Spatial interpolation method and device

Abstract

The invention provides a spatial interpolation method and device. The method comprises the following steps: acquiring actual measurement values of index objects actually measured by a plurality of sample points in a target area; performing interpolation on a predicted position in the target area according to the measured value by using an inverse distance weight interpolation method to obtain an inverse distance weight interpolation result of the predicted position; and taking the inverse distance weight interpolation result as a collaborative variable of a collaborative Kriging interpolation method, and performing secondary interpolation on the target area according to the measured value and the collaborative variable by using the collaborative Kriging interpolation method to obtain a predicted value of the index object.

Description

Technical Field

[0001] The embodiments of this specification relate to a regional scale interpolation method, and more particularly to a spatial interpolation method and device. Background Art

[0002] In the related art, commonly used spatial interpolation methods may include Inverse Distance Weight (IDW), Kriging, and Co-Kriging.

[0003] The interpolation result of the inverse distance weighted interpolation method is between the measured maximum and minimum values. Although it can retain the details to the greatest extent, it is prone to the "bull's eye" phenomenon.

[0004] The Kriging interpolation method uses spatial statistics as its theoretical basis. It can overcome the problem that errors in interpolation are difficult to analyze. It has high interpolation accuracy and strong uniqueness and extrapolation capabilities. However, the interpolation results are prone to smoothing.

[0005] Cokriging interpolation is an improved Kriging interpolation method. By considering the correlation between two or more random processes, more accurate spatial interpolation results can be obtained. However, the covariates required by the cokriging interpolation method are difficult to obtain, or it is difficult to obtain suitable covariates, which leads to the poor applicability of the cokriging interpolation method. Summary of the invention

[0006] The embodiments of this specification provide a spatial interpolation method and device.

[0007] According to a first aspect of an embodiment of this specification, a spatial interpolation method is provided, the method comprising:

[0008] Obtaining measured values ​​of the indicator object measured at a number of sample points located in the target area;

[0009] Using the inverse distance weighted interpolation method, interpolating the to-be-interpolated points in the target area according to the measured values, to obtain the inverse distance weighted interpolation results of the to-be-interpolated points;

[0010] The inverse distance weighted interpolation result is used as a cooperative variable of the co-kriging interpolation method, and the co-kriging interpolation method is used to perform secondary interpolation on the points to be interpolated in the target area according to the measured values ​​and the cooperative variables to obtain the final predicted values ​​of the points to be interpolated.

[0011] Optionally, the method further includes:

[0012] The measured values ​​and predicted values ​​are input into a random forest model, and the random forest model is subjected to multiple model parameter tuning until the relationship between the measured values ​​and the predicted values ​​is fitted, and the random forest model after the model parameter tuning is further used as a prediction model for predicting indicator objects; the prediction model is used to predict the spatial distribution of indicator objects in the target area.

[0013] Optionally, obtaining the measured value of the indicator object measured at each sample point in the target area includes:

[0014] Gridding the target area, and evenly selecting a number of grids from all grids obtained by the gridding process as sample points;

[0015] The indicator object is measured at the sample points to obtain the measured value of the indicator object measured at each sample point.

[0016] Optionally, after performing gridding processing on the target area and uniformly selecting a number of grids from all grids obtained by the gridding processing as sample points, the method further includes:

[0017] All grids except the sample points are used as points to be interpolated.

[0018] Optionally, the using of the inverse distance weighted interpolation method to interpolate the to-be-interpolated points in the target area according to the measured values ​​to obtain the inverse distance weighted interpolation results of the to-be-interpolated points includes:

[0019] For any interpolation point in the target area, inverse distance weighted interpolation is performed in the following manner:

[0020] Taking the point to be interpolated as the center, searching for sample points within a preset radius, and determining a preset number of target sample points closest to the point to be interpolated from the searched sample points;

[0021] According to the measured values ​​of the preset number of target sample points, an inverse distance weighted interpolation result of the point to be interpolated is obtained by calculating a weighted average value; wherein the weight of each target sample point in the calculated weighted average value is inversely proportional to the distance from the target sample point to the point to be interpolated.

[0022] Optionally, the preset number is 12.

[0023] Optionally, the inverse distance weighted interpolation result is used as a co-variable of a co-kriging interpolation method, and the co-kriging interpolation method is used to perform secondary interpolation on the to-be-interpolated points in the target area according to the measured values ​​and the co-variables to obtain a final predicted value of the to-be-interpolated points, including:

[0024] The inverse distance weighted interpolation result is used as the co-variable of the co-kriging interpolation method, the final predicted value of the point to be interpolated is used as the main variable of the co-kriging interpolation method, and the final predicted value of the point to be interpolated is calculated by the following formula:

[0025]

[0026] Among them, Z(x 0 ) represents the position x 0 The final predicted value of the interpolation point, Z(x 1i ) means at position x 1i p observations, T(x 2j ) means at position x 2j The q inverse distance weighted interpolation results, λ 1i and λ 2j is the weight coefficient.

[0027] Optionally, the measured value and the predicted value are input into a random forest model, and the random forest model is subjected to multiple model parameter tuning until the relationship between the measured value and the predicted value is fitted, including:

[0028] Extracting m bootstrap sample sets from the n original sample pairs consisting of the measured values ​​and the predicted values ​​in a playback manner, thereby constructing m regression trees of the random forest model; wherein the original sample pairs that are not selected during the sampling process constitute m out-of-bag data sets;

[0029] For each node on each regression tree, k split variables are randomly selected from all p explanatory variables, and the next branch is selected from them based on the branching goodness criterion;

[0030] Each regression tree recursively branches from top to bottom until the preset segmentation stop condition is reached.

[0031] Optionally, the indicator object includes at least one of the air quality of the target area, the surface plant coverage, the water and salt content of the surface, and the underground mineral content.

[0032] Optionally, the water-salt content includes water content and total salt content.

[0033] According to a second aspect of an embodiment of this specification, a spatial interpolation device is provided, the device comprising:

[0034] A measuring unit, which obtains measured values ​​of the indicator object measured at a number of sample points located in the target area;

[0035] An inverse distance weighted interpolation unit, using an inverse distance weighted interpolation method, interpolates the to-be-interpolated points in the target area according to the measured values, and obtains an inverse distance weighted interpolation result of the to-be-interpolated points;

[0036] The co-kriging interpolation unit uses the inverse distance weighted interpolation result as a co-variable of the co-kriging interpolation method, and uses the co-kriging interpolation method to perform secondary interpolation on the to-be-interpolated points in the target area according to the measured values ​​and the co-variables to obtain a final predicted value of the to-be-interpolated point.

[0037] Optionally, the device further comprises:

[0038] The model prediction unit inputs the measured value and the predicted value into the random forest model, and the random forest model is subjected to multiple model parameter tuning until the relationship between the measured value and the predicted value is fitted, and the random forest model after the model parameter tuning is further used as a prediction model for predicting the indicator object; the prediction model is used to predict the spatial distribution of the indicator object in the target area.

[0039] Optionally, the measuring unit includes:

[0040] The gridding subunit performs gridding processing on the target area and uniformly selects a number of grids from all grids obtained by the gridding processing as sample points;

[0041] The measuring subunit measures the indicator object at the sample points to obtain the measured value of the indicator object at each sample point.

[0042] Optionally, the gridding subunit is further used to use grids in all grids except the sample points as points to be interpolated.

[0043] Optionally, the inverse distance weighted interpolation unit includes:

[0044] For any interpolation point in the target area, inverse distance weighted interpolation is performed through the following search subunit and calculation subunit:

[0045] The search subunit searches for sample points within a preset radius with the point to be interpolated as the center, and determines a preset number of target sample points closest to the point to be interpolated from the searched sample points;

[0046] The calculation subunit obtains the inverse distance weighted interpolation result of the to-be-interpolated point by calculating a weighted average value based on the measured values ​​of the preset number of target sample points; wherein the weight of each target sample point in the calculated weighted average value is inversely proportional to the distance from the target sample point to the to-be-interpolated point.

[0047] Optionally, the preset number is 12.

[0048] Optionally, the co-kriging interpolation unit is further used to use the inverse distance weighted interpolation result as a co-variable of the co-kriging interpolation method, use the final predicted value of the point to be interpolated as a main variable of the co-kriging interpolation method, and calculate the final predicted value of the point to be interpolated by the following formula:

[0049]

[0050] Among them, Z(x 0 ) represents the position x 0 The final predicted value of the interpolation point, Z(x 1i ) means at position x 1i p observations, T(x 2 j) indicates that at position x 2j The q inverse distance weighted interpolation results, λ 1i and λ 2j is the weight coefficient.

[0051] Optionally, the model prediction unit includes:

[0052] Sampling subunits, extracting m bootstrap sample sets from the n original sample pairs consisting of the measured values ​​and the predicted values ​​in a playback manner, thereby constructing m regression trees of the random forest model; wherein the original sample pairs that are not selected in the sampling process constitute m out-of-bag data sets;

[0053] Branching subunit, for each node on each regression tree, randomly selects k splitting variables from all p explanatory variables, and selects the next branch from them based on the branching goodness criterion; each regression tree recursively performs branching operations in a top-down manner until the preset splitting stop condition is reached.

[0054] Optionally, the indicator object includes at least one of the air quality of the target area, the surface plant coverage, the water and salt content of the surface, and the underground mineral content.

[0055] Optionally, the water-salt content includes water content and total salt content.

[0056] The embodiment of this specification provides a spatial interpolation scheme. By using the interpolation result of the inverse distance weighted method as the co-variable of the co-kriging interpolation method, local information can be introduced to help the co-kriging interpolation better adapt to the spatial variation of actual data, which can make up for the shortcomings of few measured sample points of the main variable and the instability of the variation function.

[0057] On the other hand, the IDW interpolation method can provide local and accurate interpolation results, and the extreme values ​​are largely retained during the interpolation process, while the cokriging interpolation method can consider spatial correlation and provide more reliable estimates. Combining the two, the interpolation results can greatly weaken the "bull's eye" phenomenon of the IDW interpolation method.

[0058] On the other hand, the inverse distance weighted interpolation result used as the collaborative variable is easier to obtain.

[0059] On the other hand, the use of random forest optimization can significantly improve the accuracy of the interpolation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flow chart of a spatial interpolation method provided in an embodiment of this specification;

[0061] Figure 2 A schematic diagram of the measured water-salt distribution provided in an embodiment of this specification;

[0062] Figure 3 A schematic diagram of water-salt distribution based on the inverse distance weighted interpolation method provided in an embodiment of this specification;

[0063] Figure 4 A schematic diagram of water-salt distribution based on co-kriging interpolation method provided in an embodiment of this specification;

[0064] Figure 5 A schematic flow chart of another spatial interpolation method provided in one embodiment of this specification;

[0065] Figure 6 A schematic diagram of water-salt distribution based on random forest optimization provided in an embodiment of this specification;

[0066] Figure 7 is a hardware structure diagram of a spatial interpolation device provided in an embodiment of this specification;

[0067] Figure 8 It is a module of the spatial interpolation device provided in an embodiment of this specification. DETAILED DESCRIPTION

[0068] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with this specification. Instead, they are merely examples of devices and methods consistent with some aspects of this specification as detailed in the appended claims.

[0069] The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit this specification. The singular forms "a", "the" and "the" used in this specification and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0070] It should be understood that although the terms first, second, third, etc. may be used in this specification to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this specification, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0071] In the related art, commonly used spatial interpolation methods may include inverse distance weighted interpolation, Kriging interpolation, and co-Kriging interpolation.

[0072] The interpolation result of the inverse distance weighted interpolation method is between the measured maximum and minimum values. Although it can retain the details to the greatest extent, it is prone to the "bull's eye" phenomenon.

[0073] The Kriging interpolation method uses spatial statistics as its theoretical basis. It can overcome the problem that errors in interpolation are difficult to analyze. It has high interpolation accuracy and strong uniqueness and extrapolation capabilities. However, the interpolation results are prone to smoothing.

[0074] Cokriging interpolation is an improved Kriging interpolation method. By considering the correlation between two or more random processes, more accurate spatial interpolation results can be obtained. However, the covariates required by the cokriging interpolation method are difficult to obtain, or it is difficult to obtain suitable covariates, which leads to the poor applicability of the cokriging interpolation method.

[0075] In view of this, the present invention proposes a new spatial interpolation scheme, which uses the interpolation result of the inverse distance weighted method as a co-variable of the co-kriging interpolation method; on the one hand, local information can be introduced to help the co-kriging interpolation better adapt to the spatial variation of the actual data, and can make up for the shortcomings of the small number of measured sample points of the main variable and the instability of the variance function. On the other hand, the inverse distance weighted interpolation method can provide local and accurate interpolation results, and the extreme values ​​are largely retained during the interpolation process, while the co-kriging interpolation can consider spatial correlation and provide more reliable estimates. Combining the two, the interpolation result greatly weakens the "bull's eye" phenomenon of the inverse distance weighted method interpolation. On the other hand, the inverse distance weighted interpolation result used as a co-variable is easier to obtain.

[0076] The spatial interpolation solution provided by the present invention can be applied to various business scenarios that require spatial interpolation, such as analyzing the air quality distribution, soil salinization distribution, surface plant coverage, underground mineral distribution, etc. in a geographical area.

[0077] Taking the distribution of soil salinization as an example, there is a serious soil salinization problem worldwide. Due to the use of irrigation water, fertilizers and lime, about 1 billion hectares of soil are affected by secondary salinization. It is estimated that under the influence of climate change and human activities, by the end of the 21st century, more than 50% of the world's farmland will be affected by salinization. In arid and semi-arid areas, due to unique climatic conditions and intensive irrigation methods and management measures, the salt in the soil is significantly concentrated from the lower layer to the upper layer, forming secondary salinization of the soil. Water shortage and secondary salinization have become two important factors restricting the development of agriculture in arid and semi-arid areas; therefore, clarifying the distribution of soil water and salt is of great significance to the scientific and rational development and utilization of saline soil.

[0078] In the process of obtaining soil water and salt distribution data, it is difficult to measure soil water and salt data at all locations in a region due to factors such as time, space and cost. In practical applications, soil water and salt are usually measured at several sampling points evenly distributed in the region, and then the soil water and salt distribution in the region is predicted through spatial interpolation methods based on the measured values.

[0079] Since existing spatial interpolation methods (such as inverse distance weighted interpolation, Kriging interpolation and co-Kriging interpolation) have certain defects, the spatial interpolation method provided in this manual can predict accurate water and salt data, thereby obtaining a more accurate distribution of soil water and salt in the region.

[0080] Please refer to the following Figure 1 The spatial interpolation method flow chart shown in FIG. 1 may include the following steps:

[0081] Step 110: obtaining measured values ​​of the indicator object measured at a plurality of sample points located in the target area.

[0082] As mentioned above, the index objects in the present invention may include but are not limited to at least one of the air quality of the target area, the surface plant coverage, the water and salt content of the surface, and the underground mineral content.

[0083] In practical applications, in order to make the sampling distribution more universal, the target area can be gridded in advance, and then a number of grids can be evenly selected from all the grids obtained by the gridding process as sample points; finally, the indicator object is measured at the sample points to obtain the measured values ​​of the indicator object measured at each sample point.

[0084] In this embodiment, the water-salt content of the ground surface is taken as an example for explanation, and the water-salt content may include the surface water content and the total salt content of the target area.

[0085] Please refer to the following Figure 2 The schematic diagram of the measured water-salt distribution is shown in Figure 2 The left side shows the measured value of total salt content at each sample point. Figure 2 The right side of the figure shows the measured values ​​of the moisture content of each sample point.

[0086] Step 120: Use the inverse distance weighted interpolation method to interpolate the points to be interpolated in the target area according to the measured values ​​to obtain inverse distance weighted interpolation results of the points to be interpolated.

[0087] After obtaining the measured values ​​of the indicator objects measured at each sample point, the inverse distance weighted interpolation method can be used to perform inverse distance weighted interpolation on the points to be interpolated in the target area according to the measured values, thereby obtaining inverse distance weighted interpolation results of the points to be interpolated.

[0088] The points to be interpolated may refer to the grids other than the sample points in all the grids after the aforementioned grid processing, that is, all the grids other than the grids selected as the sample points may belong to the points to be interpolated.

[0089] In practical applications, for any point to be interpolated, the sample point closest to its geographical location has the largest contribution, that is, the contribution is inversely proportional to the distance. This is also the origin of the name of the inverse distance weighted interpolation method. In the inverse distance weighted interpolation method, the sample weight is inversely proportional to the distance between the sample and the point to be interpolated, so this interpolation method is called the inverse distance weighted interpolation method.

[0090] The following specifically introduces the use of the inverse distance weighted interpolation method to interpolate the points to be interpolated in the target area according to the measured values ​​to obtain the inverse distance weighted interpolation results of the points to be interpolated:

[0091] For any interpolation point in the target area, inverse distance weighted interpolation can be performed in the following way:

[0092] Taking the point to be interpolated as the center, searching for sample points within a preset radius, and determining a preset number of target sample points closest to the point to be interpolated from the searched sample points;

[0093] According to the measured values ​​of the preset number of target sample points, an inverse distance weighted interpolation result of the point to be interpolated is obtained by calculating a weighted average value; wherein the weight of each target sample point in the calculated weighted average value is inversely proportional to the distance from the target sample point to the point to be interpolated.

[0094] In this embodiment, the preset number may be an empirical value, which may be specifically configured according to the adoption density. Figure 2 In the example of measured water-salt distribution shown in the figure, it has been experimentally verified that when the preset number is set to 12, the interpolation result is more ideal.

[0095] The following is an explanation using a preset number of 12. After searching for 12 sample points closest to the point to be interpolated as the center, a weighted average can be performed based on the measured values ​​of the 12 sample points using the following formula 1 to obtain an inverse distance weighted interpolation result of the point to be interpolated.

[0096]

[0097] Among them, Z is the inverse distance weighted interpolation result of the point to be interpolated, Z i is the measured value of the i-th (i=1,…,12) sample point, d i is the distance between the i-th sample point and the point to be interpolated, and p is the distance d i The power of.

[0098] It should be noted that Formula 1 is for the case where the preset number is 12 sample points. If the preset number is another value n, the value 12 in Formula 1 will also be adjusted to the value n. In addition, the value of p can be an empirical value. It has been verified through experiments that the interpolation result when p is greater than 2 is not much different from the interpolation result when p is 2. Therefore, in order to reduce the amount of calculation, p can be set to 2.

[0099] In the aforementioned Figure 2 Based on the measured value of the water salt content shown in the figure, the above formula 1 can be used to calculate the following: Figure 3 Schematic diagram of water and salt distribution based on inverse distance weighted interpolation method. Figure 3 The left side shows the total salt content of the inverse distance weighted interpolation result. Figure 3 The right side of the figure shows the water content of the inverse distance weighted interpolation result.

[0100] Step 130, using the inverse distance weighted interpolation result as a co-variable of the co-kriging interpolation method, and using the co-kriging interpolation method to perform secondary interpolation on the points to be interpolated in the target area according to the measured values ​​and the co-variables, to obtain the final predicted values ​​of the points to be interpolated.

[0101] Co-kriging interpolation is an improved Kriging interpolation method suitable for processing multiple related variables. Its core idea is to use the spatial information of the co-variable (also called the secondary variable) to improve the estimation of the primary variable (also called the target variable) when constructing the interpolation model of the primary variable (also called the target variable). The primary variable can represent the target variable to be predicted, and the co-variable can represent one or more variables spatially related to the primary variable. The co-variable usually has relatively complete observation data.

[0102] As mentioned above, the problem with the cokriging interpolation method is that the required covariables are difficult to obtain, or it is difficult to obtain suitable covariables, which leads to the poor applicability of the cokriging interpolation method.

[0103] Therefore, in this embodiment, the interpolation result of the inverse distance weighted method is used as the co-variable of the co-kriging interpolation method to help the co-kriging interpolation better adapt to the spatial changes of the actual data, which can make up for the shortcomings of few measured sample points of the main variable and the instability of the variation function.

[0104] The improved co-kriging interpolation method provided by this embodiment is specifically described below: the inverse distance weighted interpolation result is used as the co-variable of the co-kriging interpolation method, and the co-kriging interpolation method is used to perform secondary interpolation on the to-be-interpolated points in the target area according to the measured value and the co-variable, so as to obtain the final predicted value of the to-be-interpolated point:

[0105] Using the inverse distance weighted interpolation result as a co-variable of the co-kriging interpolation method, and using the final predicted value of the to-be-interpolated point as a main variable of the co-kriging interpolation method;

[0106] In this embodiment, according to the basic principle of co-kriging interpolation, the spatial correlation between the measured value and other variables is taken into consideration. It is assumed that the variable Z(x) is randomly distributed in the target area. The variable Z(x) represents the value at the spatial position x. 1i There are p observations Z(x) on (i=1, 2, 3, ..., p) 1i ), then the variable Z(x) can be regarded as the main variable of interpolation; at the same time, there is another variable T(x), which represents the interpolation at different spatial positions x 2j (j=1, 2, 3, ..., q) has a corresponding observation value T(x 2j ), where T(x 2j) is the inverse distance weighted interpolation result obtained in the previous step 120.

[0107] Since T(x) covers the positions of all known sample points of the main variable Z(x), it can meet the prerequisite for being used as a collaborative variable. 0 The prediction of the main variable Z(x) at the point to be interpolated can be obtained 0 The final predicted value.

[0108] Exemplarily, the final predicted value of the point to be interpolated can be obtained by calculating the following formula 2:

[0109]

[0110] Among them, Z(x 0 ) represents the final predicted value of any point to be interpolated.

[0111] Z(x 1i ) represents the position x in space 1i p observations on T(x 2j ) represents the position x in space 2j q observations on 1i The weight coefficient of the main variable, λ 2j is the weight coefficient of the co-variable, p is the number of observation points of the main variable, and q is the number of observation points of the co-variable.

[0112] Furthermore, the weight coefficient λ here 1i and λ 2j It can be solved by the following equation:

[0113]

[0114] Among them, C ZZ is the covariance matrix between the main variables;

[0115] C ZY is the covariance matrix between the principal and covariates;

[0116] C YZ is the covariance matrix between the covariates and the principal variables (with C ZY same);

[0117] C YY is the covariance matrix between covariates;

[0118] C Z0 is the covariance vector between the main variable and the point to be interpolated;

[0119] C Y0 is the covariance vector between the covariate and the point to be interpolated;

[0120] By solving this equation, we can get the weight coefficient λ 1i and λ 2j Once the weight coefficient is determined, the co-kriging interpolation method can be used to predict the predicted value of the main variable at any interpolation point. Specifically, the weight coefficient λ can be determined. 1i and λ 2j Substitute the above formula 2 to calculate the main variable at the interpolation point x 0 The final predicted value.

[0121] In the aforementioned Figure 3 Based on the water-salinity content obtained by the inverse distance weighted interpolation method shown in FIG. 2 , the above formula can be used to calculate the following: Figure 4 Schematic diagram of water and salt distribution based on co-kriging interpolation method. Figure 4 The left side shows the predicted value of total salt content predicted by the co-kriging interpolation method. Figure 4 The right side of the figure shows the predicted value of water content predicted by the co-kriging interpolation method.

[0122] Through the above embodiments, the interpolation results of the inverse distance weighted method are used as the co-variable of the co-kriging interpolation method; on the one hand, local information can be introduced to help the co-kriging interpolation better adapt to the spatial variation of the actual data, which can make up for the shortcomings of the small number of measured sample points of the main variable and the instability of the variance function. On the other hand, the inverse distance weighted interpolation method can provide local and accurate interpolation results, and the extreme values ​​are largely retained during the interpolation process, while the co-kriging interpolation can consider spatial correlation and provide more reliable estimates. Combining the two, the interpolation results largely weaken the "bull's eye" phenomenon of the inverse distance weighted method interpolation. On the other hand, the inverse distance weighted interpolation results used as co-variables are easier to obtain.

[0123] With the continuous development of machine learning technology, the efficiency and accuracy of spatial interpolation results can be effectively improved by training the prediction model of prediction indicator objects through machine learning technology.

[0124] In view of this, the present invention Figure 1 Based on the spatial interpolation method shown in Figure 5 A flowchart of another spatial interpolation method is shown in FIG.

[0125] like Figure 5 As shown, in Figure 1 Based on the steps 110 to 130 shown, step 140 may be further included after step 130:

[0126] The measured values ​​and predicted values ​​are input into a random forest model, and the random forest model is subjected to multiple model parameter tuning until the relationship between the measured values ​​and the predicted values ​​is fitted, and the random forest model after the model parameter tuning is further used as a prediction model for predicting indicator objects; the prediction model is used to predict the spatial distribution of indicator objects in the target area.

[0127] In this step, Random Forest is selected as the introduced machine learning model. Random Forest is a supervised learning algorithm based on ensemble learning of decision trees. The basic principle of the Random Forest model is to make the final prediction by building multiple decision trees and voting (classification problem) or averaging (regression problem). The core idea of ​​Random Forest is to increase the diversity of the model, reduce overfitting, and improve the generalization ability of the model by introducing randomness.

[0128] In this embodiment, according to the basic principle of the random forest model, m self-service sample sets can be extracted from the n original sample pairs consisting of the measured values ​​and the predicted values ​​in a playback manner to construct m regression trees; wherein the original sample pairs that are not selected during the sampling process constitute m out-of-bag data sets.

[0129] For each node on each regression tree, k split variables are randomly selected from all p explanatory variables, and the optimal branch is selected based on the branching goodness criterion. Each regression tree is recursively branched in a top-down manner until the preset splitting stop condition is reached.

[0130] In the present disclosure, m bootstrap sample sets are extracted from the n original sample pairs consisting of the measured values ​​and the predicted values ​​in a playback manner, and statistical methods (such as bootstrap) can be used for sampling with replacement. Sampling with replacement can make the training data of each regression tree different, thereby enhancing the diversity of the model.

[0131] In the random forest model, the “explanatory variables” and “split variables” are closely related to the model training and decision tree construction process.

[0132] Explanatory variables are input features used to predict the target variable (usually the response variable or dependent variable). In simple terms, they are the independent variables that the model uses to make predictions. In each regression tree, the explanatory variables can be the basic data used to split the node. Each tree selects a subset of all the explanatory variables to split the node. These explanatory variables are usually individual features in the dataset. In the dataset, suppose there are p explanatory variables, then each tree of the random forest model will use these explanatory variables to split the node.

[0133] The splitting variable is a subset randomly selected from the explanatory variables to determine the splitting method of each node. Specifically, when building a regression tree or a classification tree, each node is split by selecting several (usually a small random subset) from the explanatory variables of the current node as candidate splitting variables, and then selecting one of the best variables to split the node to generate the next branch. When building each node of the tree, the splitting variable is a part of all the explanatory variables randomly selected. The purpose of this is to increase the diversity of the model and reduce the correlation between trees, thereby improving the accuracy and robustness of the entire forest model. By randomly selecting splitting variables, random forests can avoid overfitting and improve the accuracy of predictions. Usually, it is assumed that k splitting variables (where k is a hyperparameter) are randomly selected from p explanatory variables when each node is split, and then the optimal splitting variable is selected from them for node splitting according to some criteria (such as information gain, Gini index, etc.).

[0134] In the process of building each regression tree, for the splitting of each node, the random forest does not use all the features, but randomly selects some features for splitting, which can further improve the diversity of the model.

[0135] For regression tasks, random forests will average the prediction results of all m regression trees to obtain the prediction results; then the model parameters are tuned by measuring the error between the prediction results and the measured results, and the above process is repeated using the tuned model parameters. In this way, after multiple iterations, the model parameters can be gradually optimized, and the error between the prediction results and the measured values ​​can be continuously fitted (minimized) until the model converges or reaches the convergence condition. Finally, the random forest model with the tuned model parameters can be used as a prediction model, and the prediction model can be used to predict the spatial distribution of the indicator objects in the target area.

[0136] In the aforementioned Figure 2 The measured values ​​of the water salt content are shown as well as Figure 4 Based on the predicted value of the water-salinity content predicted by the co-kriging interpolation method shown in FIG. 1 , the prediction model optimized by the random forest in the embodiment shown in step 140 can predict the following: Figure 6 Schematic diagram of water-salt distribution is shown. Figure 6 The left side shows the total salt content output by the prediction model. Figure 6 Shown on the right is the water content output by the prediction model.

[0137] In summary, the present invention proposes a new spatial interpolation scheme, which uses the interpolation result of the inverse distance weighted method as a co-variable of the co-kriging interpolation method; on the one hand, local information can be introduced to help the co-kriging interpolation better adapt to the spatial variation of actual data, and can make up for the shortcomings of few measured sample points of the main variable and the instability of the variance function. On the other hand, the inverse distance weighted interpolation method can provide local and accurate interpolation results, and the extreme values ​​are largely retained during the interpolation process, while the co-kriging interpolation can consider spatial correlation and provide more reliable estimates. Combining the two, the interpolation result greatly weakens the "bull's eye" phenomenon of the inverse distance weighted interpolation method. On the other hand, the inverse distance weighted interpolation result used as a co-variable is easier to obtain. On the other hand, the use of random forest optimization can improve the accuracy of the interpolation result.

[0138] Corresponding to the above-mentioned spatial interpolation method embodiment, this specification also provides an embodiment of a spatial interpolation device. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of the device in which it is located reading the corresponding computer program in the non-volatile memory into the memory and running it. From the hardware level, if Figure 7 The figure is a hardware structure diagram of the device where the spatial interpolation device in this specification is located, except Figure 7 In addition to the processor, network interface, memory and non-volatile memory shown, the device in which the apparatus in the embodiment is located may also include other hardware according to the actual communication function, which will not be described in detail.

[0139] See also Figure 8 , is a module diagram of a spatial interpolation device provided in an embodiment of this specification, and the device corresponds to Figure 1 In the embodiment shown, the device comprises:

[0140] The measuring unit 810 obtains measured values ​​of the indicator object measured at a plurality of sample points located in the target area;

[0141] An inverse distance weighted interpolation unit 820 uses an inverse distance weighted interpolation method to interpolate the to-be-interpolated points in the target area according to the measured values ​​to obtain an inverse distance weighted interpolation result of the to-be-interpolated points;

[0142] The co-kriging interpolation unit 830 uses the inverse distance weighted interpolation result as a co-variable of the co-kriging interpolation method, and uses the co-kriging interpolation method to perform secondary interpolation on the to-be-interpolated points in the target area according to the measured values ​​and the co-variables to obtain the final predicted values ​​of the to-be-interpolated points.

[0143] Optionally, the device further comprises:

[0144] The model prediction unit 840 inputs the measured value and the predicted value into the random forest model, and the random forest model is subjected to multiple model parameter tuning until the relationship between the measured value and the predicted value is fitted, and the random forest model after the model parameter tuning is further used as a prediction model for predicting the indicator object; the prediction model is used to predict the spatial distribution of the indicator object in the target area.

[0145] Optionally, the measuring unit 810 includes:

[0146] The gridding subunit 811 performs gridding processing on the target area, and uniformly selects a number of grids from all grids obtained by the gridding processing as sample points;

[0147] The measuring subunit 813 performs measuring of the indicator object on the sample points to obtain the measured value of the indicator object measured at each sample point.

[0148] Optionally, the gridding subunit 811 is further configured to use grids in all grids except the sample points as points to be interpolated.

[0149] Optionally, the inverse distance weighted interpolation unit 820 includes:

[0150] For any point to be interpolated in the target area, inverse distance weighted interpolation is performed by the following search subunit 821 and calculation subunit 823:

[0151] The search subunit 821 searches for sample points within a preset radius with the point to be interpolated as the center, and determines a preset number of target sample points closest to the point to be interpolated from the searched sample points;

[0152] The calculation subunit 823 obtains the inverse distance weighted interpolation result of the to-be-interpolated point by calculating a weighted average value based on the measured values ​​of the preset number of target sample points; wherein the weight of each target sample point in the calculated weighted average value is inversely proportional to the distance from the target sample point to the to-be-interpolated point.

[0153] Optionally, the preset number is 12.

[0154] Optionally, the co-kriging interpolation unit 830 is further configured to use the inverse distance weighted interpolation result as a co-variable of the co-kriging interpolation method, use the final predicted value of the point to be interpolated as a main variable of the co-kriging interpolation method, and calculate the final predicted value of the point to be interpolated by the following formula:

[0155]

[0156] Among them, Z(x 0 ) represents the position x0 The final predicted value of the interpolation point, Z(x 1i ) means at position x 1i p observations, T(x 2 j) indicates that at position x 2j The q inverse distance weighted interpolation results, λ 1i and λ 2j is the weight coefficient.

[0157] Optionally, the model prediction unit 840 includes:

[0158] The sampling subunit 841 extracts m bootstrap sample sets from the n original sample pairs consisting of the measured values ​​and the predicted values ​​in a playback manner, thereby constructing m regression trees of the random forest model; wherein the original sample pairs that are not selected in the sampling process constitute m out-of-bag data sets;

[0159] The branching subunit 843 randomly selects k splitting variables from all p explanatory variables for each node on each regression tree, and selects the next branch from them based on the branching goodness criterion; each regression tree recursively performs branching operations in a top-down manner until a preset splitting stop condition is reached.

[0160] Optionally, the indicator object includes at least one of the air quality of the target area, the surface plant coverage, the water and salt content of the surface, and the underground mineral content.

[0161] Optionally, the water-salt content includes water content and total salt content.

[0162] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which may be in the form of a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email transceiver, a game console, a tablet computer, a wearable device or a combination of any of these devices.

[0163] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0164] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can refer to the partial description of the method embodiments. The device embodiments described above are only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this specification. Ordinary technicians in this field can understand and implement it without paying creative work.

[0165] Corresponding to the aforementioned spatial interpolation method embodiment, the present specification also provides an autonomous underwater vehicle, on which an autonomous unmanned system is deployed, and the autonomous unmanned system implements the aforementioned spatial interpolation method.

[0166] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0167] Those skilled in the art will readily appreciate other embodiments of the specification after considering the specification and practicing the invention disclosed herein. The specification is intended to cover any variations, uses, or adaptations of the specification that follow the general principles of the specification and include common knowledge or customary techniques in the art that are not disclosed in the specification. The specification and examples are to be considered exemplary only, and the true scope and spirit of the specification are indicated by the following claims.

[0168] It should be understood that the present description is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present description is limited only by the appended claims.

Claims

1. A spatial interpolation method, characterized in that: The method comprises: Obtaining measured values ​​of the indicator object measured at a number of sample points located in the target area; Using the inverse distance weighted interpolation method, interpolating the to-be-interpolated points in the target area according to the measured values, to obtain the inverse distance weighted interpolation results of the to-be-interpolated points; The inverse distance weighted interpolation result is used as a cooperative variable of the co-kriging interpolation method, and the co-kriging interpolation method is used to perform secondary interpolation on the points to be interpolated in the target area according to the measured values ​​and the cooperative variables to obtain the final predicted values ​​of the points to be interpolated.

2. The method according to claim 1, further comprising: The measured values ​​and predicted values ​​are input into a random forest model, and the random forest model is subjected to multiple model parameter tuning until the relationship between the measured values ​​and the predicted values ​​is fitted, and the random forest model after the model parameter tuning is further used as a prediction model for predicting indicator objects; the prediction model is used to predict the spatial distribution of indicator objects in the target area.

3. The method according to claim 1, characterized in that The obtaining of measured values ​​of the indicator objects measured at each sample point in the target area includes: Gridding the target area, and evenly selecting a number of grids from all grids obtained by the gridding process as sample points; The indicator object is measured at the sample points to obtain the measured value of the indicator object measured at each sample point.

4. The method according to claim 3, characterized in that After the target area is gridded and a number of grids are uniformly selected from all grids obtained by the gridding process as sample points, the method further includes: All grids except the sample points are used as points to be interpolated.

5. The method according to claim 1, characterized in that The method of using the inverse distance weighted interpolation method to interpolate the points to be interpolated in the target area according to the measured values ​​to obtain the inverse distance weighted interpolation results of the points to be interpolated includes: For any interpolation point in the target area, inverse distance weighted interpolation is performed in the following manner: Taking the point to be interpolated as the center, searching for sample points within a preset radius, and determining a preset number of target sample points closest to the point to be interpolated from the searched sample points; According to the measured values ​​of the preset number of target sample points, an inverse distance weighted interpolation result of the point to be interpolated is obtained by calculating a weighted average value; wherein the weight of each target sample point in the calculated weighted average value is inversely proportional to the distance from the target sample point to the point to be interpolated.

6. The method according to claim 5, characterized in that The preset number is 12.

7. The method according to claim 1, characterized in that The method uses the inverse distance weighted interpolation result as a cooperative variable of the co-kriging interpolation method, and uses the co-kriging interpolation method to perform secondary interpolation on the to-be-interpolated points in the target area according to the measured value and the cooperative variable to obtain the final predicted value of the to-be-interpolated point, including: The inverse distance weighted interpolation result is used as the co-variable of the co-kriging interpolation method, the final predicted value of the point to be interpolated is used as the main variable of the co-kriging interpolation method, and the final predicted value of the point to be interpolated is calculated by the following formula: Among them, Z(x0) represents the final predicted value of the interpolation point at position x0, and Z(x 1i ) means at position x 1i p observations, T(x 2j ) means at position x 2j The q inverse distance weighted interpolation results, λ 1i and λ 2j is the weight coefficient.

8. The method according to claim 2, characterized in that: The step of inputting the measured value and the predicted value into a random forest model, and performing multiple model parameter tuning on the random forest model until the relationship between the measured value and the predicted value is fitted, comprises: Extracting m bootstrap sample sets from the n original sample pairs consisting of the measured values ​​and the predicted values ​​in a playback manner, thereby constructing m regression trees of the random forest model; wherein the original sample pairs that are not selected during the sampling process constitute m out-of-bag data sets; For each node on each regression tree, k split variables are randomly selected from all p explanatory variables, and the next branch is selected from them based on the branching goodness criterion; Each regression tree recursively branches from top to bottom until the preset segmentation stop condition is reached.

9. The method according to claim 1, characterized in that: The indicator object includes at least one of the air quality, surface plant coverage, surface water salt content, and underground mineral content of the target area.

10. The method according to claim 9, characterized in that The water-salt content includes water content and total salt content.

11. A spatial interpolation device, characterized in that: A measuring unit, which obtains measured values ​​of the indicator object measured at a number of sample points located in the target area; An inverse distance weighted interpolation unit, using an inverse distance weighted interpolation method, interpolates the to-be-interpolated points in the target area according to the measured values, and obtains an inverse distance weighted interpolation result of the to-be-interpolated points; The co-kriging interpolation unit uses the inverse distance weighted interpolation result as a co-variable of the co-kriging interpolation method, and uses the co-kriging interpolation method to perform secondary interpolation on the to-be-interpolated points in the target area according to the measured values ​​and the co-variables to obtain a final predicted value of the to-be-interpolated point.

12. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to execute the method according to any one of claims 1 to 10.

Images