighted sum matrix;

[0127] The model building training module is used to build a karst spatiotemporal graph convolutional network model, input the weighted sum matrix into the spatial graph convolutional layer of the karst spatiotemporal graph convolutional network model, and perform network training through the temporal graph convolutional layer and the feature layer to obtain the wind and solar prediction value and calculate various indicators of wind and solar resources;

[0128] The result evaluation convolution module is used to classify the wind and solar power prediction values, perform weighted sum processing according to the karst weight factor, and perform a convolution operation on the weighted sum data to obtain a quantitative evaluation result of the wind and solar power resource endowment.

[0129] In a third aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the processor executes the computer program, it implements the steps of a method for quantitatively evaluating the endowment of wind and light resources under karst landforms.

[0130] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, the steps of a method for quantitatively evaluating the endowment of wind and solar resources in karst landforms are implemented.

[0131] Compared with the prior art, the beneficial effects of the present invention are that the present invention fully considers the particularity of karst landform and uses karst weight factor to reflect the degree of influence of karst landform on wind and solar resource endowment; the present invention adopts karst spatiotemporal graph convolution method, in which the hybrid jump propagation layer of the spatial graph convolution block integrates the karst weight factor and the Euclidean adjacency matrix to predict the wind and solar resource value and avoid the problem of incompatibility between meteorological elements and landform environment; the present invention proposes a weighted convolution based on karst landform, which can accurately evaluate the wind and solar resource endowment in the region under the influence of karst landform in the surrounding area. BRIEF DESCRIPTION OF THE DRAWINGS

[0132] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 creative work.

[0133] Figure 1 A method flow chart of a method and system for quantitatively evaluating wind and solar resource endowment in karst landforms provided by an embodiment of the present invention;

[0134] Figure 2 An internal structural diagram of a computer device of a method and system for quantitatively evaluating wind and solar resource endowment in karst landforms provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0135] In order to make the above-mentioned purposes, features and advantages of the present invention more understandable, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0136] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0137] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0138] Example 1

[0139] Reference Figure 1-Figure 2 , which is the first embodiment of the present invention, and provides a method for quantitatively evaluating the endowment of wind and solar resources in karst landforms, comprising:

[0140] Figure 1 A method flow chart of a method and system for quantitatively evaluating the endowment of wind and solar resources in karst landforms is shown, including:

[0141] S1: Collecting geographical environment data in the karst landform area and performing data preprocessing on the collected data, wherein the geographical environment data includes scenery data and landform data;

[0142] It should be noted that karst landform is a symbolic landform in Guizhou Province, but it does not mean that the whole province is karst landform. Therefore, the area located in the central part of Guizhou Province was selected, with a geographical location of 105°~106° east longitude and 26°~27° north latitude. It is a typical karst plateau landform area, and karst landform accounts for a relatively large proportion.

[0143] Furthermore, the wind and sunlight data under the karst landform in Guizhou Province are monthly reanalysis data, including the u component of 10 m wind, the v component of 10 m wind, surface solar radiation, and surface net solar radiation. The temporal resolution of the data is 1 hour, and the spatial resolution is 0.25° × 0.25°.

[0144] Furthermore, data preprocessing includes missing value processing and Z-Score normalization;

[0145] The missing value processing of the wind and solar data includes determining whether there are missing values ​​and filling the missing values. If the missing values ​​in the data set are missing for a whole day, the wind and solar data are filled with the improved adjacent mean. The filling calculation formula is as follows:

[0146]

[0147] Among them, x t is the missing fill value at the tth time step, x t-τ 、x t-1 are the values ​​of the τth and 1st time steps before the missing value, respectively. τ represents a cycle, i.e., 96 time steps per day. For any x t+iτIf it is not a missing value, let n be equal to the minimum value of i, and k be the weight coefficient of the adjacent value, k∈(0,1];

[0148] Normalize the scenery data to Z-Score, and the calculation formula is as follows:

[0149]

[0150] Among them, z i For sample data x i The Z-Score value, x i is the original data value of the sample data, μ is the mean value of the sample data, σ is the standard deviation of the sample data, and n is the number of sample data.

[0151] S2: Based on the longitude and latitude information of the collected geographic environment data, the number of hillsides is counted and the karst weight factor is determined, the Euclidean distance between each site is calculated, and the Euclidean adjacency matrix is ​​generated, and weighted and processed at the same time.

[0152] Furthermore, the DEM digital elevation data was processed and the areas with an altitude higher than 100 m were defined as having mountains. On the regular grid of longitude and latitude coordinates, each grid was marked with 1 if there was a mountain and 0 if there was no mountain. The karst weight factor W'=[ω' ij ], the weight factor calculation formula is as follows:

[0153]

[0154] Among them, h i ' j is the relative height of the grid area in row i and column j, h ij is the altitude in the grid area of ​​row i and column j, is the average altitude of the entire grid area.

[0155] Furthermore, generating the Euclidean adjacency matrix includes obtaining the Euclidean adjacency matrix by calculating the Euclidean distance, and calculating the Euclidean distance d between the coordinates i and j. ij , we get the Euclidean adjacency matrix W = [ω ij ], the calculation formula is as follows:

[0156]

[0157] Among them, r is the scale factor, d min is the threshold distance when the weight is zero;

[0158] The Euclidean adjacency matrix and the Karst weight factor are weighted and the weighted sum matrix is: The calculation formula is as follows:

[0159]

[0160] Among them, λ 1 is the karst weight matrix weight factor, λ 2 is the Euclidean adjacency matrix weight factor;

[0161] Furthermore, the initial karst weight matrix weight factor λ in the present invention is 1 and the initial Euclidean adjacency matrix weight factor λ 2 Take 0.3 and 0.7 respectively.

[0162] S3: Construct the Karst Spatiotemporal Graph Convolution (KSTGCN) network model, input the weighted sum matrix into the spatial graph convolution layer of the model, and train the network through the temporal graph convolution layer and feature layer to obtain the predicted values ​​of wind speed and light radiation, and calculate various indicators of wind and solar resources;

[0163] Furthermore, the predicted values ​​of various indicators such as wind shear, wind power density, turbulence intensity, sunshine hours, direct radiation and scattered radiation are calculated based on the predicted values;

[0164] Furthermore, the KSTGCN network model consists of the same L layers, and each KSTGCN layer contains an independent part and a shared part;

[0165] The blocks in the shared part of different layers share the same set of parameters, and the blocks in the independent part of different layers retain different sets of parameters. There are three different blocks in the shared part and the independent part, which are used to process the hidden states in the spatial, temporal and feature dimensions respectively;

[0166] When there are 6 blocks in a layer, a fusion block is used to integrate the outputs of the 6 blocks and keep the size of the hidden state consistent;

[0167] For the KSTGCN network model, is the input hidden vector of the lth layer KSTGCN, and the input data is converted into H through the fully connected layer 0 :

[0168] H 0 =W 0 *X t-P+1:t +b 0

[0169] Among them, W and b are adaptive learning parameters;

[0170] The input hidden state is processed by three blocks in separate parts:

[0171] H l,S =f l,S (H l )

[0172] H l,T =f l,T (H l )

[0173] H l,F =f l,F (H l )

[0174] Among them, f l,S 、f l,T and f l,F are the spatial block, temporal block and feature block of the lth layer of the independent part, respectively, and H l,S is the output vector of the lth layer of the independent part space layer, H l,T is the output vector of the lth layer of the independent part time layer, H l,F is the output vector of the lth layer of the independent part feature layer, H l is the input hidden vector of the lth layer;

[0175] On the other hand, the hidden state is also handled by the shared part, which is:

[0176] H l,S′ =f S (H l )

[0177] H l,T′ =f T (H l )

[0178] H l,F′ =f F (H l )

[0179] Among them, f S 、f T and f F They are the shared spatial blocks, temporal blocks, and feature blocks, but they share the same parameter set between different layers. l,S is the output vector of the lth layer of the shared spatial layer, H l,T is the output vector of the lth layer of the shared part time layer, H l,F is the output vector of the lth layer of the shared feature layer;

[0180] Then the hidden states of these six blocks are fused, and the fusion formula is as follows:

[0181] H l′ =W l，f *[H l，S ;H l，T ;H l，F ;H l，S′ ;H l，T′ ;H l，F′]+b l，f

[0182] Among them, H l′ is the fusion output hidden vector, W l,f and b l,f is the hyperparameter of the feature layer l, [;] is the cascade operation;

[0183] After each layer, the input of the next layer is obtained through the residual link, that is:

[0184] H l+1 =Batch_Norm(H l +ReLU(H l′ ))

[0185] Among them, H l+1 It is the input of the next layer of the residual link;

[0186] Moreover, the hidden state of each layer is jump-connected to the output layer for final prediction, namely:

[0187]

[0188] in, is the final prediction result, t is the current time, and Q is the predicted future time.

[0189] Furthermore, the adjacency tensors are reorganized. Given a specific tensor, it is approximated by the product of a core tensor and a factor matrix of each modality, that is:

[0190] Adjacency tensor Calculate four matrices And it can be automatically updated, the calculation formula is as follows:

[0191] A = Softmax(ReLU(E k × 1 E t × 2 E s × 3 E e ))

[0192] Among them, × 1 、× 2 、× 3 is the tensor-matrix multiplication in the i-th dimension, is the Nt time period, is the information of the starting node, is the parameter of the end node, is the unit matrix;

[0193] Build the temporal relationship between input time slots as a dynamic graph

[0194] For the temporal graph convolutional layer, given the input X t-P+1:t and The temporal graph convolution block of the relation matrix is ​​based on A φ(t) Aggregate information into time slots, A φ(t) The (i, j) entry of A reflects the impact strength of node j on node i, where φ(t) is used to calculate the corresponding index. φ(t) is the dynamic graph selected by t. The temporal convolution on the l-layer independent partial dynamic graph can be defined as:

[0195]

[0196] Where P is the input time slot, It is tensor-matrix multiplication, which is different in different layers;

[0197] In the shared part of layer l, the time block is defined as:

[0198]

[0199] in, is the temporal relationship matrix, shared by different layers.

[0200] The spatial graph convolution layer consists of two mix-hop layers, which are used to process the inflow and outflow information through each node respectively;

[0201] The net inflow information is obtained by adding the outputs of two hybrid jump propagation layers. The output of each layer in the L layer will be retained and used as the output result of KTSGCN after jump connection. The hybrid jump includes two steps: information propagation and information selection. Information propagation is the process of transforming the input data to the next step, that is, the process of transforming H(1) into H(2), ..., H(k-1) into H(k). The formula is as follows:

[0202]

[0203] in, is the Karst weight factor, k is the number of iterations within the hybrid jump propagation layer, β is a hyperparameter used to control the ratio of keeping the original state of the root node, H in is the input hidden state of the previous layer output;

[0204] The information selection in the hybrid jump propagation layer selects all the outputs of the H layer to obtain the final result. The calculation formula is as follows:

[0205]

[0206] Among them, K is the propagation depth, H out Represents the output hidden state of the current layer; H (k) is the output hidden state of the kth layer;

[0207] First propagate information horizontally, then select information vertically. A hybrid jump propagation layer represents the difference between two consecutive jumps. The specific formula is as follows:

[0208] H out =Δ(H (2) ,H (1) )=H 2 -H 1

[0209] The final H out The output of the spatial graph convolution layer in the KTSGCN network model will be fused with the output of other layers. The fused output will be used as the output of the first layer of the KTSGCN network model, and so on for each subsequent layer.

[0210] For the feature dimension layer, the feature block of each layer is implemented as a fully connected layer on the feature dimension, using huberloss as the loss function, that is,

[0211]

[0212] Among them, δ is the hyperparameter of balanced square error, ι is the average loss, is the error of a single value, N is the spatial dimension, and Q is the time dimension;

[0213] Furthermore, in the KSTGCN network model, all temporal relationship matrices are generated by the same dynamic graph generator, and the temporal graphs used are dynamic. The proposed model selects the input X at different times t by mapping Φ(t) t-P+1:t , select a different time map

[0214] Training the Karst weight matrix weight factor λ 1 and the Euclidean adjacency matrix weight factor λ 2 And output the predicted value of wind and solar power through the KSTGCN network model to obtain the matrix A;

[0215] Furthermore, the predicted values ​​of wind shear, turbulence intensity, wind power density, sunshine hours, direct radiation, and diffuse radiation are derived based on the predicted values ​​of wind speed and light radiation. Each indicator generates a matrix, collectively referred to as The following is the derivation process of each indicator:

[0216] Wind shear describes the change in wind speed with height and is calculated by taking the logarithmic vertical gradient of wind speed using the following formula:

[0217]

[0218] Where α is the wind shear coefficient, z 1 and z 2 are two different heights, v 1 and v 2 are the wind speeds at these two heights respectively;

[0219] Turbulence intensity refers to the degree of change in wind speed fluctuations caused by vortices and turbulent structures in the airflow relative to the average wind speed. It is an important parameter used to describe the degree of turbulence in the wind field. It is usually expressed as a percentage, expressed as the percentage of wind speed fluctuations relative to the average wind speed. The calculation formula is as follows:

[0220]

[0221] Where TI is the turbulence intensity, σ is the standard deviation of wind speed, is the average wind speed.

[0222] Wind power density describes the wind power per unit area. It is used to describe the wind power generation potential of the selected site and helps to evaluate the performance of wind turbines and determine the optimal number of wind turbines. The calculation formula is as follows:

[0223]

[0224] Where P is the wind power density, ρ is the air density, is the average wind speed;

[0225] Sunshine hours refers to the time that sunlight directly hits the earth's surface in a day. It is usually expressed in hours and is used to describe the sunshine conditions in a certain area or a specific time period. The calculation formula is as follows:

[0226]

[0227] Among them, S' is the sunshine hours and h is the solar altitude angle.

[0228] For direct radiation, the smaller the solar altitude angle, the larger the area where the same amount of solar radiation is distributed, and thus the smaller the solar radiation received per unit area of ​​the surface. The relationship between the attenuation of solar radiation after passing through the atmosphere and the atmospheric transparency coefficient and the mass of the atmosphere passing through can be expressed by the Bouguer formula, namely:

[0229] I=I 0 ×p×m

[0230] Among them, I is the intensity of solar radiation reaching the ground, I0 is the solar constant, p is the air transparency coefficient, and m is the atmospheric mass number. If the atmospheric transparency coefficient is constant and the atmospheric mass number increases in an arithmetic progression, then the solar radiation that passes through the atmosphere and reaches the ground decreases in a geometric progression.

[0231] The strength of scattered radiation is also related to the solar altitude angle and atmospheric transparency. When the solar altitude angle increases, the direct radiation reaching the near-ground layer increases, and the scattered radiation also increases accordingly. On the contrary, when the solar altitude angle decreases, the scattered radiation is also weak. The scattered radiation is equal to the total solar radiation minus the direct radiation. The calculation formula is as follows:

[0232] Q′=QI

[0233] Among them, Q' is diffuse radiation, Q is total solar radiation, and I is direct radiation.

[0234] S4: The predicted values ​​are classified and linked with the karst weight factors for weighted sum processing. A 3×3 convolution operation is performed on the weighted sum data to finally obtain the quantitative assessment results of wind and solar resource endowment.

[0235] The prediction results are reduced as a whole, and the minimum to maximum values ​​of each indicator are reduced to the range of 0 to 7, and divided into seven levels according to the following standards: the first level (0 to 1), the second level (1 to 2), the third level (2 to 3), the fourth level (3 to 4), the fifth level (4 to 5), the sixth level (5 to 6), and the seventh level (6 to 7), and converted into a matrix

[0236] The prediction results are weighted and added with the karst weight factor. Under the influence of the karst weight factor, a matrix reflecting the impact of the geographical environment on the actual richness of scenery resources is obtained: The formula is as follows:

[0237]

[0238] Where A is the karst weight factor, is the prediction result after the classification standard. is the weight coefficient of A, for The weight coefficient of .

[0239] After filling the edge of the weighted sum matrix and performing convolution operations, a visual display is performed;

[0240] Pair Matrix The Karst matrix filling method is performed to ensure that the output matrix has the same size as the input matrix when performing subsequent convolution operations. The formula is as follows, from top to bottom, it is the center area of ​​the matrix, the upper edge, the lower edge, the left edge, and the right edge:

[0241]

[0242] Among them, 0≤i<n, 0≤j<n, is the filled matrix, n×n is The size of the four corners is directly determined by the original matrix The values ​​of the four corners are copied from the formula, which is the upper left corner, upper right corner, lower left corner, and lower right corner:

[0243]

[0244] From the filled matrix Select 9 data of 3×3 and perform convolution operation to reflect the influence of the surrounding 8 data on the central data and generate a new matrix with more endowment significance.

[0245] The convolution kernel slides on the input matrix. Each time it slides, the 9 elements of the convolution kernel are multiplied element by element with the 9 elements at the corresponding positions on the input matrix, and then these products are added to get an output value.

[0246] exist Repeat above, Each data point is represented by The nine adjacent data points in are calculated by specific weights (i.e., elements of the convolution kernel), and the calculation formula is as follows:

[0247]

[0248] Among them, i and j are the rows and columns corresponding to the matrix, m and n are the indexes of the convolution kernel, K is the convolution kernel, and the index of K starts from 0. The index starts from i-1.

[0249] Furthermore, the matrix obtained for each indicator For visualization, insert Baidu Map as the underlying map and fill in the color of the landscape resource endowment assessment matrix at 105°~106° east longitude and 26°~27° north latitude. As the top-level graph, draw different colors corresponding to different standards of each indicator on the legend on the left.

[0250] The ERA5 wind speed data at 104°~105° east longitude and 26°~27° north latitude were selected to verify the site selection of the wind farm. There are three wind farms in this area, namely A, B and C. These three wind farms are all located in the resource-rich area in the wind resource endowment map, that is, the area with wind shear close to 0m / s, large wind power density and small turbulence coefficient, which proves that the wind speed resource endowment map is accurately evaluated; the ERA5 photovoltaic data at 105°~106° east longitude and 26°~27° north latitude were used to verify the site selection of the photovoltaic power station. There is a photovoltaic power station D in this area, which is located in the resource-rich area in the light resource endowment map, that is, the area with more sunshine hours, more direct radiation and more scattered radiation.

[0251] The ERA5 wind speed data at 104°~105° east longitude and 27°~28° north latitude were selected to visualize the quantitative assessment diagram of wind speed resource endowment, and the trend of wind speed from low to high was obtained. In this area, there are four wind speed stations E, F, G, and H, which are sorted according to the trend. The actual values ​​measured by the wind speed stations are 3.41m / s, 3.58m / s, 5.96m / s, and 6.16m / s, respectively, which is consistent with the trend of wind speed from low to high, proving that the model prediction is accurate.

[0252] Furthermore, the present embodiment also provides a quantified evaluation system for wind and solar resource endowment under karst landform, comprising: an acquisition preprocessing module, a matrix generation and calculation module, a model building and training module, and a result evaluation convolution module;

[0253] The acquisition and preprocessing module is used to acquire geographical environment data in the karst landform area and perform data preprocessing on the geographical environment data, wherein the geographical environment data includes scenery data and landform data;

[0254] The matrix generation and calculation module is used to count the number of hillsides and determine the karst weight factor according to the geomorphic data, calculate the Euclidean distance between each site, generate the Euclidean adjacency matrix, and perform weighted sum calculation to generate a weighted sum matrix;

[0255] The model building and training module is used to build a karst spatiotemporal graph convolutional network model, input the weighted sum matrix into the spatial graph convolutional layer of the karst spatiotemporal graph convolutional network model, and perform network training through the temporal graph convolutional layer and the feature layer to obtain the wind and solar prediction value and calculate various indicators of wind and solar resources;

[0256] The result evaluation convolution module is used to classify the wind and solar power prediction values, perform weighted sum processing according to the karst weight factor, and perform a convolution operation on the weighted sum data to obtain a quantitative evaluation result of the wind and solar power resource endowment.

[0257] This embodiment also provides a computer device, which may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 2 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a comprehensive evaluation method suitable for orderly access of large-scale distributed power sources is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse.

[0258] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, a method for quantitatively evaluating the endowment of wind and solar resources in karst landforms is implemented.

[0259] Example 2

[0260] Reference Figure 1 - Figure 2 , which is the second embodiment of the present invention, and this embodiment provides a method for quantitatively evaluating the endowment of wind and solar resources under karst landforms. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0261] Test preparation and detailed process

[0262] This embodiment aims to verify the effectiveness and innovation of a quantitative evaluation method for wind and solar resources under karst landform. The experimental site was selected in a typical karst landform area in Guizhou Province, China, and the geographical environment data in the area, including wind and solar data and landform data, were collected. The wind and solar data include 10m wind u component data, 10m wind v component data, surface solar radiation data and surface net solar radiation data; the landform data include DEM digital elevation data, longitude and latitude information data, terrain slope data, slope aspect data, karst distribution data, soil type data, geological structure data and vegetation coverage data.

[0263] First, the collected geographic environment data is preprocessed. The missing value processing adopts the improved adjacent mean filling method to fill the missing data of the whole day observation value. Then, the data is normalized by Z-Score to eliminate the dimension effect and enhance the comparability of the data.

[0264] Next, the number of hillsides is obtained based on the DEM digital elevation data, and the area with an altitude higher than 100m is defined as having mountains. On the regular grid of longitude and latitude coordinates, the area with mountains is marked as 1 and the area without mountains is marked as 0, and the karst weight factor is calculated;

[0265] Then, the Euclidean distance between each site is calculated to generate the Euclidean adjacency matrix, and the weighted sum is performed with the karst weight factor to generate the weighted sum matrix;

[0266] The weighted sum matrix is ​​input into the Karst Spatiotemporal Graph Convolutional Network model for training. The model uses L layers of the same KSTGCN, each layer contains an independent part and a shared part. The independent part processes the hidden state in the spatial, temporal and feature dimensions, and the shared part blocks share the same set of parameters.

[0267] Through the convolution operation, the wind and solar power prediction values ​​are classified into levels and weighted and processed according to the karst weight factor. Finally, the quantitative assessment results of wind and solar power resource endowment are obtained and visualized.

[0268] Table 1 Test data table

[0269]

[0270] Through the experimental data in the above table, we can clearly see the differences and regularities of the wind and light resource endowment under the karst landform. The following is a specific analysis:

[0271] The table shows that the values ​​of wind speed u and v vary significantly between different sites. The highest wind speed occurs at site D, where the wind speeds u and v are 2.5 m / s and 1.9 m / s, respectively. This is related to the altitude (170 m) and slope (22°) of site D. Higher altitudes and steeper slopes help to increase wind speeds.

[0272] Relationship between karst distribution and wind and solar resources: Karst distribution data shows the wind and solar resource endowment of each site. Site F has the highest karst distribution ratio (60%), and its wind speed and solar radiation values ​​are also high, at 2.4m / s and 335W / m 2 The particularity of karst landform is conducive to the increase of wind speed and the absorption of solar radiation.

[0273] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0274] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0275] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0276] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0277] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0278] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0279] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for quantitatively evaluating the endowment of wind and solar resources in karst landforms, characterized by: The method includes collecting geographical environment data in a karst landform area and performing data preprocessing on the geographical environment data, wherein the geographical environment data includes scenery data and landform data; Counting the number of hillsides and determining karst weight factors according to the geomorphic data, calculating the Euclidean distances between the sites, generating a Euclidean adjacency matrix, and performing weighted sum calculations to generate a weighted sum matrix; Constructing a karst space-time graph convolutional network model, inputting the weighted sum matrix into the karst space-time graph convolutional network model, and performing network training to obtain wind and solar prediction values ​​and calculate various indicators of wind and solar resources; The wind and solar power forecast values ​​are graded, weighted and processed according to the karst weight factor, and a convolution operation is performed on the weighted and summed data to obtain a quantitative assessment result of the wind and solar power resource endowment.

2. The method for quantitatively evaluating the endowment of wind and light resources in karst landforms according to claim 1, characterized in that: The wind and light data include u component data of 10m wind, v component data of 10m wind, surface solar radiation data and surface net solar radiation data; The landform data includes DEM digital elevation data, longitude and latitude information data, terrain slope data, aspect data, karst distribution data, soil type data, geological structure data and vegetation coverage data; The data preprocessing includes missing value processing and Z-Score normalization; The missing value processing includes performing missing value processing on the geographic environment data, including determining whether there are missing values, and filling the missing values; If the missing values ​​in the data set are missing for a whole day of observations, the geographical environment data are filled with the improved adjacent mean; The Z-Score normalization includes performing Z-Score normalization on the geographic environment data.

3. The method for quantitatively evaluating the endowment of wind and light resources in karst landforms according to claim 2, characterized in that: The statistical number of hillsides includes: The DEM digital elevation data was processed and the areas with an altitude higher than 100m were defined as having mountains. On the regular grid of longitude and latitude coordinates, each grid was marked with 1 if there was a mountain and 0 if there was no mountain. The karst weight factor W'=[ω' ij ], the weight factor calculation formula is as follows: Among them, h i ' j is the relative height of the grid area in row i and column j, h ij is the altitude in the grid area of ​​row i and column j, is the average altitude in the entire grid area; The generating of the Euclidean adjacency matrix comprises: The Euclidean distance calculation is used to obtain the Euclidean adjacency matrix, based on the Euclidean distance d between coordinates i and j. ij , we get the Euclidean adjacency matrix W = [ω ij ], the calculation formula is as follows: Among them, r is the scale factor, dmin is the threshold distance when the weight is zero; The Euclidean adjacency matrix and the Karst weight factor are weighted and the weighted sum matrix is The calculation formula is as follows: Among them, λ1 is the weight factor of the karst weight matrix, and λ2 is the weight factor of the Euclidean adjacency matrix.

4. The method for quantitatively evaluating the endowment of wind and light resources in karst landforms according to claim 3, characterized in that: The construction of the karst spatiotemporal graph convolutional network model includes: Construct L identical KSTGCN layers, each of which contains an independent part and a shared part; The blocks of the shared parts of different layers share the same set of parameters, and the blocks of the independent parts of different layers retain different sets of parameters. Each shared part and independent part has three different blocks, which are used to process the hidden states in the spatial, temporal and feature dimensions respectively; When there are six blocks in a layer, a fusion block is used to integrate the outputs of the six blocks and keep the size of the hidden state consistent; For the KSTGCN network model, is the input hidden vector of the l-th layer KSTGCN, and the input data is converted into H through the fully connected layer 0 ,Right now: H 0 =W 0 *X t-P+1:t +b 0 Among them, W and b are adaptive learning parameters; The input hidden state is processed by three blocks in separate parts: H l,S =f l,S (H l ) H l,T =f l,T (H l ) H l,F =f l,F (H l ) Among them, f l,S 、f l,T and f l,F are the spatial block, temporal block and feature block of the lth layer of the independent part, respectively, and H l,S is the output vector of the lth layer of the independent part space layer, H l,T is the output vector of the lth layer of the independent part time layer, H l,F is the output vector of the lth layer of the independent part feature layer, H l is the input hidden vector of the lth layer; The hidden state is also handled by the shared part, namely: H l,S′ =f S (H l ) H l,T′ =f T (H l ) H l,F′ =f F (H l ) Among them, f S 、f T and f F are the space block, time block and feature block of the shared part respectively, H l,S is the output vector of the lth layer of the shared spatial layer, H l,T is the output vector of the lth layer of the shared part time layer, H l,F is the output vector of the lth layer of the shared feature layer; The hidden states of the six blocks are fused, and the fusion formula is as follows: H l′ =W l，f *[H l，S ;H l，T ;H l，F ;H l，S′ ;H l，T' ;HF l，F′ ]+b l，f Among them, H l′ is the fusion output hidden vector, W l,f and b l,f is the hyperparameter of the feature layer l, [;] is the cascade operation; After each layer, the input of the next layer is obtained through the residual link, that is: H l+1 =Batch_Norm(H l +ReLU(H l′ )) Among them, H l+1 It is the input of the next layer of the residual link; Moreover, the hidden state of each layer is jump-connected to the output layer for final prediction, namely: in, is the final prediction result, t is the current time, and Q is the predicted future time.

5. The method for quantitatively evaluating the endowment of wind and light resources in karst landforms according to claim 4, characterized in that: The reorganization of the adjacent tensors includes: Define a specific tensor and approximate it by multiplying a core tensor and a factor matrix for each modality, namely: Adjacency tensor Calculate four matrices And it is automatically updated, the calculation formula is as follows: Among them, ×1, ×2, and ×3 are tensor-matrix multiplications on the i-th dimension, is the Nt time period, is the information of the starting node, is the parameter of the end node, is the unit matrix; Build the temporal relationship between input time slots as a dynamic graph For the temporal graph convolution layer, the temporal graph convolution block is constructed according to A φ(t) Aggregate information into time slots, A φ(t) The (i, j) entry of is the impact strength of node j on node i. The temporal convolution on the l-layer independent partial dynamic graph is defined as: Where P is the input time slot, is a tensor-matrix; In the shared part of layer l, the time block is defined as: in, is the temporal relationship matrix shared by different layers; The spatial graph convolution layer consists of two mix-hop layers, which are used to process the net inflow and outflow information through each node respectively; The net inflow information is obtained by adding the outputs of the two hybrid hop propagation layers; The output of each layer in the L layer will be retained and used as the output result of KTSGCN after jump connection. The hybrid jump includes information propagation and information selection; The information propagation includes the process of transforming the input data to the next step, that is, the process of transforming H(1) into H(2), ..., H(k-1) into H(k), and the formula is as follows: in, is the Karst weight factor, k is the number of iterations within the hybrid jump propagation layer, β is a hyperparameter used to control the ratio of keeping the original state of the root node, H in is the input hidden state of the previous layer output; Information selection includes selecting all the outputs in the obtained H layer to obtain the final result. The calculation formula is as follows: Among them, K is the propagation depth, H out Represents the output hidden state of the current layer; H (k) is the output hidden state of the kth layer; A hybrid hop propagation layer represents the difference between two consecutive hops, and the specific formula is as follows: H out =Δ(H (2) ,H (1) )=H 2 -H 1 The final H out The output of the spatial graph convolution layer in the KTSGCN network model will be fused with the output of other layers. The fused output will be used as the output of the first layer of the KTSGCN network model, and so on for each layer below. For the feature dimension layer, the feature block of each layer is implemented as a fully connected layer in the feature dimension, and huberloss is used as the loss function, that is, Among them, δ is the hyperparameter of balanced square error, ι is the average loss, is the error of a single value, N is the spatial dimension, and Q is the time dimension; Select input X at different time t by mapping Φ(t) t-P+1:t , select a different time map Train the karst weight matrix weight factor λ1 and the Euclidean adjacency matrix weight factor λ2, and output the scenery prediction value through the KSTGCN network model to obtain the matrix A.

6. The method for quantitatively evaluating the endowment of wind and light resources in karst landforms according to claim 5, characterized in that: The various indicators for calculating wind and solar resources include: The wind and light prediction values ​​include wind speed and light radiation prediction values; Based on the wind and solar forecast values, the forecast values ​​of wind shear, wind power density, turbulence intensity, sunshine hours, direct radiation and diffuse radiation are calculated, and a matrix is ​​generated according to each indicator, collectively referred to as Specifically include: By calculating the logarithmic vertical gradient of wind speed, the calculation formula is as follows: Among them, α is the wind shear coefficient, z1 and z2 are two different heights, v1 and v2 are the wind speeds at the two heights z1 and z2 respectively; Turbulence intensity refers to the degree of change in wind speed fluctuations caused by vortices and turbulent structures in the airflow relative to the average wind speed. The calculation formula is as follows: Where TI is the turbulence intensity, σ is the standard deviation of wind speed, is the average wind speed; Wind power density describes the wind power per unit area and determines the optimal number of wind turbines. The calculation formula is as follows: Where P is the wind power density, is the air density, is the average wind speed; Sunshine hours refers to the time in a day when sunlight directly hits the earth's surface. It is used to describe the sunshine conditions in a certain area or a specific period of time. The calculation formula is as follows: Among them, S' is the sunshine hours, h is the solar altitude angle; The relationship between the attenuation of solar radiation after passing through the atmosphere and the atmospheric transparency coefficient and the mass of the atmosphere passing through is expressed by the Bouguer formula, that is: I=I0×p×m Among them, I is the solar radiation intensity reaching the ground, I0 is the solar constant, p is the air transparency coefficient, and m is the atmospheric mass number; The intensity of scattered radiation is also related to the solar altitude angle and atmospheric transparency. When the solar altitude angle increases, the direct radiation reaching the near-ground layer increases, and the scattered radiation increases. Similarly, when the solar altitude angle decreases, the scattered radiation weakens. The scattered radiation is equal to the total solar radiation minus the direct radiation. The calculation formula is as follows: Q′=QI Among them, Q' is diffuse radiation, Q is total solar radiation, and I is direct radiation.

7. The method for quantitatively evaluating the endowment of wind and light resources in karst landforms according to claim 6, characterized in that: The grading includes: Narrow the prediction results as a whole, and reduce the minimum to maximum values ​​of each indicator to the range of 0 to 7; And divided into seven levels, the division criteria include: when the prediction result is greater than or equal to 0 and less than 1, it is set to the first level; when the prediction result is greater than or equal to 1 and less than 2, it is set to the second level; when the prediction result is greater than or equal to 2 and less than 3, it is set to the third level; when the prediction result is greater than or equal to 3 and less than 4, it is set to the fourth level; when the prediction result is greater than or equal to 4 and less than 5, it is set to the fifth level; when the prediction result is greater than or equal to 5 and less than 6, it is set to the sixth level; when the prediction result is greater than or equal to 6 and less than or equal to 7, it is set to the seventh level, and converted into a matrix The prediction results are weighted and added with the karst weight factor, and a matrix reflecting the impact of the geographical environment on the actual richness of scenery resources is obtained based on the karst weight factor. The calculation formula is as follows: Where A is the karst weight factor, is the prediction result after the classification standard. is the weight coefficient of A, for The weight coefficient of After filling the edge of the weighted sum matrix and performing convolution operations, the matrix is ​​visualized. Pair Matrix The Karst matrix filling method is performed to ensure that the output matrix has the same size as the input matrix when performing subsequent convolution operations. The formula is as follows, from top to bottom, it is the center area of ​​the matrix, the upper edge, the lower edge, the left edge, and the right edge: Among them, 0≤i<n, 0≤j<n, is the filled matrix, n×n is size; The values ​​of the top, bottom, left, and right corners are directly obtained from the original matrix The calculation formula is as follows: the upper left corner, the upper right corner, the lower left corner, and the lower right corner: From the filled matrix Select 9 data of 3×3 and perform convolution operation to generate a new matrix with more endowment meaning The convolution kernel slides on the input matrix. Each time it slides, the 9 elements of the convolution kernel are multiplied element by element with the 9 elements at the corresponding positions on the input matrix, and the products are added to obtain an output value. exist Repeat above, Each data point is represented by The 9 adjacent data points in are calculated by specific weights, and the calculation formula is as follows: Among them, i and j are the rows and columns corresponding to the matrix, m and n are the indices of the convolution kernel, and K is the convolution kernel.

8. A quantified evaluation system for wind and solar resources in karst landforms, based on the quantified evaluation method for wind and solar resources in karst landforms according to any one of claims 1 to 7, characterized in that: Including, acquisition preprocessing module, matrix generation calculation module, model building training module and result evaluation convolution module; The acquisition and preprocessing module is used to acquire geographical environment data in the karst landform area and perform data preprocessing on the geographical environment data, wherein the geographical environment data includes scenery data and landform data; The matrix generation and calculation module is used to count the number of hillsides and determine the karst weight factor according to the geomorphic data, calculate the Euclidean distance between each site, generate a Euclidean adjacency matrix, and perform weighted sum calculation to generate a weighted sum matrix; The model building training module is used to build a karst spatiotemporal graph convolutional network model, input the weighted sum matrix into the spatial graph convolutional layer of the karst spatiotemporal graph convolutional network model, and perform network training through the temporal graph convolutional layer and the feature layer to obtain the wind and solar prediction value and calculate various indicators of wind and solar resources; The result evaluation convolution module is used to classify the wind and solar power prediction values, perform weighted sum processing according to the karst weight factor, and perform a convolution operation on the weighted sum data to obtain a quantitative evaluation result of the wind and solar power resource endowment.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for quantitatively evaluating the wind and light resource endowment under karst landform described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for quantitatively evaluating the wind and light resource endowment under karst landform described in any one of claims 1 to 7 are implemented.

Images